CN103136221A - Method capable of generating requirement template and requirement identification method and device - Google Patents
Method capable of generating requirement template and requirement identification method and device Download PDFInfo
- Publication number
- CN103136221A CN103136221A CN2011103793355A CN201110379335A CN103136221A CN 103136221 A CN103136221 A CN 103136221A CN 2011103793355 A CN2011103793355 A CN 2011103793355A CN 201110379335 A CN201110379335 A CN 201110379335A CN 103136221 A CN103136221 A CN 103136221A
- Authority
- CN
- China
- Prior art keywords
- demand type
- inquiry
- described demand
- template
- subquery
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Abstract
The invention provides a method capable of generating a requirement template, a requirement identification method and a device. The method capable of generating the requirement template includes that obtaining seed query of need type from a search log; generalizing the seed query of the need type into a candidate template of the need type; selecting a final template of the need type from the candidate template of the need type. The requirement identification method includes obtaining user query; confirming the final template which is obtained from the method of the need template and matched with user query, and enabling the need type, corresponding to the final template which is matched with the user query to be used as a requirement of the user query. By the method, expandability and maintenance of a requirement identification procedure are greatly improved.
Description
[technical field]
The present invention relates to natural language processing technique, a kind of particularly method that generates the demand template, method and the device thereof of demand identification.
[background technology]
Along with the widespread use of search engine, search engine technique has obtained very great development, and search engine of today not only has been satisfied with and is returned to the content that is complementary with user's inquiry, but attempts to return the content relevant to user's query demand.
Return to the content relevant to user's query demand, at first need to understand the demand of user when search, namely need user's request is identified, in the prior art, the rule-based mode of common employing is identified in user's request realized.For example the developer makes discovery from observation, and the inquiry with " mp3 " ending in user's inquiry is all generally the demand of music class, and the developer just is written to " demand of mp3 ending is the music demand " this rule in the demand recognizer.There are following two problems in this method: at first, the foundation of rule relies on people's observation, therefore need to expend a large amount of manpower and materials, and is difficult to set up the rule that comprehensively covers various demands, thereby causes application program to be difficult to identify user's various demands; Secondly, adopts this method identification user's request, because rule is embedded on line in demand recognizer code, therefore can cause extensibility and all reductions greatly of maintainability of demand recognizer.
[summary of the invention]
Technical matters to be solved by this invention is to provide a kind of method of demand template, method and device thereof of demand identification of generating, to solve in prior art, the program that user's request is identified is difficult to identify user's various demands comprehensively, and extensibility and maintainable all relatively poor defectives.
The present invention is that the technical scheme that the technical solution problem adopts is to provide a kind of method that generates the demand template, comprising: the kind subquery that obtains the demand type from the search daily record; Be the candidate template of described demand type with the seed Query generalization of described demand type; Choose the final template of described demand type from the candidate template of described demand type.
The preferred embodiment one of according to the present invention, the step of obtaining the kind subquery of demand type comprises: the initial seed inquiry of obtaining default described demand type; All inquiries of recording in the search daily record are carried out cluster according to the method for hierarchical clustering; Determine a cluster level, make the initial seed inquiry that is no less than predetermined ratio under this level minimum by the poly-inquiry sum that comprises in same class X and at this level lower class X, all inquiries that the class X under this level is comprised are as the kind subquery of described demand type.
The preferred embodiment one of according to the present invention, the step of obtaining the kind subquery of described demand type comprises: the initial seed inquiry of obtaining default described demand type; Use the iterative learning device to satisfy the inquiry of preset requirement from search daily record learning and the similarity between the inquiry of described initial seed, and the inquiry that will learn and described initial seed are inquired about in the lump the kind subquery as described demand type.
The preferred embodiment one of according to the present invention, the step of obtaining the kind subquery of demand type comprises: cause daily record in the clicked inquiry of the page of described demand type from search, choose the highest N1 of inquiry times inquiry as the kind subquery of described demand type, described N1 is default positive integer; Perhaps, extract the kind subquery of described demand type from the search daily record of the vertical search of described demand type.
The preferred embodiment one of according to the present invention, the step that is the candidate template of described demand type with the seed Query generalization of described demand type comprises: will replace to the asterisk wildcard of classification under described default entity word in the kind subquery of described demand type with the part of the corresponding default entity word coupling of described demand type; Perhaps, with the part that is identified by the classification recognition function in the kind subquery of described demand type replace to described classification recognition function the asterisk wildcard of corresponding classification, wherein said classification recognition function is to be used for such other function of identification according to the attribute definition of a classification.
The preferred embodiment one of according to the present invention, the step that is the template of described demand type with the seed Query generalization of described demand type further comprises: the word that in the kind subquery of described demand type, the contribution degree of described demand type is required lower than default contribution degree is replaced with length asterisk wildcard for restriction word length.
The preferred embodiment one of according to the present invention, when choosing the final template of described demand type from the candidate template of described demand type, carry out according at least one in the following characteristics of the candidate template of described demand type: click feature is used for characterizing the clicked probability of the page that inquiry that the candidate template of described demand type covers can cause described demand type; The similarity feature is used for characterizing the general character degree of all candidate template of a candidate template of described demand type and described demand type; The matching capacity feature is used for characterizing the ability of inquiry of the described demand type of candidate template coupling of described demand type.
The preferred embodiment one of according to the present invention, the click feature of the candidate template W of described demand type adopt following manner to calculate:
Wherein Click (W) represents the click feature of W,
All inquiries that expression W covers in the search daily record cause the clicked number of times of described demand type page,
All inquiries that expression W covers in the search daily record cause the number of times that all pages are clicked.
The preferred embodiment one of according to the present invention, the similarity feature of the candidate template W of described demand type adopt following manner to calculate:
Wherein, the similarity feature of Similarity (W) expression W,
Similarity sum between the every other candidate template of expression W and described demand type.
The preferred embodiment one of according to the present invention, the matching capacity feature of the candidate template W of described demand type adopt following manner to calculate:
Wherein, the matching capacity feature of Match (W) expression W,
The quantity of the inquiry that expression W matches in the search daily record of the inquiry formation of described demand type,
The quantity of the inquiry that expression W matches in the search daily record of the inquiry formation of various demand types.
The present invention also provides a kind of method of demand identification, comprising: obtain user's inquiry; Determine the final template that is complementary with described user's inquiry and the demand that will have as described user's inquiry with the corresponding demand type of final template that described user's inquiry is complementary in the final template that the method for the described generation demand of preamble template obtains.
The present invention also provides a kind of device that generates the demand template, comprising: the seed acquiring unit, for obtain the kind subquery of demand type from the search daily record; Extensive unit, the seed Query generalization that is used for described demand type is the candidate template of described demand type; Choose the unit, be used for choosing from the candidate template of described demand type the final template of described demand type.
The preferred embodiment one of according to the present invention, described seed acquiring unit comprises: first chooses the unit, is used for obtaining the initial seed inquiry of default described demand type; Cluster cell is used for all inquiries that the search daily record is recorded are carried out cluster according to the method for hierarchical clustering; Determining unit, be used for determining a cluster level, make the initial seed inquiry that is no less than predetermined ratio under this level minimum by the poly-inquiry sum that comprises in same class X and at this level lower class X, all inquiries that the class X under this level is comprised are as the kind subquery of described demand type.
The preferred embodiment one of according to the present invention, described seed acquiring unit comprises: second chooses the unit, be used for obtaining default described demand type the initial seed inquiry; Unit be used for to use the iterative learning device to satisfy the inquiry of preset requirement from search daily record learning and the similarity between the inquiry of described initial seed, and the inquiry that will learn and described initial seed are inquired about in the lump the kind subquery as described demand type.
The preferred embodiment one of according to the present invention, described seed acquiring unit is when obtaining the kind subquery of described demand type, specifically cause daily record in the clicked inquiry of the page of described demand type from search, choose the highest N1 of inquiry times inquiry as the kind subquery of described demand type, described N1 is default positive integer; Perhaps, extract the kind subquery of described demand type from the search daily record of the vertical search of described demand type.
The preferred embodiment one of according to the present invention, when described extensive unit is the candidate template of described demand type at the seed Query generalization with described demand type, specifically the part of corresponding with described demand type default entity word coupling in the kind subquery of described demand type is replaced to the asterisk wildcard of classification under described default entity word; Perhaps, with the part that is identified by the classification recognition function in the kind subquery of described demand type replace to described classification recognition function the asterisk wildcard of corresponding classification, wherein said classification recognition function is to be used for such other function of identification according to the attribute definition of a classification.
The preferred embodiment one of according to the present invention, described extensive unit also is used for the word that the kind subquery of described demand type requires lower than default contribution degree the contribution degree of described demand type is replaced with length asterisk wildcard for restriction word length.
The preferred embodiment one of according to the present invention, described when choosing the unit choosing the final template of described demand type from the candidate template of described demand type, carry out according at least one in the following characteristics of the candidate template of described demand type: click feature is used for characterizing the clicked probability of the page that inquiry that the candidate template of described demand type covers can cause described demand type; The similarity feature is used for characterizing the general character degree of all candidate template of a candidate template of described demand type and described demand type; The matching capacity feature is used for characterizing the ability of inquiry of the described demand type of candidate template coupling of described demand type.
The preferred embodiment one of according to the present invention, the described unit of choosing adopts following manner to calculate the click feature of the candidate template W of described demand type:
Wherein Click (W) represents the click feature of W,
All inquiries that expression W covers in the search daily record cause the clicked number of times of described demand type page,
All inquiries that expression W covers in the search daily record cause the number of times that all pages are clicked.
The preferred embodiment one of according to the present invention, the described unit of choosing adopts following manner to calculate the similarity feature of the candidate template W of described demand type:
Wherein, the similarity feature of Similarity (W) expression W,
Similarity sum between the every other candidate template of expression W and described demand type.
The preferred embodiment one of according to the present invention, the described unit of choosing adopts following manner to calculate the matching capacity feature of the candidate template W of described demand type:
Wherein, the matching capacity feature of Match (W) expression W,
The quantity of the inquiry that expression W matches in the search daily record of the inquiry formation of described demand type,
The quantity of the inquiry that expression W matches in the search daily record of the inquiry formation of various demand types.
The present invention also provides a kind of device of demand identification, comprising: the inquiry acquiring unit is used for obtaining user's inquiry; Matching unit, determine to inquire about with described user the final template that is complementary for the final template that the device in the described generation demand of preamble template obtains, and will inquire about the demand that have as described user with the corresponding demand type of final template that described user's inquiry is complementary.
As can be seen from the above technical solutions, by the way, need not the developer and manually recognition rule is write in the demand recognizer, but automatically utilize the query generation demand template that records in the search daily record that user's request is identified by machine.In program, after getting user's inquiry, utilize the demand template that generates under line on line, can judge well user's demand type.In the mode of this identification user's request, the demand template is automatically to generate, and has saved manpower and materials, simultaneously, realize separating under the line and on line owing to generating the identification of demand template and demand, made extensibility and all raisings greatly of maintainability of demand recognizer.
[description of drawings]
Fig. 1 is the schematic flow sheet that generates the embodiment of the method for demand template and the method that demand is identified in the present invention;
Fig. 2 carries out the structural representation of cluster to the inquiry in the search daily record in the present invention;
Fig. 3 is the schematic diagram of coupling tree in the present invention;
Fig. 4 is the structural representation block diagram that generates the embodiment of the device of demand template and the device that demand is identified in the present invention;
Fig. 5 is the structural representation block diagram of an embodiment of seed acquiring unit in the present invention;
Fig. 6 is the structural representation block diagram of another embodiment of seed acquiring unit in the present invention.
[embodiment]
In order to make the purpose, technical solutions and advantages of the present invention clearer, describe the present invention below in conjunction with the drawings and specific embodiments.
Please refer to Fig. 1, Fig. 1 is the schematic flow sheet that generates the embodiment of the method for demand template and the method that demand is identified in the present invention.As shown in Figure 1, the present embodiment is divided under line part on part and line, and its center line bottom is divided into the schematic flow sheet of embodiment of the method for generation demand template, and line top is divided into the schematic flow sheet of embodiment of the method for demand identification.The line bottom divides the method for generation demand template to comprise:
Step S101: the kind subquery that obtains the demand type from search daily record (querylog).
Step S102: be the candidate template of corresponding demand type with the seed Query generalization of corresponding demand type.
Step S103: the final template of choosing corresponding demand type from the candidate template of corresponding demand type.
The below is specifically described above-mentioned steps.
In step S101, the inquiry of a kind of seed of demand type comprises the inquiry that can reflect this demand type of expressing in every way.For example plant subquery " soul-stirring theme song step by step " and " three cun paradise ", although expression way is different, the demand of expressing is identical, is all the same song of inquiry.
The kind subquery that obtains the demand type can have various ways, below by specific embodiment, the mode of obtaining kind of subquery is introduced.
Obtain a kind of embodiment one of kind subquery of demand type:
Step S1011: obtain the initial seed inquiry of default this kind demand type, and all inquiries of recording in the search daily record are carried out cluster according to the method for hierarchical clustering.
Step S1012: determine a cluster level, make the initial seed inquiry that is no less than predetermined ratio under this level minimum by the poly-inquiry sum that comprises in same class X and at this level lower class X, all inquiries that the class X under this level is comprised are as the kind subquery of this kind demand type.Be appreciated that, predetermined ratio wherein is as a ratio value, span should be more than or equal to zero and less than or equal to 1, and is 1 when (namely 100%) when the predetermined ratio value, and in fact the initial seed that is no less than predetermined ratio is inquired about is exactly that whole initial seeds is inquired about.
The initial seed inquiry can have the inquiry of this kind demand type to obtain by artificial selecting.The present embodiment can adopt bottom-up or top-down dual mode to carry out hierarchical clustering.Bottom-up hierarchy clustering method is: at first each is inquired about as a class, then by iteration, constantly merge two the most similar classes, until a class is merged in all inquiries; Top-down hierarchy clustering method is: at first regard all inquiries as a class, then by iteration, find out the most dissimilar inquiry and divide away and become two classes, until constitute a class by itself to each inquiry.
In cluster, know the similarity between different inquiries, a kind of mode be by calculate with different inquiry characteristic of correspondence term vectors between the cosine similarity obtain, wherein with each inquiry characteristic of correspondence word, can extract from the result for retrieval of this inquiry correspondence and obtain.In addition, inquire about corresponding user click data from each and also can be used as the feature of calculating similarity between different inquiries.The present invention does not limit the method that how to judge the similarity between different inquiries in cluster, can adopt the any-mode known to those skilled in the art to carry out.
Below by an above-mentioned process of obtaining a kind of kind subquery of demand type of instantiation explanation.The initial seed inquiry of supposing music demand type is " soul-stirring theme song step by step ", " three cun paradise mp3 " and " loving the free audition of my Chinese song ".The inquiry of search in daily record has: 1, soul-stirring caudal flexure step by step, 2, soul-stirring theme song step by step, 3, three cun paradise, 4, tight skill red three cun paradise, 5, three cun paradise mp3,6, pleasing to the ear song, 7 love the free auditions of my Chinese song, 8, step by step soul-stirringly watch online, 9, step by step soul-stirring the 30th the collection, 10, step by step soul-stirring without abreviation version, 11, to pass through novel step by step soul-stirring, the result of hierarchical clustering is carried out in above inquiry please refer to Fig. 2.
Fig. 2 carries out the result schematic diagram of cluster to the inquiry in the search daily record in the present invention.As shown in Figure 2, on the ground floor of cluster result, inquiry 1,2 is classes, and inquiry 3,4 is classes, and inquiry 8,9 is classes, and all the other inquire about each class naturally.On the second layer of cluster result, inquiry 1,2 is classes, and inquiry 3,4,5 is classes, and inquiry 8,9,10 is classes, and all the other inquire about each class naturally.On the 3rd layer of inquiry, inquiry 1,2,3,4,5 is classes, and inquiry 8,9,10 is classes, and all the other inquire about each class naturally.On the 4th layer of inquiry, inquiry 1,2,3,4,5,6 is classes, and inquiry 8,9,10 is classes, and all the other inquire about each class naturally.On the layer 5 of inquiry, inquiry 1,2,3,4,5,6,7 is classes, and inquiry 8,9,10 is classes, and inquiry 11 constitutes a class by itself.On the layer 6 of inquiry, inquiry 1,2,3,4,5,6,7,8,9,10 is classes, and inquiry 11 constitutes a class by itself.On the layer 7 of inquiry, all inquiries form a class.Can find out, the initial seed inquiry is respectively inquiry 2,5,7, if the said predetermined ratio in front gets 1, namely require to determine a level, under this level, all initial seed inquiries (namely inquiring about 2,5,7) are in same class, and the number of queries that this class comprises should be minimum, obviously, there is the class that satisfies this condition in layer 5, all inquiries that this class comprises are respectively inquiries 1,2,3,4,5,6,7, are exactly the final kind subquery of music demand type so inquire about 1,2,3,4,5,6,7.
Obtain a kind of embodiment two of kind subquery of demand type:
Step S101a: the initial seed inquiry of obtaining default this kind demand type.
Step S102b: the inquiry of using the iterative learning device to meet the demands from search daily record learning and similarity between the initial seed inquiry, and the inquiry that will learn and initial seed are inquired about in the lump the kind subquery as this kind demand type.
With similar in embodiment one, the initial seed inquiry can have the inquiry of this kind demand type to obtain by artificial selecting, and in the present embodiment, need to obtain equally the similarity between different inquiries, and introduction similar in the computing method of similarity and embodiment one between different inquiries do not repeat them here.The iterative learning device can adopt has supervision or unsupervised machine learning method to obtain arbitrarily, and the present invention no longer does emphasis and describes.
Except top said embodiment one and embodiment two, when obtaining a kind of kind subquery of demand type, can also the inquiry when causing the page of this kind demand type clicked determine.For example: choose the highest N1 of inquiry times inquiry as the kind subquery of this kind demand type from the clicked inquiry of the page that causes this kind demand type, wherein N1 is default positive integer.For example the page that pop music is downloaded, clicked in a large number by inquiries such as " download of Zhou Jielun special edition ", " Zhou Jielun chrysanthemum platform ", " still Fan Texi ", and these inquiries just can be used as the kind subquery of pop music demand type.In addition, can also extract the kind subquery of this kind demand type from a kind of search daily record of vertical search of demand type.Vertical search is the search for certain industry or certain field, as the user in the search of tour site the time, expression be exactly and the relevant demand of travelling and can not be the demand relevant to food and drink.Therefore when the kind subquery that need to obtain with the related needs of travelling, just can directly extract from the search daily record of the search field of travelling and obtain.
Please continue with reference to figure 1.In step S102, be the process of candidate template with Query generalization, with asterisk wildcard, inquiry limited to generate exactly the process of candidate template.
Particularly, the embodiment of step S102 comprises following several, referring to embodiment three, four and five.
Embodiment three:
The asterisk wildcard of classification under this default entity word will be replaced to the part of the corresponding default entity word coupling of corresponding demand type in the kind subquery of corresponding demand type.
For example the kind subquery of music demand type has " soul-stirring caudal flexure step by step " and " the red three cun paradise of tight skill ", default entity word corresponding to music demand type has: soul-stirring (belong to film and television acute title classification), Yan Yidan (belonging to singer's title classification) and three cun paradise (belonging to the song title classification) step by step, just can to distinguish extensive be " [film and television play title] sheet caudal flexure " and " [singer's title] [song title] " two candidate template for inquiry " soul-stirring caudal flexure step by step " and " the red three cun paradise of tight skill ".
Embodiment four:
With the part that is identified by the classification recognition function in the kind subquery of corresponding demand type replace to the classification recognition function the asterisk wildcard of corresponding classification, wherein the classification recognition function is to be used for such other function of identification according to the attribute definition of a classification.
The classification recognition function comprises the name recognition function, Symbol recognition function, English recognition function, digital recognition function, date recognition function, marque recognition function etc., the attribute of the corresponding classification of the expression such as name wherein, symbol, English.Should be appreciated that, under the thought of using a classification of function identification, the classification recognition function is not limited in above kind, and the classification recognition function that every those skilled in the art can realize all should comprise within the scope of the invention.
Classification recognition function in embodiment four also can be combined with the default entity word in embodiment three, can strengthen the correctness of asterisk wildcard in candidate template.
Below the kind subquery of 5 video requirement types:
1, " recent film that article is drilled "
2, " film of pursuing a goal with determination that Ma Yi Li drilled "
3, " film that grandson pari Deng Chao Mr. and Mrs act the leading role together "
4, " model ice is iced the video of packing luggage of oneself bending over "
5, " all HD videos of Michael Jackson "
If the name recognition function is combined with the default vocabulary that comprises star's class instance word, can obtain following candidate template: film, [Star] that the film of pursuing a goal with determination that the recent film that [Star] drills, [Star] drilled, [Star] [Star] Mr. and Mrs act the leading role together oneself bend over to pack luggage all HD videos of video, [F:name], wherein [Star] is the asterisk wildcard that satisfies star's name of default star's class instance word, and [F:name] is the asterisk wildcard of the name that can be identified by the name recognition function.But be in the example of " the where is it film review article that THE SUN ALSO RISES " in another inquiry, " although article " and default star's class instance word coupling, but because the name recognition function is not identified as name with " article " in this inquiry, therefore " article " in this inquiry just can be by not extensive, thereby improved the correctness of candidate template.
The name recognition function can define according to the co-occurrence probabilities of word, for example add up the large-scale corpus resource, the possibility size that occurs as name according to this word of probabilistic determination of a word and context word co-occurrence is just regarded as name with this word when possibility during greater than setting threshold.The classification recognition function of other kinds also can define according to the characteristics of identification types, repeats no more here.
Embodiment five:
On the basis of embodiment three and embodiment four, the contribution degree to corresponding demand type in the kind subquery of corresponding demand type is replaced with for the length asterisk wildcard that limits word length lower than the word that default contribution degree requires.
5 kind subqueries of given example in embodiment four for example, " all high definitions " in " that drills is up-to-date ", " that drilled pursues a goal with determination " in inquiry 2 of inquiry in 1, " Mr. and Mrs act the leading role together " in inquiry 3, " oneself bending over to pack luggage " in inquiry 4, inquiry 5, these words are lower to the contribution degree that judges an inquiry and whether belong to the video requirement type, therefore, these words can be replaced with the length asterisk wildcard." that drills is up-to-date " that for example will inquire about in 1 replaces with [W:1-4], and wherein [W:1-4] means that length is the asterisk wildcard of 1 to 4 word.5 kind subqueries in embodiment four on the basis that embodiment three and embodiment manage, can obtain following candidate template after further the contribution degree of video requirement type being replaced lower than the word of default contribution degree requirement everywhere:
1, [Star] [W:1-4] film
2, [Star] [W:1-5] film
3, [Star] [Star] [W:1-7] film
4, [Star] [W:1-8] video
5, [F:name] [W:1-5] video
Above-mentioned candidate template 1 and candidate template 2 can also be according to certain consolidation strategies, for example the ultimate range with the matching length interval of the length asterisk wildcard in candidate template to be combined merges, and candidate template 1 and candidate template 2 is merged into: [Star] [W:1-5] film.
The contribution degree of word in inquiry to corresponding demand type, cosine distance between the vector that the vector that in can inquiring about by calculating, the word of n-gram granularity consists of and the word of corresponding demand type consist of obtains, the fragment that becomes of the n-gram minimum particle size morphology that refers to independently to be expressed the meaning by n wherein, for example " that drills is up-to-date " be exactly one by the minimum particle size word " is drilled " and " up-to-date " forms 2-gram.Concept about n-gram can with reference to existing various participle techniques, no longer describe in detail at this.The word of determining corresponding demand type also can adopt various prior aries to carry out, and for example manually chooses or excavates etc. in the language material of corresponding demand type, owing to not being emphasis of the present invention, does not repeat them here.
above narration can be found out, by step S102, can obtain the candidate template of corresponding demand type, but, owing to might there be excessively extensive situation in extensive process, as a template " [singer's title] [W:1-4] ", not only can mate " the red three cun paradise of tight skill " such inquiry, can also mate " height of Yan Yidan " such inquiry, suppose that this template is the template of music demand type, obviously, the demand of a rear query express of its coupling is not music type, illustrate and want to obtain enough demand template accurately, choose in the candidate template that also needs to obtain from step S102.Therefore, in step S103, can choose the candidate template that generates in step S102, to obtain final template.
In step S103, candidate template is chosen, can be adopted the mode of sorter to carry out, namely utilize sorter that candidate template is divided into correct template and wrong template, correct template wherein is exactly the final template that will be chosen for corresponding demand type.
Concerning sorter, the most important factor that affects the classification results quality is the feature that candidate template is extracted.In the present invention, the feature that can extract includes but not limited to: click feature, similarity feature and matching capacity feature.Wherein click feature can cause the clicked probability of the page of corresponding demand type for the inquiry of the candidate template covering that characterizes corresponding demand type, the similarity feature is for the general character degree of all candidate template of a candidate template that characterizes corresponding demand type and corresponding demand type, and the matching capacity feature is for the ability of the inquiry of the corresponding demand type of candidate template coupling that characterizes corresponding demand type.
Particularly, the click feature of candidate template W can be used formula (1) expression:
Wherein, the click feature of Click (W) expression W,
All inquiries that expression W covers in the search daily record cause the clicked number of times of corresponding demand type page (URL),
All inquiries that expression W covers in the search daily record cause the number of times that all pages are clicked.
For example, all inquiries that candidate template " [the acute title of film and television] theme song is downloaded " covers in the search daily record comprise " soul-stirring theme song is downloaded step by step ", " download of Water Margin theme song ", the point that causes in these two inquiries hits, what 100 click sensings were arranged is music site, have 5 clicks to point to other websites, " [the acute title of film and television] theme song is downloaded " this candidate template is exactly 100/105 with respect to the click feature value of music demand.and the inquiry that " [the acute title of film and television] online reading " this candidate template covers has " soul-stirring online reading step by step ", " Water Margin online reading ", the point that causes in these two inquiries hits, what only have 3 click sensings is music site, there are 100 clicks to point to other websites (being mainly reading website), obviously, " [the acute title of film and television] online reading " this candidate template is exactly 3/103 with respect to the click feature value of music demand, this template should be very high with respect to the click feature value of reading requirement also to be easy to judge " [the acute title of film and television] online reading " by above-mentioned this method, thereby the demand that reading more may be arranged.
The similarity feature of candidate template W, available formula (2) calculates:
Wherein Similarity (W) represents the similarity feature of W,
Similarity sum between the every other candidate template of expression W and corresponding demand type.
A, B, three candidate template of C are for example arranged under music demand type, the similarity feature Similarity (A) of candidate template A=S (A, B)+S (A, C), in like manner, Similarity (B)=S (B, A)+S (B, C), Similarity (C)=S (C, A)+S (C, B).
Similarity between a candidate template X and another candidate template Y can be by calculating cosine between the term vector that term vector that X obtains and Y obtain apart from obtaining.And the term vector of candidate template X or Y, can adopt various ways to obtain, for example extracting keywords consists of term vector from the inquiry of X or Y coupling, perhaps extracting keywords consists of term vector from the result for retrieval that the inquiry of X or Y coupling causes, the mode that extracts can adopt the any-mode that those skilled in the art expect to carry out, and the present invention does not limit this.
The matching capacity feature of candidate template W, available formula (3) calculates:
Wherein, the matching capacity feature of Match (W) expression W,
The quantity of the inquiry that expression W matches in the search daily record of the inquiry formation of corresponding demand type,
The quantity of the inquiry that expression W matches in the search daily record of the inquiry formation of various demand types.
The search daily record that the inquiry of corresponding demand type consists of refers to the journal file of the inquiry that only records this kind demand type.For example, from the search daily record that a music site is obtained, record should be all the inquiry of music demand type obviously.
Please continue with reference to figure 1, on the line in Fig. 1, the method for the demand of part identification comprises:
Step S201: obtain user's inquiry.
Step S202: determine the final template that is complementary with user's inquiry and the demand that will have as user's inquiry with the corresponding demand type of final template that user's inquiry is complementary in the final template that the method for the described generation demand of preamble template obtains.
In step S202, determine to inquire about with the user the final template that is complementary, can adopt the algorithm of tree construction to mate.In tree construction, each node represents a kind of state, and root node wherein represents original state, and leaf node represents the state of template matches success, and intermediate node represents the intermediateness in matching process.Connect the limit between two nodes, be called the state transitions condition.
Please refer to Fig. 3, Fig. 3 is the schematic diagram of coupling tree in the present invention.For inquiry " three cun paradise 2011 ", original state is 1, due to " three cun paradise " match state jump condition " [song title] ", so transfer to state 3 from state 1, again because " 2011 " match state jump condition " [F:time] " (expression can be identified by the time recognition function), so transfer to state 9 (be leaf node, the match is successful in expression) from state 3 again.The state transitions conditional combination that consists of the state transitions route has just formed the template that is complementary with inquiry.
Because the corresponding demand type of template that is complementary with inquiry is known, after matching process finishes, just can determine that the user inquires about and has the demand consistent with the template that is complementary.
Should be appreciated that, adopt the query tree algorithm to give an example in above given example, be not for matching process of the present invention is limited, in fact, any known matching algorithm can adopt at this, for example adopt the mode of regular expression to mate, because matching algorithm belongs to the technology that those skilled in the art can be known, the present invention no longer is described in detail at this.
Please refer to Fig. 4.Fig. 4 device that to be the device that generates the demand template in the present invention identify with demand the structural representation block diagram of embodiment, wherein, the line bottom is divided into the schematic diagram of the device of generation demand template, line top is divided into the schematic diagram of demand recognition device.As shown in Figure 4, the device that generates the demand template comprises: seed acquiring unit 301, extensive unit 302 and choose unit 303.
Wherein the seed acquiring unit 301, are used for obtaining the kind subquery of demand type.Extensive unit 302, the seed Query generalization that is used for corresponding demand type is the candidate template of corresponding demand type.Choose unit 303, be used for choosing from the candidate template of corresponding demand type the final template of corresponding demand type.
Please refer to Fig. 5, Fig. 5 is the structural representation block diagram of an embodiment of seed acquiring unit in the present invention.As shown in Figure 5, seed acquiring unit 301 comprises: first chooses unit 3011, cluster cell 3012 and determining unit 3013.
Wherein first choose the initial seed inquiry that unit 3011 is used for obtaining default corresponding demand type.Cluster cell 3012 is used for all inquiries that the search daily record is recorded are carried out cluster according to the method for hierarchical clustering.Determining unit 3013 is used for determining a cluster level, make the initial seed inquiry that is no less than predetermined ratio under this level minimum by the poly-inquiry sum that comprises in same class X and at this level lower class X, all inquiries that the class X under this level is comprised are as the kind subquery of corresponding demand type.
Please refer to Fig. 6, Fig. 6 is the structural representation block diagram of another embodiment of seed acquiring unit in the present invention.As shown in Figure 6, seed acquiring unit 301 comprises: second chooses unit 301a and unit 301b.
Wherein second choose the initial seed inquiry that unit 301a is used for obtaining default corresponding demand type.Unit 301b be used for to use the iterative learning device to satisfy the inquiry of preset requirement from search daily record learning and the similarity between the inquiry of described initial seed, and the inquiry that will learn and described initial seed are inquired about in the lump the kind subquery as corresponding demand type.
Except Fig. 5 and mode shown in Figure 6, seed acquiring unit 301 can also cause choosing the highest N1 of inquiry times inquiry as the kind subquery of corresponding demand type in the clicked inquiry of the page of corresponding demand type from the search daily record, perhaps, extract the kind subquery of corresponding demand type from the search daily record of the vertical search that reflects corresponding demand type.
Please continue with reference to figure 4.When extensive unit 302 is the candidate template of corresponding demand type with the seed Query generalization of corresponding demand type, specifically have:
Mode one: will replace to the asterisk wildcard of classification under this default entity word in the kind subquery of corresponding demand with the part of the corresponding default entity word coupling of corresponding demand type, perhaps
Mode two: with the part that is identified by the classification recognition function in the kind subquery of corresponding demand type replace to the classification recognition function the asterisk wildcard of corresponding classification, wherein said classification recognition function is to be used for such other function of identification according to the attribute definition of a classification.
When extensive unit 302 is the candidate template of corresponding demand type with the seed Query generalization of corresponding demand type, comprise further also that on the basis of above dual mode the word that will in the kind subquery of corresponding demand type, the contribution degree of corresponding demand type be required lower than default contribution degree replaces with the length asterisk wildcard for restriction word length.
Choose unit 303 when choosing the final template of corresponding demand type from the candidate template of corresponding demand type, at least one in the following characteristics of the candidate template of the corresponding demand type of foundation carried out:
One, click feature is used for characterizing the clicked probability of the page that inquiry that the candidate template of corresponding demand type covers can cause corresponding demand type.
Two, similarity feature is used for characterizing a candidate template of corresponding demand type in the general character degree of all candidate template of corresponding demand type.
Three, matching capacity feature is used for characterizing the ability of inquiry of the corresponding demand type of candidate template coupling of corresponding demand type.
Particularly, choosing unit 303 adopts following manner to calculate the click feature of the candidate template W of corresponding demand type:
Particularly, choosing unit 303 adopts following manner to calculate the similarity feature of the candidate template W of corresponding demand type:
Particularly, choosing unit 303 adopts following manner to calculate the matching capacity feature of the candidate template W of corresponding demand type:
Wherein, the matching capacity feature of Match (W) expression W,
The quantity of the inquiry that expression W matches in the search daily record of the inquiry formation of corresponding demand type,
The quantity of the inquiry that expression W matches in the search daily record of the inquiry formation of various demand types.
As shown in Figure 4, on line, the demand recognition device of part comprises: inquiry acquiring unit 401 and matching unit 402.Wherein inquire about acquiring unit 401 and be used for obtaining user's inquiry, the final template that matching unit 402 obtains for the device in the described generation demand of preamble template is determined to inquire about with the user the final template that is complementary, and will inquire about the demand that have as the user with the corresponding demand type of final template that user's inquiry is complementary.
The above is only preferred embodiment of the present invention, and is in order to limit the present invention, within the spirit and principles in the present invention not all, any modification of making, is equal to replacement, improvement etc., within all should being included in the scope of protection of the invention.
Claims (22)
1. a method that generates the demand template, is characterized in that, described method comprises:
Obtain the kind subquery of demand type from the search daily record;
Be the candidate template of described demand type with the seed Query generalization of described demand type;
Choose the final template of described demand type from the candidate template of described demand type.
2. method according to claim 1, is characterized in that, the step of obtaining the kind subquery of demand type comprises:
Obtain the initial seed inquiry of default described demand type;
All inquiries of recording in the search daily record are carried out cluster according to the method for hierarchical clustering;
Determine a cluster level, make the initial seed inquiry that is no less than predetermined ratio under this level minimum by the poly-inquiry sum that comprises in same class X and at this level lower class X, all inquiries that the class X under this level is comprised are as the kind subquery of described demand type.
3. method according to claim 1, is characterized in that, the step of obtaining the kind subquery of described demand type comprises:
Obtain the initial seed inquiry of default described demand type;
Use the iterative learning device to satisfy the inquiry of preset requirement from search daily record learning and the similarity between the inquiry of described initial seed, and the inquiry that will learn and described initial seed are inquired about in the lump the kind subquery as described demand type.
4. method according to claim 1, is characterized in that, the step of obtaining the kind subquery of demand type comprises:
Cause daily record choosing the highest N1 of inquiry times inquiry as the kind subquery of described demand type, the positive integer of described N1 for presetting in the clicked inquiry of the page of described demand type from search; Perhaps,
Extract the kind subquery of described demand type from the search daily record of the vertical search of described demand type.
5. method according to claim 1, is characterized in that, the step that is the candidate template of described demand type with the seed Query generalization of described demand type comprises:
The asterisk wildcard of classification under described default entity word will be replaced to the part of the corresponding default entity word coupling of described demand type in the kind subquery of described demand type; Perhaps,
With the part that is identified by the classification recognition function in the kind subquery of described demand type replace to described classification recognition function the asterisk wildcard of corresponding classification, wherein said classification recognition function is to be used for such other function of identification according to the attribute definition of a classification.
6. method according to claim 5, is characterized in that, the step that is the template of described demand type with the seed Query generalization of described demand type further comprises:
Contribution degree to described demand type in the kind subquery of described demand type is replaced with for the length asterisk wildcard that limits word length lower than the word that default contribution degree requires.
7. method according to claim 1, is characterized in that, when choosing the final template of described demand type from the candidate template of described demand type, at least one in the following characteristics of the candidate template of the described demand type of foundation carried out:
Click feature is used for characterizing the clicked probability of the page that inquiry that the candidate template of described demand type covers can cause described demand type;
The similarity feature is used for characterizing the general character degree of all candidate template of a candidate template of described demand type and described demand type;
The matching capacity feature is used for characterizing the ability of inquiry of the described demand type of candidate template coupling of described demand type.
8. method according to claim 7, is characterized in that, the click feature of the candidate template W of described demand type adopts following manner to calculate:
9. method according to claim 7, is characterized in that, the similarity feature of the candidate template W of described demand type adopts following manner to calculate:
10. method according to claim 7, is characterized in that, the matching capacity feature of the candidate template W of described demand type adopts following manner to calculate:
11. the method for a demand identification is characterized in that, described method comprises:
Obtain user's inquiry;
Determine in the final template that the method for the described generation demand of arbitrary claim template obtains in claim 1 to 10 to inquire about with described user the final template that is complementary, and the demand that will have as described user's inquiry with the corresponding demand type of final template that described user's inquiry is complementary.
12. a device that generates the demand template is characterized in that, described device comprises:
The seed acquiring unit is for obtain the kind subquery of demand type from the search daily record;
Extensive unit, the seed Query generalization that is used for described demand type is the candidate template of described demand type;
Choose the unit, be used for choosing from the candidate template of described demand type the final template of described demand type.
13. device according to claim 12 is characterized in that, described seed acquiring unit comprises:
First chooses the unit, is used for obtaining the initial seed inquiry of default described demand type;
Cluster cell is used for all inquiries that the search daily record is recorded are carried out cluster according to the method for hierarchical clustering;
Determining unit, be used for determining a cluster level, make the initial seed inquiry that is no less than predetermined ratio under this level minimum by the poly-inquiry sum that comprises in same class X and at this level lower class X, all inquiries that the class X under this level is comprised are as the kind subquery of described demand type.
14. device according to claim 12 is characterized in that, described seed acquiring unit comprises:
Second chooses the unit, be used for obtaining default described demand type the initial seed inquiry;
Unit be used for to use the iterative learning device to satisfy the inquiry of preset requirement from search daily record learning and the similarity between the inquiry of described initial seed, and the inquiry that will learn and described initial seed are inquired about in the lump the kind subquery as described demand type.
15. device according to claim 12, it is characterized in that, described seed acquiring unit is when obtaining the kind subquery of described demand type, specifically cause daily record in the clicked inquiry of the page of described demand type from search, choose the highest N1 of inquiry times inquiry as the kind subquery of described demand type, described N1 is default positive integer; Perhaps,
Extract the kind subquery of described demand type from the search daily record of the vertical search of described demand type.
16. device according to claim 12, it is characterized in that, when described extensive unit is the candidate template of described demand type at the seed Query generalization with described demand type, specifically the part of corresponding with described demand type default entity word coupling in the kind subquery of described demand type is replaced to the asterisk wildcard of classification under described default entity word; Perhaps,
With the part that is identified by the classification recognition function in the kind subquery of described demand type replace to described classification recognition function the asterisk wildcard of corresponding classification, wherein said classification recognition function is to be used for such other function of identification according to the attribute definition of a classification.
17. device according to claim 16, it is characterized in that, described extensive unit also is used for the word that the kind subquery of described demand type requires lower than default contribution degree the contribution degree of described demand type is replaced with length asterisk wildcard for restriction word length.
18. device according to claim 12, it is characterized in that, described when choosing the unit choosing the final template of described demand type from the candidate template of described demand type, carry out according at least one in the following characteristics of the candidate template of described demand type:
Click feature is used for characterizing the clicked probability of the page that inquiry that the candidate template of described demand type covers can cause described demand type;
The similarity feature is used for characterizing the general character degree of all candidate template of a candidate template of described demand type and described demand type;
The matching capacity feature is used for characterizing the ability of inquiry of the described demand type of candidate template coupling of described demand type.
19. device according to claim 18 is characterized in that, the described unit of choosing adopts following manner to calculate the click feature of the candidate template W of described demand type:
20. device according to claim 18 is characterized in that, the described unit of choosing adopts following manner to calculate the similarity feature of the candidate template W of described demand type:
21. device according to claim 18 is characterized in that, the described unit of choosing adopts following manner to calculate the matching capacity feature of the candidate template W of described demand type:
22. the device of a demand identification is characterized in that, described device comprises:
The inquiry acquiring unit is used for obtaining user's inquiry;
Matching unit, be used for determining in final template that the device in the described generation demand of the arbitrary claim of claim 12 to 21 template obtains the final template that is complementary with described user's inquiry and the demand that will have as described user's inquiry with the corresponding demand type of final template that described user's inquiry is complementary.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201110379335.5A CN103136221B (en) | 2011-11-24 | 2011-11-24 | A kind of method for generating requirement templet, demand know method for distinguishing and its device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201110379335.5A CN103136221B (en) | 2011-11-24 | 2011-11-24 | A kind of method for generating requirement templet, demand know method for distinguishing and its device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN103136221A true CN103136221A (en) | 2013-06-05 |
| CN103136221B CN103136221B (en) | 2017-06-06 |
Family
ID=48496057
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201110379335.5A Active CN103136221B (en) | 2011-11-24 | 2011-11-24 | A kind of method for generating requirement templet, demand know method for distinguishing and its device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103136221B (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104331456A (en) * | 2014-10-31 | 2015-02-04 | 百度在线网络技术（北京）有限公司 | Method and device for mining sort named entities |
| CN104991943A (en) * | 2015-07-10 | 2015-10-21 | 百度在线网络技术（北京）有限公司 | Music searching method and apparatus |
| CN106294476A (en) * | 2015-06-05 | 2017-01-04 | 北京搜狗科技发展有限公司 | A kind of Feature Words Relation acquisition method and device |
| WO2017107738A1 (en) * | 2015-12-24 | 2017-06-29 | 北京国双科技有限公司 | Rule matching method and device |
| CN110209780A (en) * | 2018-08-07 | 2019-09-06 | 腾讯科技（深圳）有限公司 | A kind of question template generation method, device, server and storage medium |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101984423A (en) * | 2010-10-21 | 2011-03-09 | 百度在线网络技术（北京）有限公司 | Hot-search word generation method and system |
| CN102096717A (en) * | 2011-02-15 | 2011-06-15 | 百度在线网络技术（北京）有限公司 | Search method and search engine |
| CN102368260A (en) * | 2011-10-12 | 2012-03-07 | 北京百度网讯科技有限公司 | Method and device of producing domain required template |
-
2011
- 2011-11-24 CN CN201110379335.5A patent/CN103136221B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101984423A (en) * | 2010-10-21 | 2011-03-09 | 百度在线网络技术（北京）有限公司 | Hot-search word generation method and system |
| CN102096717A (en) * | 2011-02-15 | 2011-06-15 | 百度在线网络技术（北京）有限公司 | Search method and search engine |
| CN102368260A (en) * | 2011-10-12 | 2012-03-07 | 北京百度网讯科技有限公司 | Method and device of producing domain required template |
Non-Patent Citations (4)
| Title |
|---|
| 刘亮亮，林乐宇: "《基于查询模板的特定领域中文问答系统的研究与实现》", 《江苏科技大学学报》 * |
| 梁正平，纪震等: "基于语义模板的问答系统研究", 《深圳大学学报理工版》 * |
| 蔡岳，袁津生: "用户行为聚类的搜索引擎算法与实现", 《用户行为聚类的搜索引擎算法与实现》 * |
| 钱兵，王永成等: "面向搜索引擎的自然语言理解的设计与实现", 《计算机应用研究》 * |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104331456A (en) * | 2014-10-31 | 2015-02-04 | 百度在线网络技术（北京）有限公司 | Method and device for mining sort named entities |
| CN104331456B (en) * | 2014-10-31 | 2018-02-09 | 百度在线网络技术（北京）有限公司 | Classification proper name method for digging and device |
| CN106294476A (en) * | 2015-06-05 | 2017-01-04 | 北京搜狗科技发展有限公司 | A kind of Feature Words Relation acquisition method and device |
| CN104991943A (en) * | 2015-07-10 | 2015-10-21 | 百度在线网络技术（北京）有限公司 | Music searching method and apparatus |
| WO2017107738A1 (en) * | 2015-12-24 | 2017-06-29 | 北京国双科技有限公司 | Rule matching method and device |
| US10796077B2 (en) | 2015-12-24 | 2020-10-06 | Beijing Gridsum Technology Co., Ltd. | Rule matching method and device |
| CN110209780A (en) * | 2018-08-07 | 2019-09-06 | 腾讯科技（深圳）有限公司 | A kind of question template generation method, device, server and storage medium |
| CN110209780B (en) * | 2018-08-07 | 2023-03-10 | 腾讯科技（深圳）有限公司 | Question template generation method and device, server and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN103136221B (en) | 2017-06-06 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN101404015B (en) | Automatically generating a hierarchy of terms | |
| CN100465954C (en) | Reinforced clustering of multi-type data objects for search term suggestion | |
| CN103164454B (en) | Keyword group technology and system | |
| CN102508859B (en) | Advertisement classification method and device based on webpage characteristic | |
| CN101876981B (en) | A kind of method and device building knowledge base | |
| CN1871597B (en) | System and method for associating documents with contextual advertisements | |
| CN102567464B (en) | Based on the knowledge resource method for organizing of expansion thematic map | |
| CN102419778B (en) | Information searching method for discovering and clustering sub-topics of query statement | |
| WO2017070656A1 (en) | Video content retrieval system | |
| US20130060769A1 (en) | System and method for identifying social media interactions | |
| CN105095187A (en) | Search intention identification method and device | |
| US20120158703A1 (en) | Search lexicon expansion | |
| JP2005526317A (en) | Method and system for automatically searching a concept hierarchy from a document corpus | |
| CN103605729A (en) | POI (point of interest) Chinese text categorizing method based on local random word density model | |
| CN101727447A (en) | Generation method and device of regular expression based on URL | |
| CN101802776A (en) | Method and apparatus for relating datasets by using semantic vectors and keyword analyses | |
| CN103970748A (en) | Related keyword recommending method and device | |
| CN103136221A (en) | Method capable of generating requirement template and requirement identification method and device | |
| CN101458692A (en) | Strategic material industry knowledge base platform and construct method thereof | |
| CN103778206A (en) | Method for providing network service resources | |
| CN103927177A (en) | Characteristic-interface digraph establishment method based on LDA model and PageRank algorithm | |
| CN105205163A (en) | Incremental learning multi-level binary-classification method of scientific news | |
| CN114997288A (en) | Design resource association method | |
| Machanavajjhala et al. | Collective extraction from heterogeneous web lists | |
| CN103761286A (en) | Method for retrieving service resources on basis of user interest |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |