WO2016107455A1

WO2016107455A1 - Information matching processing method and apparatus

Info

Publication number: WO2016107455A1
Application number: PCT/CN2015/098247
Authority: WO
Inventors: 王涛; 黄鹏; 林锋
Original assignee: 阿里巴巴集团控股有限公司; 王涛; 黄鹏; 林锋
Priority date: 2014-12-29
Filing date: 2015-12-22
Publication date: 2016-07-07
Also published as: CN105808541B; CN105808541A

Abstract

Disclosed is an information matching processing method. The method comprises: acquiring various search key words and various pieces of product information, and search key word and product information feature pairs being constituted of the various search key words and the various pieces of product information in pairs (S101); calculating the correlation of each of the search key word and product information feature pairs, and determining a correlation level of each of the search key word and product information feature pairs according to a correlation calculation result (S102); calculating a pre-estimated click rate of each of the search key word and product information feature pairs, and determining a pre-estimated click rate level corresponding to the pre-estimated click rate of each of the search key word and product information feature pairs by using a quantile (S103); and according to the correlation level and the pre-estimated click rate level, determining a score of each of the search key word and product information feature pairs, the score being used for characterizing a matching degree between the search key word and the product information (S104).

Description

Information matching processing method and device

Technical field

The present invention relates to the field of data processing technologies, and in particular, to an information matching processing method and apparatus.

Background technique

With the development of computer and Internet technologies, e-commerce websites have developed rapidly. In an e-commerce website, a large amount of data or products are usually stored. In order to improve the efficiency of a user searching for a product of interest, the website server often recommends a product matching the search term to the user according to the search term input by the user. Among the products recommended to the user that match the search term, some products that match the search term with high quality, good quality, and are promoted are often recommended to the user. In order to increase the sales volume of products, sellers often choose good quality products for advertising promotion. When the seller promotes the advertisement, the corresponding search keyword needs to be purchased for the published product information. If the product information published by the seller matches the search keyword, the probability that the product is searched by the user is larger, and the buyer user also has a higher probability. It is more likely to find products that match the search term, so that you can get useful information in the information ocean.

Therefore, accurately determining the matching degree between the product information and the search term can not only improve the effectiveness of the product promotion by the seller user, but also reduce the data interaction between the client and the server brought by the buyer user repeatedly searching for the product, thereby improving the user experience and improving at the same time. Server performance.

The method for judging the matching degree between product information and search words existing in the prior art is often to calculate the correlation between the search term and the advertisement product, determine the matching degree between the search term and the published product information according to the relevance score, and recommend the seller to purchase the match. High search keywords.

However, this method existing in the prior art only considers the relevance of the search term to the advertisement product, and does not consider the degree to which the advertisement product is preferred by the user, and thus the matching calculated by this is not accurate. Inaccurate matching calculation results not only lead to the seller not effectively promoting its products, but also the products recommended by the website to the buyer users are not products that exactly match their needs and interests. The buyer has to repeatedly search to get the real products. Interested products, which increase the data interaction between the client and the server, increase the data processing load of the server, and reduce the processing performance of the server. And seriously occupy valuable Internet bandwidth resources.

Summary of the invention

In order to solve the above technical problem, the present invention discloses an information matching processing method and device, which can improve the objectivity and accuracy of information matching, improve the user experience, reduce the data processing load of the server, improve the processing performance of the server, and save. Valuable Internet bandwidth resources.

The technical solutions are as follows:

According to a first aspect of the embodiments of the present invention, a product information matching processing method is disclosed, the method comprising:

Obtaining each search keyword and product information, and combining the search keywords and product information into a search keyword and a product information feature pair;

Calculating a correlation between each of the search keyword and the product information feature pair, and determining a correlation gear position of each of the search keyword and the product information feature pair according to the correlation calculation result;

Calculating an estimated click rate of each of the search keyword and the product information feature pair, and using the quantile to determine an estimated click rate file corresponding to the estimated click rate of each of the search keyword and the product information feature pair;

And determining, according to the correlation gear position and the estimated click rate gear position, a score of each of the search keyword and product information feature pairs, the score being used to represent a degree of matching between the search keyword and product information.

According to a second aspect of the embodiments of the present invention, a product information matching processing apparatus is disclosed, the apparatus comprising:

An obtaining unit, configured to obtain each search keyword and product information, and combine the search keywords and product information into a search keyword and a product information feature pair;

a correlation gear determining unit, configured to calculate a correlation between each of the search keyword and the product information feature pair, and determine a correlation gear position of each of the search keyword and the product information feature pair according to the correlation calculation result;

The estimated click rate gear determining unit is configured to calculate an estimated click rate of each of the search keyword and the product information feature pair, and determine, by using the quantile point, the pair of the search keyword and the product information feature Estimated click rate rate corresponding to the estimated click rate;

a matching determining unit, configured to determine, according to the correlation gear position and the estimated click rate gear position, a score of each of the search keyword and product information feature pairs, wherein the score is used to represent the search keyword and The degree of matching of product information.

An advantageous aspect that can be achieved by an aspect of the embodiments of the present invention is that the method and apparatus provided by the present invention not only considers the relevance of the search keyword and the product information, but also considers the degree of matching between the search keyword and the product information. The degree to which the product is preferred by the user, the estimated click-through rate factor that can objectively reflect the degree to which the product is preferred by the user is used to calculate the estimated click-through rate, and is also determined according to a preset proportional rule (for example, a normal distribution rule). The click rate file corresponding to the probability that the advertisement product is clicked by the user under the search keyword, and the matching degree of the search keyword and the product information is comprehensively determined by the relevant gear position and the click rate gear position, thereby obtaining more accurate matching. result. Therefore, not only can the seller user improve the effectiveness of the product promotion, but also reduce the data interaction between the client and the server brought by the buyer user repeatedly searching for the product, improve the user experience, reduce the data processing load of the server, and improve the processing of the server. Performance, saving valuable internet bandwidth resources.

DRAWINGS

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the embodiments or the description of the prior art will be briefly described below. Obviously, the drawings in the following description are only It is a few embodiments described in the present invention, and other drawings can be obtained from those skilled in the art without any inventive effort.

1 is a schematic flowchart of an information matching processing method according to an embodiment of the present invention;

2 is a schematic diagram of a standard normal distribution quantile table according to an embodiment of the present invention;

FIG. 3 is a schematic diagram of an estimated click rate rate distribution according to an embodiment of the present invention; FIG.

FIG. 4 is a schematic diagram of an information matching processing apparatus according to an embodiment of the present invention.

detailed description

In order to make those skilled in the art better understand the technical solutions in the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. The embodiment is only a part of the embodiment of the invention, not the entire implementation. example. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative efforts shall fall within the scope of the present invention.

The invention discloses an information matching processing method and device, which not only considers the correlation between the search keyword and the product information, but also considers the degree to which the product is favored by the user, and introduces an estimated click that reflects the degree of the product being favored by the user. The rate factor is used to calculate the estimated click rate, and according to the normal distribution law, the click rate file corresponding to the probability that the advertising product is clicked by the user under the search keyword is determined, and the relevant gear position and the click rate gear are comprehensively determined. Search for the match between keywords and product information to get more accurate matching results.

In an application scenario of the present invention, in an e-commerce website, a seller needs to purchase a search keyword to promote an advertisement product, and the method provided by the embodiment of the present invention can be applied to a website server for judging a search keyword and a seller. The degree of matching of the published product information, so as to recommend to the seller to purchase the search keyword with high matching degree, so as to improve the effectiveness of the seller user to promote the product, and further improve the probability that the seller user product is clicked by the buyer user; on the other hand, Improve the efficiency of buyers and users searching for products, reduce the data interaction between the client and the server brought by the repeated search of the product by the buyer, improve the user experience, reduce the data processing load of the server, improve the processing performance of the server, and save valuable Internet bandwidth. Resources.

FIG. 1 is a schematic flowchart diagram of an information matching processing method according to an embodiment of the present invention.

S101. Acquire each search keyword and product information, and combine the search keywords and product information into a search keyword and a product information feature pair.

Usually for sellers, the products they operate are diverse and may belong to different categories. In this case, the seller's product information can be processed separately to obtain one or more words that can describe their product information, and search with Key words two-two composition search keywords and product information feature pairs. For example, the seller's product information includes MP3 player, iphone6, Note4, headphones, and so on. Search keywords for mobile phones, the composition of the search keywords and product information features include (mobile phones, MP3 players), (mobile phones, iphone6), (mobile phones, Note4), (mobile phones, headsets). Of course, the above is merely illustrative and is not to be construed as limiting the invention. The product information may specifically be advertisement product information.

It should be noted that, before performing step S102 and step S103, the search keywords and product information may be preprocessed, and the preprocessing includes performing words required for matching each feature. Extraction of semantic features. The specific processing manner may be various and is not limited herein.

In addition, there is no necessary sequential execution order between step S102 and step S103, and the two may be executed in parallel or may be performed in reverse.

S102. Calculate a correlation between each of the search keyword and the product information feature pair, and determine a correlation gear position of each of the search keyword and the product information feature pair according to the correlation calculation result.

Among them, the correlation calculation is mainly obtained by searching the keyword category and the relevance of the article of the advertisement product and the text correlation. Among them, the category relevance refers to the degree of matching between the click category of the search keyword and the category of the advertising product; the text relevance includes various aspects, mainly referring to the matching degree between the core word of the search keyword and the core word of the advertisement product title. And the attribute matching attribute in the search keyword and the attribute matching in the description of the advertisement product, and the comprehensive category matching and the text matching can obtain the relevance score.

In a specific implementation, the step S102 may include: performing matching determination on each feature of the search keyword and the product information feature pair; and determining the search keyword and product information feature according to the matching judgment result of each feature The relevant gear position.

In a specific implementation, when performing the correlation calculation, the search keyword and the product information feature perform matching determination of each feature: at least one of a category feature matching judgment and a text feature matching judgment.

Further, the category feature matching is determined to determine whether the search keyword and the product information belong to the same category. In a specific implementation of the present invention, the category feature matching judgment generally refers to a category judgment according to the meaning of the text. If the search keyword category is the same as the category in which the product information is published, the result of the category feature matching judgment is “Yes”; otherwise, the result of the category feature matching judgment is “No”. Wherein, a special case in which the result of the category feature matching judgment is “No” is that the search keyword has no category, and for a search keyword without a category, the long tail is generally serious, and the long tail is very Search keywords that are less searched by users. For example, if the search keyword is “mp3” and the published product is “audio player”, the two belong to the same category, and the result of the category feature matching judgment is “yes”. The search keyword is “mp3”, and the published product is “radio”, the two do not belong to the same category, and the result of the category feature matching judgment is “no”.

Further, the text feature matching is determined to determine whether the search keyword and the text content of the published product information are associated. Specifically, the text feature matching judgment of the present invention includes: an exact matching judgment, a partial matching judgment, a central word matching judgment, a central word complete matching judgment, and a hidden word. At least one of a matching judgment and a reverse preposition matching judgment. Of course, the text feature matching judgment may further include a method of extracting the text feature vector and calculating the similarity of the text vector by using the cosine angle formula. The invention is not limited thereto.

After determining the matching of the features according to the search keyword and the product information feature, the correlation gears of the search keyword and the product information feature pair may be determined according to the matching judgment result of the features. In the present invention, the correlation gear is divided into three grades of excellent difference.

As shown in Table 1, a schematic description of the division of the relevant gears, of course, other gear division methods may also be adopted, which are not limited herein.

Table 1

S103. Calculate an estimated click rate of each of the search keyword and the product information feature pair, and use the quantile to determine an estimated click rate file corresponding to the estimated click rate of each of the search keyword and the product information feature pair. .

In a specific implementation, the step S103 may include: predetermining a scale factor corresponding to each gear position of the estimated click rate gear; determining a value of the quantile according to the proportional coefficient; and according to the search keyword and the product information feature The estimated click rate of the pair and the value of the quantile determine the gear range in which the estimated click rate is located.

Preferably, the quantile is a normal distribution quantile.

The following is described in detail with reference to an example.

The standard normal distribution quantile is first introduced. The standard normal distribution, also known as the Gaussian distribution, is a normal distribution with 0 as the mean and 1 as the standard deviation, denoted as N(0,1), which is a probability distribution curve showing a bell shape, which is small at both ends. The middle is large, and the total area under the curve is 1, which is defined as: if it changes randomly The quantity X obeys a probability distribution with a positional parameter of μ and a scale parameter of σ, which is recorded as:

X～N(μ,σ ² ) (1)

Its probability density function is

Then, f is subject to 0 as the average and 1 is the standard normal distribution of the standard deviation.

The normal distribution quantile is used to characterize the rule of the curve area under the normal distribution. The upper α quantile of the standard normal distribution is defined as: X~N(0,1), for any given α, (0 <α<1), a point Za that satisfies P(X>Z _a )=α is referred to as an upper α-fraction of _a standard normal distribution. For example, look at the diagram of the normal distribution table shown in Figure 2. When Z _a =1, find α = 0.158655.

The commonly used quantile of a normal distribution has the following rules:

The area of 68.268949% under the function curve is within a standard deviation of the average.

The area of 95.449974% is within the range of two standard deviations of 2σ around the mean.

The area of 99.730020% is within the range of three standard deviations of 3σ around the average.

The area of 99.993666% is within the range of four standard deviations of 4σ around the mean.

The invention divides the gear position of the estimated click rate by applying the normal distribution law.

Among them, the estimated click rate eCTR is to establish a mathematical probability model through multiple exposures and clicks in history, and use the model to predict whether a future exposure will produce a click. The final value is given to a certain Under the word, the probability that a product is clicked by the user after exposure, therefore, it is a value between 0 and 1. The larger the value, the more likely it is to be clicked.

The eCTR is estimated using the industry standard LR model, which includes feature extraction and model training. The calculating the estimated click rate of each of the search keyword and the product information feature pair comprises: extracting features of the search keyword and the product information feature pair, and obtaining feature weights corresponding to each feature according to the training model; The estimated click rate is calculated by the feature and the feature weight corresponding to the feature.

The feature extraction feature includes one or any combination of the following: text information of the search keyword, category information of the search keyword, title of the product information, and the The attribute of the product information, the relevance of the search keyword to the product information.

Then, after the feature weight is obtained through the model training, the estimated click rate eCTR of the advertisement pair (Query, offer) can be estimated. Among them, Query is the search keyword and offer is the product information.

The LR model belongs to the generalized linear model, which is obtained by changing the linear model through the Logistic formula.

Where w _i is the feature weight, f _i is the feature value, and y is the final calculated estimated click rate. The formula limits the final result to (0, 1), which coincides with the click probability.

In theory, the estimated eCTR should conform to the Gaussian normal distribution. The eCTR is used to classify the eCTR of the ad pair using the keyword and the global dimension. The eCTR of each ad pair will fall on the corresponding interval of the overall eCTR distribution. The interval determines the estimated click rate file for the pair of ads. According to the method for estimating the click rate rate provided by the present invention, the rating of the advertising products of most customers is averaged, and the advertising products of a small number of customers are at a better or worse level.

In the embodiment of the present invention, according to actual business analysis and empirical determination, it is determined that the estimated click rate gears are divided into good, medium, and poor gears, and the scale coefficients corresponding to each gear are 3:4:3, that is, The proportion of good advertising products is 30%, the proportion of advertising products in the gear position is 40%, and the proportion of advertising products with poor gear is 30%. The corresponding score is 5 stars. 4 stars and 3 stars. For details, please refer to Figure 3, which is a schematic diagram of the estimated click rate. Wherein, the abscissa is the estimated click rate value, the ordinate is the frequency, and the curve area corresponds to the probability (ie, the proportional value).

In the specific implementation, when the global or keyword dimension estimated click-through rate eCTR distribution is divided according to the ratio of 3:4:3, the distribution area under the curve that deviates from the average range is 0.4, and the two sides are symmetrically 0.3, according to the law of common distribution of commonly used quantile points:

Where μ is the average, σ is the standard deviation, and Z _a is the normal distribution quantile.

That is to say, after determining the scale factor corresponding to each gear position of the estimated click rate gear, it can be rooted The value of the normal distribution quantile is determined according to the scale factor.

Suppose that Figure 3 obeys the standard normal distribution, that is, X~N(0,1). For any given α, (0<α<1), the point Zα that satisfies P(X>Zα)=α is the standard normal state. The upper α-fraction of the distribution, Z(1-α) corresponds to the lower α-point.

Zα is a value. When X~N(0,1), then P(X>Zα)=α. For example, it is found that α is found in the normal distribution table, and Zα is detected correspondingly. For example, to check the value of Z0.025, it is necessary to check the Z value corresponding to 1-0.025=0.975, and find the normal distribution table shown in Fig. 2, just to find that the Z value corresponding to 0.9750 is 1.96, so Z0.025=1.96 To check the alpha value corresponding to Zα=1.96, you need to check 1.96 first, corresponding to 0.975, 1-0.975=0.025= is the alpha value.

It can be seen from Fig. 3 that a1 and a2 correspond to two quantized points of the standard normal distribution, respectively, and can be respectively corresponding to Zα1 and Zα2 by the scale values in Fig. 3, and Zα1 can be obtained by the above method. The value of Zα2, under the standard normal distribution, Zα1 corresponds to the α-fraction point, and Zα2 corresponds to the lower α-point.

In the specific implementation, when the estimated click rate gears are divided according to the ratio of 3:4:3, the distribution area under the curve with a certain range deviating from the average is 0.4, and the left and right sides are symmetric. Then each is 0.3, then the area of the right curve corresponding to the a2 quantile in the standard normal distribution bitmap of Figure 3 is 0.3, that is, the value of Z _0,3 is checked, that is, the Z corresponding to 1-0.3=0.7 needs to be checked. value. Looking up the normal distribution quantile table shown in Fig. 2, the Z value corresponding to 0.7 is 0.52, then Z _0,3 = 0.52, that is, a2 is 0.52; similarly, it can be determined that the value of a1 is -0.52. A2 and a1 correspond to the two quantiles in the normal distribution. Of course, the values of the normal distribution quantile points Zα1 and Zα2 can also be calculated according to the formula (4). Since Figure 3 satisfies the standard normal distribution quantile, there is X~N(0,1), that is, μ is equal to 0, and σ is equal to 1, which is calculated by the formula (4), Z _a = ±0.5, corresponding to Figure 3 , ie a1=-0.5, a2=0.5.

The estimated click rate is in accordance with the general normal distribution law. Corresponding to the general normal distribution (μ is not equal to 0, σ is not equal to 1), the corresponding quantile can be approximated by the law of the normal distribution quantile, and the general normal distribution corresponds to the ratio 3:4: The quantile of 3 thus gives the following formula:

Where μ is the average and σ is the standard deviation. Among them, μ and σ can be calculated by actual data samples. Specifically, after obtaining the estimated click-through rate value, the average value μ of all the estimated click-through rates and the corresponding variance σ can be obtained. For the specific calculation method, refer to the existing method. Then, based on the average value μ and the variance σ, the value of the general normal distribution quantile is obtained according to the formula (4).

After determining the value of the general normal distribution quantile, the range of the estimated click rate may be determined according to the estimated click rate and the value of the normal distribution quantile. For example, when the estimated hit rate is (0, μ-σ/2) according to the standard normal distribution quantile table, the corresponding estimated click rate is the difference; the estimated click rate belongs to (μ-σ/ 2, μ+σ/2), the corresponding estimated click rate is medium; when the estimated click rate is [μ+σ/2,1), the corresponding estimated click rate is it is good.

It should be noted that the above description is made by taking the proportional coefficient of 3:4:3 as an example. When the determined proportional coefficient is other ratios, the calculation can be performed with reference to the idea of the above method.

S104. Determine, according to the correlation gear position and the estimated click rate gear position, a score of each of the search keyword and the product information feature pair, where the score is used to represent the matching degree between the search keyword and the product information. .

In a specific implementation, the specific calculation method of the score may be various, for example, a weighted average method is used to obtain a score or other implementation manner, which is not limited by the present invention.

Referring to Table 2, an implementation of star rating.

Table 2

Among them, according to the actual business analysis, you can choose to use the ratio of 3:4:3 with good difference to classify the ads with good correlation, which are 5 stars, 4 stars and 3 stars respectively. Good The advertisements are divided according to the ratio of 1:1, corresponding to 2 stars and 1 star respectively. The division of excellent advertisement pairs is shown in Table 2. The good advertisements are relatively simple because of only two files, and the distribution mean points are taken. Yes, good advertising is centered, greater than the mean is 2 stars, less than the average of 1 star.

In the embodiment of the present invention, the correlation calculation and the estimated click rate are combined to calculate the matching degree between the search keyword and the advertisement product, not only to inform the seller user of the advertisement quality and the matching degree, but also objectively reflect that the buyer user searches for the product on the website. The probability that the advertising product will be clicked by the buyer, the higher the rating star rating, the higher the ranking, the greater the possibility of the buyer clicking, the more exposure and feedback, and the return on investment of the advertiser. The higher the rate, the more effective the seller is to promote the product. For website buyers, advertisers' optimization of advertisements will lead to an increase in product quality. The direct result is that users' experience on the website will become better, and the data interaction between the client and the server will be less. The data processing load of the server improves the processing performance of the server and saves valuable Internet bandwidth resources.

FIG. 4 is a schematic diagram of a product information matching processing apparatus according to an embodiment of the present invention.

A product information matching processing device 400, the device comprising:

The obtaining unit 401 is configured to obtain each search keyword and product information, and combine the search keywords and product information into a search keyword and a product information feature pair.

The correlation gear determining unit 402 is configured to calculate a correlation between each of the search keyword and the product information feature pair, and determine a correlation gear position of each of the search keyword and the product information feature pair according to the correlation calculation result.

The estimated click rate gear determining unit 403 is configured to calculate an estimated click rate of each of the search keyword and product information feature pairs, and use the quantile to determine an estimate of each of the search keywords and product information feature pairs. The estimated click rate file corresponding to the click rate.

The matching determining unit 404 is configured to determine a score of each of the search keyword and the product information feature pair according to the relevance gear and the estimated click rate gear, the score being used to represent the search keyword The degree of matching with product information.

Further, the estimated click rate rate determining unit includes an estimated click rate calculating subunit and a gear determining subunit, wherein the estimated click rate calculating subunit comprises:

a model establishing subunit for performing feature extraction on the search keyword and the product information feature pair Taking, according to the training model, obtaining feature weights corresponding to each feature;

The calculating subunit is configured to calculate the estimated click rate by using the extracted feature and the feature weight corresponding to the feature.

Further, the feature extracted by the model establishing subunit includes one or any combination of the following: text information of the search keyword, category information of the search keyword, title of the product information, The attribute of the product information, the relevance of the search keyword to the product information.

Further, the estimated click rate gear determining unit includes an estimated click rate calculating subunit and a gear determining subunit, wherein the gear determining unit includes:

a proportional coefficient determining subunit, configured to predetermine a scale factor corresponding to each gear position of the estimated click rate gear position;

a quantile determining subunit for determining a value of the quantile according to the scaling factor;

a gear interval determining subunit, configured to determine, according to the estimated click rate of each of the search keyword and the product information feature pair and the value of the quantile, the gear range in which the estimated click rate is located.

Wherein, the quantile is a normal distribution quantile.

Further, the correlation gear determining unit includes:

a feature matching sub-unit, configured to perform matching matching between the search keyword and the product information feature pair;

Determining a subunit, configured to determine a correlation gear of the search keyword and the product information feature pair according to the matching judgment result of the each feature.

Further, the matching determination of each feature performed by the feature matching subunit includes: at least one of a category feature matching judgment and a text feature matching judgment;

The category feature matching is determined to determine whether the search keyword and the product information belong to the same category;

The text feature matching is determined to determine whether the search keyword and the text content of the product information are associated.

The functions of the above units may correspond to the processing steps of the above method described in detail in FIG. 1, and details are not described herein again. It should be noted that, due to the detailed description of the method embodiments, the device is implemented. The description of the examples is relatively simple, and those skilled in the art will appreciate that the device embodiments of the present invention can be constructed with reference to method embodiments. Other implementations obtained by those skilled in the art without creative efforts are within the scope of the present invention.

It should be understood by those skilled in the art that the above embodiments of the method and the device are exemplified, and the above is not considered as a limitation of the present invention. Other implementations obtained by those skilled in the art without creative efforts belong to the present invention. The scope of protection.

It should be noted that, in this context, relational terms such as first and second are used merely to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply such entities or operations. There is any such actual relationship or order between them. Furthermore, the term "comprises" or "comprises" or "comprises" or any other variations thereof is intended to encompass a non-exclusive inclusion, such that a process, method, article, or device that comprises a plurality of elements includes not only those elements but also Other elements, or elements that are inherent to such a process, method, item, or device. An element that is defined by the phrase "comprising a ..." does not exclude the presence of additional equivalent elements in the process, method, item, or device that comprises the element. The invention may be described in the general context of computer-executable instructions executed by a computer, such as a program module. Generally, program modules include routines, programs, objects, components, data structures, and the like that perform particular tasks or implement particular abstract data types. The invention may also be practiced in distributed computing environments where tasks are performed by remote processing devices that are connected through a communication network. In a distributed computing environment, program modules can be located in both local and remote computer storage media including storage devices.

The various embodiments in the specification are described in a progressive manner, and the same or similar parts between the various embodiments may be referred to each other, and each embodiment focuses on the differences from the other embodiments. In particular, for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the description of the method embodiment. The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separate, and the components displayed as units may or may not be physical units, ie may be located A place, or it can be distributed to multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of the embodiment. Those of ordinary skill in the art can understand and implement without any creative effort. as above It is to be understood that the present invention may be modified and modified without departing from the principles of the invention. The scope of protection of the present invention.

Claims

An information matching processing method, characterized in that the method comprises:

Obtaining each search keyword and product information, and combining the search keywords and product information into a search keyword and a product information feature pair;

Calculating a correlation between each of the search keyword and the product information feature pair, and determining a correlation gear position of each of the search keyword and the product information feature pair according to the correlation calculation result;

Calculating an estimated click rate of each of the search keyword and the product information feature pair, and using the quantile to determine an estimated click rate file corresponding to the estimated click rate of each of the search keyword and the product information feature pair;

And determining, according to the correlation gear position and the estimated click rate gear position, a score of each of the search keyword and product information feature pairs, the score being used to represent a degree of matching between the search keyword and product information.
The method according to claim 1, wherein said calculating a predicted click rate of each of said search keyword and product information feature pairs comprises:

Feature extraction is performed on the search keyword and product information feature pairs, and feature weights corresponding to each feature are obtained according to the training model;

The estimated click rate is calculated using the extracted features and the feature weights corresponding to the features.
The method according to claim 2, wherein said extracted features comprise one or any combination of the following: text information of said search keyword, category information of said search keyword, The title of the product information, the attribute of the product information, and the relevance of the search keyword to the product information.
The method according to claim 1, wherein said using the quantile to determine an estimated click rate file corresponding to an estimated click rate of each of said search keyword and product information feature pairs comprises:

Predetermining the scale factor corresponding to each gear position of the estimated click rate gear;

Determining a value of the quantile according to the scale factor;

And determining, according to the estimated click rate of each of the search keyword and product information feature pairs and the value of the quantile, the gear range in which the estimated click rate is located.
The method of claim 4 wherein said quantile is a normal distribution quantile.
The method according to claim 1, wherein the correlation between each of the search keyword and the product information feature pair is calculated, and the correlation file of each of the search keyword and the product information feature pair is determined according to the correlation calculation result. Bits include:

Matching the search keyword and the product information feature to perform matching determination of each feature;

And determining a correlation gear of the search keyword and the product information feature pair according to the matching judgment result of the features.
The method according to claim 6, wherein the matching determination of each feature comprises: at least one of a category feature matching determination and a text feature matching determination;

The category feature matching is determined to determine whether the search keyword and the product information belong to the same category;

The text feature matching is determined to determine whether the search keyword and the text content of the product information are associated.
An information matching processing device, characterized in that the device comprises:

An obtaining unit, configured to obtain each search keyword and product information, and combine the search keywords and product information into a search keyword and a product information feature pair;

a correlation gear determining unit, configured to calculate a correlation between each of the search keyword and the product information feature pair, and determine a correlation gear position of each of the search keyword and the product information feature pair according to the correlation calculation result;

The estimated click rate gear determining unit is configured to calculate an estimated click rate of each of the search keyword and the product information feature pair, and use the quantile to determine an estimated click with each of the search keyword and product information feature pairs. The estimated click rate rate corresponding to the rate;

a matching determining unit, configured to determine, according to the correlation gear position and the estimated click rate gear position, a score of each of the search keyword and product information feature pairs, wherein the score is used to represent the search keyword and The degree of matching of product information.
The apparatus according to claim 8, wherein said estimated click rate rate determining unit comprises an estimated click rate calculating subunit and a gear determining subunit, wherein said estimated click rate calculating subunit comprises :

a model establishing subunit, configured to perform feature extraction on the search keyword and the product information feature pair, and obtain a feature weight corresponding to each feature according to the training model;

The calculating subunit is configured to calculate the estimated click rate by using the extracted feature and the feature weight corresponding to the feature.
The apparatus according to claim 9, wherein the feature extracted by the model establishing subunit comprises one or any combination of the following: text information of the search keyword, class of the search keyword The title information, the title of the product information, the attribute of the product information, and the relevance of the search keyword to the product information.
The apparatus according to claim 8, wherein the estimated click rate position determining unit comprises an estimated click rate calculating subunit and a gear determining subunit, wherein the gear determining unit comprises:

a proportional coefficient determining subunit, configured to predetermine a scale factor corresponding to each gear position of the estimated click rate gear position;

a quantile determining subunit for determining a value of the quantile according to the scaling factor;

a gear interval determining subunit, configured to determine, according to the estimated click rate of each of the search keyword and the product information feature pair and the value of the quantile, the gear range in which the estimated click rate is located.
The apparatus of claim 11 wherein said quantile is a normal distribution quantile.
The apparatus according to claim 8, wherein the correlation gear determining unit comprises:

a feature matching sub-unit, configured to perform matching matching between the search keyword and the product information feature pair;

Determining a subunit, configured to determine a correlation gear of the search keyword and the product information feature pair according to the matching judgment result of the each feature.
The device according to claim 13, wherein the matching determination of each feature performed by the feature matching sub-unit comprises: at least one of a category feature matching determination and a text feature matching determination;

The category feature matching is determined to determine whether the search keyword and product information belong to the same Category

The text feature matching is determined to determine whether the search keyword and the text content of the product information are associated.