A method and apparatus for information recommendation

Abstract

The specification discloses a method and device for information recommendation. After a service platform receives a search request sent by a user, the search information carried in the search request is used to determine each candidate recommendation information corresponding to the search information, and a first feature model is used to determine a feature vector corresponding to the search information. Then, for each candidate recommendation information, at least one comment information associated with the search information is determined from each comment information as target comment information corresponding to the candidate recommendation information according to the feature vector corresponding to the search information and the feature vector of each comment information corresponding to the candidate recommendation information, the feature vector corresponding to the candidate recommendation information is determined according to the target comment information corresponding to the candidate recommendation information, and finally, the to-be-recommended information is determined from each candidate recommendation information and recommended to the user, thereby improving the accuracy of information recommendation for the user.

Description

TECHNICAL FIELD

[0001] The present specification relates to the technical field of computer, and particularly relates to a method and device for information recommendation. BACKGROUND

[0002] With the popularization of Internet technology, users can search in an online website to view information that the users want to search, so as to facilitate the users to perform the business that the users want to perform online.

[0003] For example, a third-party consumption review website contains information of a plurality of merchants and comments under each merchant, and a user can search for "a park where cherry blossoms can be viewed" through the website. The website can recommend the merchants related to the search information (i.e., a park where cherry blossoms can be viewed) to the user and show the comments related to the search information under the merchants to the user.

[0004] Taking this scenario as an example, in the prior art, the third-party consumption review website can determine the merchants to be recommended to the user according to the relevance between all the comments under each merchant and the search information, and recommend the merchants and part of the comments of the merchants to the user, but this way cannot accurately reflect the relevance between the merchants and the search information, so the merchants recommended to the user by this way are relatively inaccurate.

[0005] Therefore, how to improve the accuracy of information recommendation to users is a problem to be solved. SUMMARY

[0006] The present specification provides a method and device for information recommendation to partially solve the above problems in the prior art.

[0007] The present specification adopts the following technical solutions:

[0008] The present specification provides a method for information recommendation, comprising:

[0009] receiving a search request sent by a user;

[0010] determining each candidate recommendation information corresponding to the search information according to the search information carried in the search request, and determining a feature vector corresponding to the search information through a pre-trained first feature model;

[0011] For each candidate recommendation information, at least one comment information associated with the search information is determined from each comment information corresponding to the candidate recommendation information as target comment information corresponding to the candidate recommendation information according to the feature vector corresponding to the search information and the feature vector of each comment information corresponding to the candidate recommendation information determined, and the feature vector corresponding to the candidate recommendation information is determined according to the target comment information corresponding to the candidate recommendation information.

[0012] According to the feature vectors corresponding to each candidate recommendation information, a to-be-recommended information is determined from the candidate recommendation information, and the to-be-recommended information is recommended to the user.

[0013] Optionally, the first feature model is trained, and specifically includes:

[0014] A training sample is obtained.

[0015] Text information contained in the training sample is input into a preset second feature model to obtain a semantic analysis result for the text information.

[0016] The second feature model is trained with the optimization target of minimizing the difference between the labeled information contained in the training sample and the semantic analysis result.

[0017] The first feature model is constructed according to the network structure of the trained second feature model, and the number of neural network layers contained in the constructed first feature model is less than that of the trained second feature model.

[0018] Historical search information is obtained, and the first feature model is trained through the historical search information, wherein the feature vectors determined by the trained first feature model and the trained second feature model for the same search information are similar.

[0019] Optionally, the first feature model is trained, and specifically includes:

[0020] Historical search information, historical comment information, and a labeled similarity between the historical search information and the historical comment information are obtained.

[0021] The historical search information is input into the first feature model to obtain a feature vector corresponding to the historical search information, and the historical comment information is input into a preset third feature model to obtain a feature vector corresponding to the historical comment information.

[0022] The first feature model and the third feature model are trained with the optimization target of minimizing the difference between the similarity of the feature vector corresponding to the historical search information and the feature vector corresponding to the historical comment information and the labeled similarity.

[0023] Optionally, the feature vector corresponding to the candidate recommendation information is determined according to the target comment information corresponding to the candidate recommendation information, and specifically includes:

[0024] determining, for each target review information of the candidate recommendation information, a similarity between the feature vector corresponding to the search information and a feature vector corresponding to the target review information, as a similarity corresponding to the target review information;

[0025] determining, according to the similarity corresponding to each target review information of the candidate recommendation information, a correlation parameter corresponding to the candidate recommendation information, the correlation parameter being used to represent the similarity between the target review information of the candidate recommendation information, and / or the similarity between the target review information of the candidate recommendation information and the search information as a whole;

[0026] determining, according to the similarity corresponding to each target review information of the candidate recommendation information, the feature vector corresponding to each target review information of the candidate recommendation information, and the correlation parameter, a feature vector corresponding to the candidate recommendation information.

[0027] Optionally, the information to be recommended is recommended to the user, and specifically includes:

[0028] determining, according to the target review information corresponding to the information to be recommended, additional information for the information to be recommended, the additional information including at least one of an information theme and an information abstract corresponding to the information to be recommended;

[0029] recommending the information to be recommended and the additional information to the user.

[0030] Optionally, according to the feature vector corresponding to the search information and the feature vector of each review information corresponding to the candidate recommendation information, at least one review information associated with the search information is determined from each review information corresponding to the candidate recommendation information, as target review information corresponding to the candidate recommendation information, and specifically includes:

[0031] determining, for each review information corresponding to the candidate recommendation information, a similarity between the feature vector corresponding to the search information and a feature vector corresponding to the review information, as a similarity corresponding to the review information;

[0032] determining an information selection quantity corresponding to a current business period as a target quantity, and selecting, from each review information of a business object corresponding to the candidate recommendation information, the target quantity of review information as target review information corresponding to the candidate recommendation information, according to the similarity corresponding to each review information corresponding to the candidate recommendation information.

[0033] Optionally, the information selection quantity corresponding to the current business period is determined, and specifically includes:

[0034] For each historical business cycle, determine historical recommended information recommended to each user in the historical business cycle based on the information selection quantity corresponding to the historical business cycle, and determine historical execution results of each user in the historical business cycle for the historical recommended information as the historical execution results corresponding to the historical business cycle;

[0035] According to the historical execution results corresponding to each historical business cycle and the information selection quantity corresponding to each historical business cycle, determine the information selection quantity corresponding to the current business cycle.

[0036] The present specification provides an information recommendation device, comprising:

[0037] A receiving module is configured to receive a search request sent by a user.

[0038] A first determining module is configured to determine each candidate recommended information corresponding to search information according to the search information carried in the search request, and determine a feature vector corresponding to the search information by using a pre-trained first feature model.

[0039] A second determining module is configured to, for each candidate recommended information, determine at least one comment information associated with the search information from each comment information corresponding to the candidate recommended information as target comment information corresponding to the candidate recommended information according to the feature vector corresponding to the search information and the feature vector of each comment information corresponding to the candidate recommended information, and determine the feature vector corresponding to the candidate recommended information according to the target comment information corresponding to the candidate recommended information.

[0040] A recommendation module is configured to determine a to-be-recommended information from each candidate recommended information according to the feature vector corresponding to each candidate recommended information, and recommend the to-be-recommended information to the user.

[0041] The present specification provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the above-mentioned information recommendation method.

[0042] The present specification provides an electronic device, which comprises a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the above-mentioned information recommendation method when executing the program.

[0043] The above-mentioned at least one technical solution adopted by the present specification can achieve the following beneficial effects:

[0044] In the information recommendation method provided in this specification, after receiving a search request from a user, the business platform can determine the candidate recommendation information corresponding to the search information based on the search information carried in the search request, and determine the feature vector corresponding to the search information through a pre-trained first feature model. Then, for each candidate recommendation information, based on the feature vector corresponding to the search information and the determined feature vectors of each comment information corresponding to the candidate recommendation information, at least one comment information associated with the search information can be determined from the comments information corresponding to the candidate recommendation information as the target comment information corresponding to the candidate recommendation information. Based on the target comment information, the feature vector corresponding to the candidate recommendation information is determined. Based on the feature vectors corresponding to each candidate recommendation information, the information to be recommended can be determined from the candidate recommendation information and recommended to the user.

[0045] As can be seen from the above method, for a candidate recommendation, the business platform can select some comments related to the search information from the comments corresponding to that candidate recommendation. Based on these comments, the feature vector of the candidate recommendation can be determined, thus enabling the selection of the recommendation to be presented to the user from among the candidate recommendations. Compared to existing technologies, this method does not simply determine the recommendation to be presented to the user based on all the comments of each candidate recommendation, therefore, it can improve the accuracy of the determined recommendation. Attached Figure Description

[0046] The accompanying drawings, which are included to provide a further understanding of this specification and form part of this specification, illustrate exemplary embodiments and are used to explain this specification, but do not constitute an undue limitation thereof. In the drawings:

[0047] Figure 1 This is a flowchart illustrating one method of information recommendation in this specification;

[0048] Figure 2 This is a schematic diagram of an interface for recommending information to users, as provided in this specification.

[0049] Figure 3 This is a schematic diagram of an information recommendation device described in this specification;

[0050] Figure 4 The corresponding information provided in this specification Figure 1 A schematic diagram of an electronic device. Detailed Implementation

[0051] In order to make the purpose, technical solutions and advantages of the present application clearer, the technical solutions of the present application will be described in detail below with reference to the embodiments of the present application and the corresponding drawings. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of protection of the present application.

[0052] In the prior art, the business platform can determine the feature vector of each comment information of each candidate recommendation information, and then take the average feature vector of each comment information of the candidate recommendation information as the feature vector of the candidate recommendation information. The similarity between the feature vector of the candidate recommendation information and the feature vector of the search information input by the user can determine the to-be-recommended information to be recommended to the user.

[0053] Still taking the candidate recommendation information as the merchant information and the search information as "a park where cherry blossoms can be enjoyed" as an example, the user can be recommended in this way. There may be a situation that most of the comment information of a certain merchant is irrelevant to cherry blossom, but there are sporadic comments mentioning cherry blossom, and finally it is determined that the similarity between the merchant and the search information is low, and the business platform does not recommend the merchant information of the merchant to the user. However, in actual application, the business platform can recommend the merchant information of the merchant to the user. Therefore, it can be seen that the accuracy of the to-be-recommended information determined by the prior art is low.

[0054] The technical solutions provided by the embodiments of the present application will be described in detail below with reference to the drawings.

[0055] Figure 1 The flowchart of the method for recommending information in the present application specifically includes the following steps:

[0056] S101: receiving a search request sent by a user.

[0057] S102: determining each candidate recommendation information corresponding to the search information according to the search information carried in the search request, and determining the feature vector corresponding to the search information through a pre-trained first feature model.

[0058] In actual application, the user can search information through the business platform, the business platform can determine the search information input by the user through the terminal, so as to determine the information the user wants to view and recommend to the user. For example, if the business platform is a third-party consumer review website, the business platform can determine the merchant related to the search information input by the user and recommend the merchant information of the merchant to the user. For another example, if the business platform is a news website, the business platform can determine the news related to the search information input by the user and recommend the news to the user. For another example, the business platform can be a shopping website, and the business platform can recommend the product information to the user. Of course, there can be many related business scenarios, which will not be illustrated one by one here. The terminal mentioned above can be a mobile phone, a notebook computer, a tablet computer or the like.

[0059] Based on this, the business platform can receive the search request sent by the user, and determine each candidate recommendation information corresponding to the search information according to the search information carried in the search request. That is to say, the business platform can determine the search information input by the user through the terminal after the user searches information, and determine the candidate recommendation information having certain correlation with the search information through the search information, so as to subsequently determine the to-be-recommended information to be recommended to the user from the candidate recommendation information.

[0060] For example, if the business platform is a third-party consumer review website, and the search information input by the user is “a park where cherry blossoms can be viewed”, the business platform can preliminarily determine the park information of all parks in the city where the user is located as each candidate recommendation information through the search information carried in the search request. For another example, if the business platform is a shopping website, and the search information input by the user is “jeans that do not fade”, the business platform can determine the product information of 1000 pieces of jeans as the candidate recommendation information.

[0061] The business platform also needs to determine the feature vector of the search information through the first feature model trained in advance, so as to filter the to-be-recommended information to be recommended to the user from the candidate recommendation information. The first feature model can be a conventional feature extraction model in the field of natural language processing, such as a BERT model, a Doc2vec model or the like. The business platform can pre-train the first feature model in a supervised manner, so that the first feature model has certain semantic analysis capability, and then deploy the trained first feature model to the server through TensorFlow Serving, so as to determine the feature vector of the search information carried in the search request through the deployed first feature model after receiving the search request sent by the user.

[0062] Since the number of neural network layers included in some feature extraction models is large, it takes a long time to determine a feature vector of search information online through such a feature extraction model, therefore, the business platform can perform model distillation in the training process, that is, the business platform can first train a second feature model which has a larger size than the first feature model, then perform model distillation on the second feature model to obtain the first feature model which has a smaller size, and perform feature extraction on the search information.

[0063] Based on this, the business platform can obtain a training sample, and input text information included in the training sample into a preset second feature model, the second feature model can extract a feature vector corresponding to the text information, the business platform can determine a semantic analysis result for the text information according to the feature vector, and train the second feature model with the optimization goal of minimizing the difference between the labeled information included in the training sample and the semantic analysis result. The semantic analysis result mentioned here can be various, for example, the semantic analysis result can be to determine the next sentence of the text information. For another example, the semantic analysis can be to determine the word removed from the text information by the second feature model. Through a large number of training samples to perform such semantic analysis training on the second feature model, the second feature model can have good feature extraction capability.

[0064] Then, the business platform can construct a first feature model according to the network structure of the trained second feature model, and obtain historical search information, and train the first feature model through the historical search information, wherein the number of neural network layers included in the first feature model is less than the number of neural network layers included in the trained second feature model, and the feature vectors determined by the trained first feature model and the trained second feature model for the same search information are similar. That is, after the business platform trains a second feature model with good performance, a first feature model with similar network structure but fewer neural network layers than the second feature model can be constructed, and after training the first feature model, the first feature model can have a semantic analysis capability similar to that of the second feature model.

[0065] Specifically, the business platform can train the first feature model based on the semantic analysis result of the historical search information by the trained second feature model, that is, the semantic analysis result of the historical search information by the second feature model is taken as the labeled information of the historical search information, the historical search information and the semantic analysis result of the historical search information are taken as the training sample for training the first feature model, the first feature model is trained, that is, the prediction result for the historical search information is determined by the first feature model, and the first feature model is trained with the optimization goal of minimizing the difference between the prediction result and the semantic analysis result of the historical search information.

[0066] In addition, the business platform can train the first feature model through other model training manners. In the subsequent process, the business platform can determine the target comment information corresponding to the candidate recommendation information by searching for the similarity between the information and each comment information corresponding to the candidate recommendation information. Therefore, the business platform can jointly train the first feature vector for determining the search information and the third feature model for determining the feature vector of each comment information.

[0067] Specifically, the business platform can obtain historical search information, historical comment information, and a labeled similarity between the historical search information and the historical comment information. Then, the business platform can input the historical search information into the first feature model to obtain a feature vector corresponding to the historical search information, and input the historical comment information into the third feature model to obtain a feature vector corresponding to the historical comment information. The business platform can minimize the difference between the similarity between the feature vector corresponding to the historical search information and the feature vector corresponding to the historical comment information and the labeled similarity, and use the difference as an optimization target to train the first feature model and the third feature model.

[0068] That is to say, the purpose of this training manner is to enable the first feature model and the third feature model to determine a more accurate similarity between each comment information and the search information after training. The labeled similarity between the historical search information and the historical comment information mentioned above can be determined in various ways. For example, the labeled similarity can be labeled by manual labeling. For another example, the historical search information can be segmented to obtain the words contained in the historical search information. The business platform can determine the number or proportion of the words in the historical search information contained in the historical comment information, thereby determining the labeled similarity between the historical search information and the historical comment information. For another example, when the historical recommendation information is recommended to the user in the past, part of the historical comment information can be displayed to the user together. If the user clicks on the link (such as a merchant link) of the business object contained in a certain historical recommendation information, it can be determined that the historical comment information is related to the historical search information input by the user, and the labeled similarity between the historical search information and the historical comment information is set to 1.

[0069] S103: For each candidate recommendation information, at least one comment information associated with the search information is determined from the comment information corresponding to the candidate recommendation information as the target comment information corresponding to the candidate recommendation information according to the feature vector corresponding to the search information and the feature vector of the comment information corresponding to the candidate recommendation information determined, and the feature vector corresponding to the candidate recommendation information is determined according to the target comment information corresponding to the candidate recommendation information.

[0070] After determining the feature vector corresponding to the search information, the business platform can determine at least one comment information associated with the search information from each comment information corresponding to the candidate recommendation information according to the feature vector corresponding to the search information and the feature vectors of the comment information determined to correspond to the candidate recommendation information, as the target comment information corresponding to the candidate recommendation information, and determine the feature vector corresponding to the candidate recommendation information according to the target comment information corresponding to the candidate recommendation information.

[0071] That is, for each candidate recommendation information, the business platform can determine the comment information in the comment information of the candidate recommendation information that is more relevant to the search information as the target comment information corresponding to the candidate recommendation information. Among them, the business platform can determine the similarity between the feature vector corresponding to the search information and the feature vector corresponding to each comment information corresponding to the candidate recommendation information as the similarity corresponding to the comment information, and determine the information selection quantity corresponding to the current business period as the target quantity. The business platform can select the target quantity of comment information from the comment information of the business object corresponding to the candidate recommendation information as the target comment information corresponding to the candidate recommendation information according to the similarity of the comment information corresponding to the candidate recommendation information. The business period mentioned here can be set in advance.

[0072] As can be seen, this way is to determine the similarity between each comment information corresponding to the candidate recommendation information and the search information, and then sort each comment information according to the similarity from large to small, and select the target quantity of comment information ranked in the front as the target comment information corresponding to the candidate recommendation information. Of course, in addition to this way, the business platform can also determine the target comment information corresponding to the candidate recommendation information through other ways, for example, the business platform can determine the comment information corresponding to the candidate recommendation information and the search information similarity not less than the set similarity as the target comment information corresponding to the candidate recommendation information.

[0073] In this specification, the feature vectors of the comment information corresponding to the candidate recommendation information can be determined in advance through the third feature model. After determining the feature vector of each comment information, the feature vectors can be stored in the Faiss index. When it is necessary to determine the target comment information of the candidate recommendation information, the business platform can determine the feature vector with high similarity to the feature vector of the search information from the Faiss index through the Approximate Nearest Neighbor (ANN) method, thereby determining each target comment information. Of course, the business platform can also determine the similarity between the feature vectors of each comment information and the feature vector of the search information through other ways, thereby determining each target comment information.

[0074] It should be noted that the target quantity determines how many target comment information is selected from each comment information of the candidate recommended information. The greater the target quantity, the more target comment information is selected, and the more the feature vector of the candidate recommended information determined by the target comment information reflects the relevance between the overall comment information of the candidate recommended information and the search information. Therefore, the feature vector of the candidate recommended information may not reflect the features related to the search information in the comment information of the candidate recommended information. However, the smaller the target quantity, the less target comment information is selected, which may reflect the relevance between the candidate recommended information and the search information in a more one-sided manner. Therefore, the business platform needs to determine a more appropriate target quantity.

[0075] The business platform can experiment with a plurality of different information selection quantities in history to determine the target quantity. Specifically, the business platform can determine, for each historical business period, historical recommended information recommended to each user in the historical business period based on the information selection quantity corresponding to the historical business period, and determine historical execution results of each user performing a business on the historical recommended information in the historical business period as the historical execution results corresponding to the historical business period. The business platform can determine the information selection quantity corresponding to the current business period as the target quantity according to the historical execution results corresponding to each historical business period and the information selection quantity corresponding to each historical business period.

[0076] That is, different information selection quantities can be set for different historical business periods, and the historical recommended information recommended to the user is determined by the information selection quantity corresponding to the historical business period in the historical business period. After each user sees the historical recommended information recommended by the business platform in each historical business period, the execution result of the historical recommended information can reflect the advantages and disadvantages of the information selection quantity to some extent.

[0077] For example, each business period is 7 days, the information selection quantity set for historical business period a is 2, the information selection quantity set for historical business period b is 4, and the information selection quantity set for historical business period c is 6. In historical business period a, the business platform can select 2 target comment information corresponding to each candidate recommended information for the candidate recommended information, and determine the feature vector corresponding to the candidate recommended information according to the 2 target comment information. The business platform can determine the feature vector corresponding to each candidate recommended information in this way, and so on. In historical business period b, the business platform can select 4 target comment information corresponding to each candidate recommended information, and in historical business period c, the business platform can select 6 target comment information corresponding to each candidate recommended information.

[0078] In each historical business cycle, after the business platform recommends the historical recommendation information to the user, it can be determined whether the user performs the business in response to the historical recommendation information. For example, if the historical recommendation information is merchant information, after the business platform recommends the merchant information to the user, it can be determined whether the user clicks the merchant link contained in the merchant information to enter the page of the merchant. If the user clicks the merchant link, it can be indicated to some extent that the merchant information recommended to the user meets the search information input by the user. Therefore, the business platform can determine the situation of the user clicking the merchant link. Different information selection quantities are applied in different historical business cycles. The business platform can take the information selection quantity corresponding to the historical business cycle in which the user has a larger probability of clicking the merchant link (or the user clicks the merchant link more times) as the target quantity.

[0079] S104: Determine the to-be-recommended information from the candidate recommendation information according to the feature vector corresponding to each candidate recommendation information, and recommend the to-be-recommended information to the user.

[0080] After the business platform determines the feature vector corresponding to each candidate recommendation information, it can determine the to-be-recommended information from the candidate recommendation information according to the feature vector, and recommend the to-be-recommended information to the user. Specifically, the business platform can input the feature vector corresponding to each candidate recommendation information into a preset ranking model, so that the ranking model ranks each candidate recommendation information. The business platform can determine the to-be-recommended information according to the ranking result of the ranking model on each candidate recommendation information, and recommend the to-be-recommended information to the user.

[0081] The information contained in the feature vector of the candidate recommendation information determined by the business platform can be determined according to actual needs. For example, the business platform can determine the similarity between each target comment information of the candidate recommendation information and the search information as the similarity corresponding to the target comment information. The feature vector corresponding to the candidate recommendation information can contain the similarity corresponding to each target comment information and the feature vector corresponding to each target comment information.

[0082] For another example, the business platform can determine the similarity corresponding to each target review information of the candidate recommendation information, and determine the relevant parameter corresponding to the candidate recommendation information according to the similarity corresponding to each target review information of the candidate recommendation information. The business platform can determine the feature vector corresponding to the candidate recommendation information according to the similarity corresponding to each target review information of the candidate recommendation information, the feature vector corresponding to each target review information of the candidate recommendation information, and the relevant parameter. The relevant parameter mentioned herein is used to represent the similarity between the target review information of the candidate recommendation information, and / or the similarity between the target review information of the candidate recommendation information and the search information as a whole. In actual application, the relevant parameter can refer to the average similarity of the target review information of the candidate recommendation information, the variance between the similarity corresponding to the target review information of the candidate recommendation information and the average similarity, and / or the extreme value in the similarity corresponding to each target review information of the candidate recommendation information.

[0083] It should be noted that in addition to recommending the to-be-recommended information to the user, the business platform can also determine additional information for the to-be-recommended information, and recommend the to-be-recommended information and the additional information to the user. The additional information mentioned herein can include at least one of the information theme and the information abstract of the to-be-recommended information. The information theme mentioned herein can refer to the theme of the business object (such as a merchant, a commodity, etc.) corresponding to the to-be-recommended information summarized through the target review information. The information abstract can refer to an abstract that briefly describes the business object corresponding to the to-be-recommended information, which is spliced from several sentences of the target review information, such as Figure 2 .

[0084] Figure 2 An interface diagram for recommending information to a user is provided in the present specification.

[0085] As can be seen from Figure 2 , the search information input by the user is “A city can enjoy cherry in the park”, and the business platform determines the merchants (i.e., business objects) corresponding to each to-be-recommended information as a park and b park. For the a park, the business platform combines the information theme as “Cherry blossom first choice a park cherry blossom forest” and the information abstract as a paragraph combined by the target review sentence corresponding to the a park, such as Figure 2 , “There are many parks in A city where you can enjoy cherry blossoms, such as a park, c park, etc.; a park is the most famous cherry blossom park in A city; go to the cherry blossom square to enjoy cherry blossoms”, and the information theme and the information abstract of the b park are also similar. The business platform can determine the information theme and the information abstract of the b park through the target review information under the b park.

[0086] The method of information recommendation provided by the specification is exemplified in the above-mentioned scenarios of recommending merchants, and can also be applied to other business scenarios in actual application, such as scenarios of recommending goods to users and scenarios of recommending news to users, and the specific implementation manner is similar to that of the scenario of recommending merchants, and thus will not be described herein.

[0087] As can be seen from the above method, for each candidate recommendation information, the business platform can determine the part of target comment information in the comment information corresponding to the candidate recommendation information that is more strongly associated with the search information, and determine the feature vector corresponding to each candidate recommendation information through the target comment information, so as to determine the to-be-recommended information to be recommended to the user. Compared with the prior art of directly determining the feature vector of the candidate recommendation information through the comment information corresponding to the candidate recommendation information, the accuracy of information recommendation to the user is improved to a certain extent, and the business execution efficiency of the user is improved.

[0088] The above is the method of information recommendation provided by one or more embodiments of the specification, and based on the same idea, the specification also provides a corresponding information recommendation device, as shown in Figure 3 .

[0089] Figure 3 The information recommendation device provided by the specification is a schematic diagram, and specifically includes:

[0090] The receiving module 301 is configured to receive a search request sent by a user.

[0091] The first determining module 302 is configured to determine, according to search information carried in the search request, each candidate recommendation information corresponding to the search information, and determine a feature vector corresponding to the search information through a pre-trained first feature model.

[0092] The second determining module 303 is configured to, for each candidate recommendation information, determine, according to the feature vector corresponding to the search information and the feature vector of each comment information corresponding to the candidate recommendation information, at least one comment information associated with the search information from each comment information corresponding to the candidate recommendation information as target comment information corresponding to the candidate recommendation information, and determine a feature vector corresponding to the candidate recommendation information according to the target comment information corresponding to the candidate recommendation information.

[0093] The recommendation module 304 is configured to determine, according to the feature vector corresponding to each candidate recommendation information, to-be-recommended information from the candidate recommendation information, and recommend the to-be-recommended information to the user.

[0094] Optionally, the device further includes:

[0095] The training module 305 is configured to: obtain training samples; input text information contained in the training samples into a preset second feature model to obtain semantic analysis results for the text information; train the second feature model by taking minimization of a difference between labeled information contained in the training samples and the semantic analysis results as an optimization target; construct a first feature model according to a network structure of the trained second feature model, the first feature model constructed containing fewer neural network layers than the trained second feature model; and obtain historical search information, and train the first feature model by using the historical search information, wherein the first feature model trained and the second feature model trained are similar in feature vectors determined for the same search information.

[0096] Optionally, the training module 305 is specifically configured to: obtain historical search information, historical review information, and a labeled similarity between the historical search information and the historical review information; input the historical search information into the first feature model to obtain a feature vector corresponding to the historical search information, and input the historical review information into a preset third feature model to obtain a feature vector corresponding to the historical review information; and train the first feature model and the third feature model by taking minimization of a difference between the labeled similarity and a similarity between the feature vector corresponding to the historical search information and the feature vector corresponding to the historical review information as an optimization target.

[0097] Optionally, the second determination module 303 is specifically configured to: for each target review information of the candidate recommendation information, determine a similarity between the feature vector corresponding to the search information and a feature vector corresponding to the target review information as a similarity corresponding to the target review information; determine a correlation parameter corresponding to the candidate recommendation information according to the similarity corresponding to each target review information of the candidate recommendation information, the correlation parameter being used to represent a similarity between the target review information of the candidate recommendation information and / or a similarity between the target review information of the candidate recommendation information and the search information as a whole; and determine a feature vector corresponding to the candidate recommendation information according to the similarity corresponding to each target review information of the candidate recommendation information, the feature vector corresponding to each target review information of the candidate recommendation information, and the correlation parameter.

[0098] Optionally, the recommendation module 304 is specifically configured to: determine additional information for the to-be-recommended information according to the target review information corresponding to the to-be-recommended information, the additional information including at least one of an information theme and an information abstract corresponding to the to-be-recommended information; and recommend the to-be-recommended information and the additional information to the user.

[0099] Optionally, the second determining module 303 is specifically used to: for each comment information corresponding to the candidate recommendation information, determine the similarity between the feature vector corresponding to the search information and the feature vector corresponding to the comment information, as the similarity corresponding to the comment information; determine the number of information selections corresponding to the current business cycle, as the target number; and select the target number of comment information from each comment information of the business object corresponding to the candidate recommendation information based on the similarity corresponding to each comment information of the candidate recommendation information, as the target comment information corresponding to the candidate recommendation information.

[0100] Optionally, the device further includes:

[0101] The quantity selection module 306 is used to determine, for each historical business cycle, the historical recommendation information recommended to each user based on the information selection quantity corresponding to the historical business cycle, and to determine the historical execution results of each user's business execution based on the historical recommendation information within the historical business cycle, as the historical execution results corresponding to the historical business cycle; and to determine the information selection quantity corresponding to the current business cycle based on the historical execution results and the information selection quantity corresponding to each historical business cycle.

[0102] This specification also provides a computer-readable storage medium storing a computer program that can be used to execute the above-described... Figure 1 The information provided is recommended using this method.

[0103] This instruction manual also provides Figure 4 The diagram shows a schematic structural representation of the electronic device. Figure 4 At the hardware level, the electronic device includes a processor, internal bus, network interface, memory, and non-volatile memory, and may also include other hardware required for the business operations. The processor reads the corresponding computer program from the non-volatile memory into memory and then runs it to achieve the above-mentioned functions. Figure 1 The method of information recommendation described herein. Of course, in addition to software implementation, this specification does not exclude other implementation methods, such as logic devices or a combination of hardware and software, etc. In other words, the execution subject of the following processing flow is not limited to individual logic units, but can also be hardware or logic devices.

[0104] In the 1990s, it was quite obvious to distinguish whether an improvement in a technology was in hardware (e.g., improvement in circuit structures of diodes, transistors, switches, etc.) or in software (improvement in method flow). However, as technology has evolved, many improvements in method flow today can be considered as direct improvements in hardware circuit structures. Designers almost always obtain the corresponding hardware circuit structures by programming the improved method flow into hardware circuits. Therefore, it cannot be said that an improvement in a method flow cannot be implemented by hardware entity modules. For example, a programmable logic device (PLD) (e.g., a field programmable gate array (FPGA)) is an integrated circuit whose logic function is determined by user programming of the device. A digital system is "integrated" on a PLD by the designer programming it, rather than by asking a chip manufacturer to design and fabricate a custom integrated circuit chip. Moreover, instead of manually fabricating integrated circuit chips, this programming is now mostly implemented by "logic compiler" software, which is similar to software compilers used in program development, and the original code to be compiled is written in a specific programming language, which is called a hardware description language (HDL), and there are many such languages, such as ABEL (Advanced Boolean Expression Language), AHDL (Altera Hardware Description Language), Confluence, CUPL (Cornell University Programming Language), HDCal, JHDL (Java Hardware Description Language), Lava, Lola, MyHDL, PALASM, RHDL (Ruby Hardware Description Language), etc., and the most commonly used are VHDL (Very-High-Speed Integrated Circuit Hardware Description Language) and Verilog. Those skilled in the art should be aware that, as long as the method flow is logically programmed in the above-mentioned hardware description languages and programmed into an integrated circuit, a hardware circuit implementing the logical method flow can be easily obtained.

[0105] The controller can be implemented in any suitable way, for example, the controller can take the form of a microprocessor or processor and a computer readable medium storing computer readable program code, such as software or firmware, executable by the (micro)processor, logic gates, switches, an application specific integrated circuit (ASIC), a programmable logic controller and an embedded microcontroller, examples of which include but are not limited to the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20 and Silicone Labs C8051F320, the memory controller can also be implemented as part of the control logic of the memory. Those skilled in the art will also know that, in addition to being implemented in pure computer readable program code, the controller can equally well be implemented to perform the same functions using logic gates, switches, an application specific integrated circuit, a programmable logic controller and an embedded microcontroller, etc. by means of a logical programming of the method steps. The controller can thus be considered as a hardware component, and the means comprised therein for performing the various functions can be considered as structures within the hardware component. Alternatively, the means for performing the various functions can even be considered as both a software module implementing the method and a structure within the hardware component.

[0106] The systems, apparatuses, modules or units illustrated by the above embodiments can be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, the computer can be a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

[0107] For the sake of description, the above apparatuses are described in various units with functions respectively. Of course, the functions of the units can be implemented in one or more software and / or hardware in implementing the present specification.

[0108] Those skilled in the art will understand that the embodiments of the present application can be provided as a method, a system or a computer program product. Therefore, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage etc.) containing computer usable program code.

[0109] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 one or more flowcharts and / or blocks in the flowcharts and / or combination thereof. Figure 1 one or more flowcharts and / or blocks in the flowcharts and / or combination thereof.

[0110] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 one or more flowcharts and / or blocks in the flowcharts and / or combination thereof. Figure 1 one or more flowcharts and / or blocks in the flowcharts and / or combination thereof.

[0111] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 one or more flowcharts and / or blocks in the flowcharts and / or combination thereof. ​ one or more flowcharts and / or blocks in the flowcharts and / or combination thereof.

[0112] In one typical configuration, the computing device includes one or more processors (CPUs), input / output interfaces, network interfaces, and memory.

[0113] The memory can include non-persistent memory and / or volatile memory, such as random access memory (RAM) and / or cache memory, non-volatile memory, such as read-only memory (ROM), EPROM, and / or flash memory. The memory is an example of computer-readable media.

[0114] Computer-readable media includes permanent and non-permanent, movable and non-movable media that can be implemented by any method or technology to store information. The information can be computer-readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassette, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information accessible to a computing device. According to the definition herein, computer-readable media does not include transitory media such as modulated data signals and carriers.

[0115] It should also be noted that the terms "comprising", "containing", or any other variant thereof are intended to cover non-exclusive inclusion, such that a process, method, article or apparatus that comprises a list of elements does not only include those elements, but can also include other elements not expressly listed or inherent to such process, method, article or apparatus. Without more limitations, the element defined by the statement "comprising a" does not exclude the presence of additional identical elements in the process, method, article or apparatus that includes the element.

[0116] Those skilled in the art will appreciate that embodiments of the present specification can be provided as methods, systems or computer program products. Therefore, the present specification can take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Moreover, the present specification can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0117] The present specification can be described in the general context of computer-executable instructions, such as program modules, executed by computers. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform particular tasks or implement particular abstract data types. The present specification can also be practiced in distributed computing environments where tasks are performed by remote processing devices 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.

[0118] The various embodiments described in this specification are described using a numbering of embodiments approach: these are each individually integrated contributions pertaining to different aspects of the description. For each embodiment, the description focuses on the differences from the other embodiments. In particular, the description of the system embodiments is relatively brief, as the system embodiments are largely analogous to the method embodiments. The relevant parts of the description of the method embodiments are therefore referred to.

[0119] The above description is embodied in the form of embodiments only and is not intended to limit the present specification. The present specification can be variously changed and modified by those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present specification should be included in the scope of the claims of the present specification.

Claims

1. A method for information recommendation, characterized in that, include: Receive search requests sent by users; Based on the search information carried in the search request, determine each candidate recommendation information corresponding to the search information, and determine the feature vector corresponding to the search information through a pre-trained first feature model; For each candidate recommendation, based on the feature vector corresponding to the search information and the feature vectors of each comment information corresponding to the candidate recommendation, at least one comment information associated with the search information is determined from each comment information corresponding to the candidate recommendation, which is used as the target comment information corresponding to the candidate recommendation, and the feature vector corresponding to the candidate recommendation is determined based on the target comment information. Based on the feature vectors corresponding to each candidate recommendation information, the information to be recommended is determined from the candidate recommendation information, and the information to be recommended is recommended to the user. Based on the target comment information corresponding to the candidate recommendation information, the feature vector corresponding to the candidate recommendation information is determined, specifically including: For each target comment information in the candidate recommendation information, the similarity between the feature vector corresponding to the search information and the feature vector corresponding to the target comment information is determined, and this similarity is used as the similarity between the target comment information and the target comment information. Based on the similarity of each target comment in the candidate recommendation information, relevant parameters are determined for the candidate recommendation information. These relevant parameters are used to characterize the similarity between each target comment in the candidate recommendation information, and / or the overall similarity between each target comment in the candidate recommendation information and the search information. The feature vector corresponding to the candidate recommendation information is determined based on the similarity of each target comment information in the candidate recommendation information, the feature vector corresponding to each target comment information in the candidate recommendation information, and the relevant parameters.

2. The method as described in claim 1, characterized in that, Training the first feature model specifically includes: Obtain training samples; The text information contained in the training samples is input into a preset second feature model to obtain the semantic analysis results of the text information; The second feature model is trained with the optimization objective of minimizing the difference between the annotation information contained in the training samples and the semantic analysis results; The first feature model is constructed based on the network structure of the trained second feature model. The number of neural network layers in the constructed first feature model is less than the number of neural network layers in the trained second feature model. Historical search information is obtained, and the first feature model is trained using the historical search information, wherein the feature vectors determined by the trained first feature model and the trained second feature model for the same search information are similar.

3. The method as described in claim 1 or 2, characterized in that, Training the first feature model specifically includes: acquiring historical search information, historical comment information, and labeled similarity between the historical search information and the historical comment information; The historical search information is input into the first feature model to obtain the feature vector corresponding to the historical search information, and the historical comment information is input into the preset third feature model to obtain the feature vector corresponding to the historical comment information. The first feature model and the third feature model are trained with the optimization objective of minimizing the difference between the similarity between the feature vector corresponding to the historical search information and the feature vector corresponding to the historical comment information and the difference between the labeled similarity.

4. The method as described in claim 1, characterized in that, Recommending the information to be recommended to the user specifically includes: determining additional information for the information to be recommended based on the target comment information corresponding to the information to be recommended, wherein the additional information includes at least one of the information topic and information summary corresponding to the information to be recommended; The information to be recommended, along with the additional information, is recommended to the user.

5. The method as described in claim 1, characterized in that, Based on the feature vector corresponding to the search information and the feature vectors of each comment information corresponding to the candidate recommendation information, at least one comment information associated with the search information is determined from the comment information corresponding to the candidate recommendation information, and is used as the target comment information corresponding to the candidate recommendation information. Specifically, this includes: For each comment corresponding to the candidate recommendation information, the similarity between the feature vector corresponding to the search information and the feature vector corresponding to the comment information is determined, and this similarity is used as the similarity between the two comments. Determine the number of information selections corresponding to the current business cycle as the target number, and select the target number of comment information from the comment information of the business object corresponding to the candidate recommendation information based on the similarity of each comment information.

6. The method as described in claim 5, characterized in that, Determine the number of information items to be selected for the current business cycle, specifically including: For each historical business cycle, determine the number of historical recommendation information to recommend to each user based on the information corresponding to the historical business cycle, and determine the historical execution results of each user performing business based on the historical recommendation information within the historical business cycle, as the historical execution results corresponding to the historical business cycle; The number of information selections for the current business cycle is determined based on the historical execution results and the number of information selections for each historical business cycle.

7. An information recommendation device, characterized in that, include: The receiving module is used to receive search requests sent by users; The first determining module is used to determine each candidate recommendation information corresponding to the search information based on the search information carried in the search request, and to determine the feature vector corresponding to the search information through a pre-trained first feature model. The second determining module is used, for each candidate recommendation information, to determine, based on the feature vector corresponding to the search information and the determined feature vectors of each comment information corresponding to the candidate recommendation information, at least one comment information associated with the search information from the comment information corresponding to the candidate recommendation information, as the target comment information corresponding to the candidate recommendation information, and to determine the feature vector corresponding to the candidate recommendation information based on the target comment information. Specifically, this includes: For each target comment information in the candidate recommendation information, the similarity between the feature vector corresponding to the search information and the feature vector corresponding to the target comment information is determined, and this similarity is used as the similarity between the target comment information and the target comment information. Based on the similarity of each target comment in the candidate recommendation information, relevant parameters are determined for the candidate recommendation information. These relevant parameters are used to characterize the similarity between each target comment in the candidate recommendation information, and / or the overall similarity between each target comment in the candidate recommendation information and the search information. The feature vector corresponding to the candidate recommendation information is determined based on the similarity of each target comment information in the candidate recommendation information, the feature vector corresponding to each target comment information in the candidate recommendation information, and the relevant parameters. The recommendation module is used to determine the information to be recommended from the candidate recommendation information based on the feature vectors corresponding to each candidate recommendation information, and recommend the information to be recommended to the user.

8. A computer-readable storage medium, characterized in that, The storage medium stores a computer program, which, when executed by a processor, implements the method described in any one of claims 1 to 6.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the method described in any one of claims 1 to 6.

Images