Recommendation method, device, electronic device and storage medium for searching information

By obtaining the word segmentation characteristics of the search information input by the user, calculating the similarity to the feature vectors in the recall pool dictionary, and determining the alternative recommendation information, solving the problem that new search information cannot be recalled and improving the recommendation effect of search information.

CN113282831BActive Publication Date: 2024-12-06BEIJING IQIYI TECH CO LTD
View PDF 3 Cites 0 Cited by

Patent Information

Application Number
CN202110646922.X
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2021-06-10
Publication Date
2024-12-06
Estimated Expiration
2041-06-10

AI Technical Summary

Technical Problem

In the prior art, the new search information entered by the user cannot be recalled, which affects the recommendation effect of the search information, because the search information in the index dictionary only exists in the feature vector dictionary, and the new information cannot be found during the recall process.

Method used

By obtaining the word segmentation characteristics of the target search information entered by the user, the target feature vector is determined, and the similarity of the feature vector in the pre-obtained recall pool dictionary is matched, and the alternative recommendation information is determined from the recall pool and added to the candidate information set. , calculate the similarity to recommend search information.

Benefits of technology

Even for new search information, recommended search information can be determined from the candidate collection, which improves the recall and recommendation effect of search information, meets users' search needs and achieves traffic diversion purposes.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN113282831B_ABST
    Figure CN113282831B_ABST
Patent Text Reader

Abstract

The embodiment of the present invention provides a method, device, electronic device and storage medium for recommending search information, the method comprising: obtaining target search information input by a user, determining a target feature vector corresponding to the target search information based on the word segmentation features of the target search information; determining candidate recommended information based on the similarity between the target feature vector and the feature vector of the search information; adding the candidate recommended information to a candidate information set, and determining a feature vector corresponding to each candidate information; calculating the similarity between the feature vector corresponding to each candidate information and the target feature vector, and determining recommended search information from the candidate information set based on the similarity; and recommending the recommended search information to the user. Since it is not necessary to determine the search information based on the corresponding relationship between the search information and its recalled search information, even if it is new search information input by the user, the recall of the search information can be successfully completed, thereby improving the recommendation effect of the search information.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of information search technology, and in particular to a method, device, electronic device and storage medium for recommending information search. Background Art

[0002] For websites with search functions, accurate search information recommendations for different users can be made to attract users to click on the recommended search information for search, which can effectively achieve the purpose of attracting traffic. The search information recommendation method in the related art can be divided into two stages: recall and sorting. Among them, the sorting stage is to score and sort the recalled candidate recommendation information, and finally determine the search information to be recommended to the user based on the scoring and sorting results.

[0003] In the current recall process, a feature vector is first trained for each search information to form a feature vector dictionary including the correspondence between the search information and the feature vector. Then, an index dictionary including the correspondence between the search information and its recalled search information is constructed offline through the similarity between each feature vector. When recalling online, after obtaining the target search information in the user's search behavior, the recalled search information corresponding to the target search information is searched from the index dictionary, thereby realizing the recall of the search information.

[0004] In the above method, since the search information in the index dictionary can only be the search information existing in the feature vector dictionary, if the search information input by the user is new search information, which is not the search information already in the feature vector dictionary, then it must not be in the index dictionary, and the search information cannot be recalled, which affects the recommendation effect of the search information. Summary of the invention

[0005] The purpose of the embodiments of the present invention is to provide a method, device, electronic device and storage medium for recommending search information, so as to recall the search information when the search information input by the user is new search information, and improve the recommendation effect of the search information. The specific technical solution is as follows:

[0006] In a first aspect, an embodiment of the present invention provides a method for recommending search information, the method comprising:

[0007] Acquire target search information input by a user, and determine a target feature vector corresponding to the target search information based on the word segmentation features of the target search information;

[0008] Determining candidate recommendation information from the search information included in the recall pool dictionary according to the similarity between the target feature vector and the feature vector of the search information included in the recall pool dictionary obtained in advance;

[0009] Adding the candidate recommendation information to an existing candidate information set, and determining a feature vector corresponding to each candidate information in the candidate information set;

[0010] Calculating the similarity between the feature vector corresponding to each candidate information and the target feature vector, and determining the target candidate information from the candidate information set based on the similarity as the recommended search information;

[0011] The recommended search information is recommended to the user.

[0012] Optionally, the step of determining a feature vector corresponding to each candidate information in the candidate information set includes:

[0013] For each candidate information included in the candidate information set, if the candidate information exists in the recall pool dictionary, determine the feature vector corresponding to the candidate information in the recall pool dictionary as the feature vector corresponding to the candidate information;

[0014] In the case that the candidate information does not exist in the recall pool dictionary, the candidate information is input into a pre-trained feature vector model to obtain a feature vector corresponding to the candidate information output by the feature vector model.

[0015] Optionally, the step of determining a target feature vector corresponding to the target search information based on the word segmentation feature of the target search information includes:

[0016] Performing word segmentation processing on the target search information to obtain a word segmentation feature of the target search information as a first word segmentation feature;

[0017] The first word segmentation feature is input into a pre-trained feature vector model to obtain a feature vector output by the feature vector model based on the first word segmentation feature as a target feature vector corresponding to the target search information.

[0018] Optionally, the recall pool dictionary is established by:

[0019] Obtaining multiple pre-collected search information;

[0020] Perform word segmentation processing on each search information to obtain a word segmentation feature of each search information as a second word segmentation feature;

[0021] Inputting the second word segmentation feature into a pre-trained feature vector model, obtaining a feature vector output by the feature vector model based on the second word segmentation feature as a feature vector corresponding to the search information;

[0022] The search information and its corresponding feature vector are recorded accordingly to obtain a recall pool dictionary.

[0023] Optionally, the method further includes:

[0024] updating the feature vector model;

[0025] Based on the updated feature vector model, the recall pool dictionary is updated.

[0026] Optionally, the step of determining target candidate information from the candidate information set based on the similarity includes:

[0027] Input each candidate information in the candidate information set and its corresponding similarity into a pre-trained ranking model to obtain a score corresponding to each candidate information output by the ranking model based on the similarity;

[0028] According to the score corresponding to each candidate information, target candidate information is determined from the candidate information set.

[0029] In a second aspect, an embodiment of the present invention provides a device for recommending search information, the device comprising:

[0030] A first vector determination module, configured to obtain target search information input by a user, and determine a target feature vector corresponding to the target search information based on the word segmentation features of the target search information;

[0031] a candidate recommendation information determination module, configured to determine candidate recommendation information from the search information included in the recall pool dictionary according to the similarity between the target feature vector and the feature vector of the search information included in the recall pool dictionary obtained in advance by the recall pool establishment module;

[0032] A second vector determination module, used for adding the candidate recommendation information to the existing candidate information set, and determining the feature vector corresponding to each candidate information in the candidate information set;

[0033] A recommended search information determination module, used to calculate the similarity between the feature vector corresponding to each candidate information and the target feature vector, and determine the target candidate information from the candidate information set based on the similarity as the recommended search information;

[0034] The search information recommendation module is used to recommend the recommended search information to the user.

[0035] Optionally, the second vector determination module includes:

[0036] a first vector determination unit, configured to determine, for each candidate information included in the candidate information set, a feature vector corresponding to the candidate information in the recall pool dictionary as a feature vector corresponding to the candidate information if the candidate information exists in the recall pool dictionary;

[0037] The second vector determination unit is used to input the candidate information into a pre-trained feature vector model when the candidate information does not exist in the recall pool dictionary, and obtain a feature vector corresponding to the candidate information output by the feature vector model.

[0038] Optionally, the first vector determination module includes:

[0039] A first word segmentation processing unit, configured to perform word segmentation processing on the target search information to obtain a word segmentation feature of the target search information as a first word segmentation feature;

[0040] A vector determination unit is used to input the first word segmentation feature into a pre-trained feature vector model to obtain a feature vector output by the feature vector model based on the first word segmentation feature as a target feature vector corresponding to the target search information.

[0041] Optionally, the recall pool establishment module includes:

[0042] An information acquisition unit, used to acquire a plurality of pre-collected search information;

[0043] A second word segmentation processing unit, configured to perform word segmentation processing on each search information, and obtain a word segmentation feature of each search information as a second word segmentation feature;

[0044] A third vector determination unit is used to input the second word segmentation feature into a pre-trained feature vector model to obtain a feature vector output by the feature vector model based on the second word segmentation feature as a feature vector corresponding to the search information;

[0045] The recall pool establishing unit is used to record the search information and its corresponding feature vector to obtain a recall pool dictionary.

[0046] Optionally, the device further comprises:

[0047] A first model updating module, used for updating the feature vector model;

[0048] The second model updating module is used to update the recall pool dictionary based on the updated feature vector model.

[0049] Optionally, the recommended search information determination module includes:

[0050] A score determination unit, used for inputting each candidate information in the candidate information set and its corresponding similarity into a pre-trained sorting model, and obtaining a score corresponding to each candidate information output by the sorting model based on the similarity;

[0051] The candidate information determining unit is used to determine target candidate information from the candidate information set according to the score corresponding to each candidate information.

[0052] In a third aspect, an embodiment of the present invention provides an electronic device, including a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other via the communication bus;

[0053] Memory, used to store computer programs;

[0054] The processor is used to implement any method step described in the first aspect when executing the program stored in the memory.

[0055] In a fourth aspect, an embodiment of the present invention provides a computer-readable cache medium, in which a computer program is cached. When the computer program is executed by a processor, any method step described in the first aspect is implemented.

[0056] In the scheme provided by the embodiment of the present invention, the electronic device can obtain the target search information input by the user, and determine the target feature vector corresponding to the target search information based on the word segmentation features of the target search information, determine the alternative recommended information from the search information included in the recall pool dictionary according to the similarity between the target feature vector and the feature vector of the search information included in the recall pool dictionary in advance, add the alternative recommended information to the existing candidate information set, and determine the target feature vector corresponding to each candidate information in the candidate information set, calculate the similarity between the feature vector corresponding to each candidate information and the target feature vector, and determine the target candidate information from the candidate information set based on the similarity as the recommended search information, and then recommend the recommended search information to the user.

[0057] Since the electronic device adds the alternative recommended information determined based on the similarity between feature vectors to the candidate information set, the candidate set includes multiple candidate information with high similarity to the target search information. Furthermore, the electronic device determines the target candidate information based on the similarity between the feature vector corresponding to each candidate information and the target feature vector, and can further select the target candidate information with high similarity to the target feature vector from the candidate information set as the recommended search information. In this way, since there is no need to determine the recalled search information based on the correspondence between the search information and its recalled search information, even if the target search information is new search information input by the user, the electronic device can determine the recommended search information from the candidate set, successfully complete the recall of the search information, and improve the recommendation effect of the search information. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art are briefly introduced below.

[0059] Figure 1 A flowchart of a method for recommending information search provided by an embodiment of the present invention;

[0060] Figure 2 for Figure 1 A specific flow chart of step S101 in the illustrated embodiment;

[0061] Figure 3 Based on Figure 1 A flow chart of the method of establishing a recall pool dictionary in the illustrated embodiment;

[0062] Figure 4 Based on Figure 1 A flow chart of the model updating method of the illustrated embodiment;

[0063] Figure 5 for Figure 1 A specific flow chart of step S104 in the illustrated embodiment;

[0064] Figure 6 A schematic diagram of the structure of a device for recommending search information provided by an embodiment of the present invention;

[0065] Figure 7 for Figure 6 A schematic diagram of a specific structure of the first vector determination module in the illustrated embodiment;

[0066] Figure 8 for Figure 6 A structural schematic diagram of a recall pool establishment module in the illustrated embodiment;

[0067] Fig. 9 The present invention is a schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0068] The technical solutions in the embodiments of the present invention will be described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0069] In order to recall search information when the search information input by the user is new search information and improve the recommendation effect of search information, an embodiment of the present invention provides a search information recommendation method, device, electronic device, computer-readable storage medium and computer program product. A search information recommendation method provided by an embodiment of the present invention is introduced below.

[0070] The method for recommending search information provided by the embodiment of the present invention can be applied to any electronic device that needs to recommend search information, such as computers, tablet computers, mobile phones and other electronic devices, which are not specifically limited here.

[0071] like Figure 1 As shown, a method for recommending searching information includes:

[0072] S101, obtaining target search information input by a user, and determining a target feature vector corresponding to the target search information based on the word segmentation features of the target search information;

[0073] S102, determining candidate recommendation information from the search information included in the recall pool dictionary according to the similarity between the target feature vector and the feature vector of the search information included in the recall pool dictionary obtained in advance;

[0074] S103, adding the candidate recommendation information to the existing candidate information set, and determining the target feature vector corresponding to each candidate information in the candidate information set;

[0075] S104, calculating the similarity between the feature vector corresponding to each candidate information and the target feature vector, and determining the target candidate information from the candidate information set based on the similarity as the recommended search information;

[0076] S105: Recommend the recommended search information to the user.

[0077] It can be seen that in the scheme provided by the embodiment of the present invention, the electronic device can obtain the target search information input by the user, and determine the target feature vector corresponding to the target search information based on the word segmentation features of the target search information, determine the alternative recommended information from the search information included in the recall pool dictionary according to the similarity between the target feature vector and the feature vector of the search information included in the pre-acquired recall pool dictionary, add the alternative recommended information to the existing candidate information set, and determine the target feature vector corresponding to each candidate information in the candidate information set, calculate the similarity between the feature vector corresponding to each candidate information and the target feature vector, and determine the target candidate information from the candidate information set based on the similarity as the recommended search information, and then recommend the recommended search information to the user. Since the electronic device adds the alternative recommended information determined based on the similarity between feature vectors to the candidate information set, the candidate set includes multiple candidate information with high similarity to the target search information. Furthermore, the electronic device determines the target candidate information based on the similarity between the feature vector corresponding to each candidate information and the target feature vector, and can further select the target candidate information with high similarity to the target feature vector from the candidate information set as the recommended search information. In this way, since there is no need to determine the recalled search information based on the correspondence between the search information and its recalled search information, even if the target search information is new search information input by the user, the electronic device can determine the recommended search information from the candidate set, successfully complete the recall of the search information, and improve the recommendation effect of the search information.

[0078] When users use the search function provided by the website to search for information, the website server can recommend relevant search information to users on the search page, attracting users to search based on the recommended search information, which can meet the user's search needs and effectively achieve the purpose of attracting traffic.

[0079] When the electronic device recommends search information to the user, it is generally based on the search information input by the user, and recommends search information related to the search information input by the user. Then, in the above step S101, the electronic device can obtain the search information input by the user and use it as the target search information. The target search information is the search information that the user wants to search. The target search information can be any one of a word, a phrase or a sentence, which is not specifically limited here.

[0080] For example, when a user uses the search function provided by the website to search for information, he enters the search information "TV series A" that he wants to search for, and the electronic device can obtain the search information "TV series A". The obtained search information "TV series A" is the target search information.

[0081] After obtaining the target search information input by the user, in order to determine the target feature vector corresponding to the target search information, the electronic device can obtain the word segmentation feature of the target search information, and then determine the target feature vector corresponding to the target search information based on the word segmentation feature.

[0082] In one embodiment, the electronic device can perform word segmentation processing on the target search information, and then obtain the word segmentation features of the target search information, wherein the word segmentation processing can adopt any word segmentation processing method in the information processing field, as long as the word segmentation features of the target search information can be obtained, and no specific limitation or explanation is made here.

[0083] The target feature vector is a vector that can represent the features of the target search information input by the user. The electronic device can determine the target feature vector corresponding to the target search information based on the word segmentation features using a deep learning model or other methods. No specific limitation is made here.

[0084] In order to expand the sources of recalled search information, a recall pool dictionary may be pre-established, which may include the correspondence between pre-collected search information and its corresponding feature vectors. The pre-collected search information may be high-quality user historical search information obtained through screening.

[0085] Since the similarity between the two feature vectors is higher, it means that the correlation between the search information corresponding to the two feature vectors is greater. Therefore, in the above step S102, the electronic device can determine the alternative recommended information from the search information included in the recall pool dictionary based on the similarity between the target feature vector and the feature vector of the search information included in the recall pool dictionary obtained in advance.

[0086] In one embodiment, the electronic device can calculate the similarity between the target feature vector and the feature vector of each search information included in the recall pool dictionary, and then select the top N search information with higher similarity as alternative recommended information, where N is a positive integer, for example, it can be 3, 5, 10, etc., and is not specifically limited here.

[0087] In another embodiment, a similarity threshold may be preset, for example, 80%, 90%, 95%, etc. The electronic device may calculate the similarity between the target feature vector and the feature vector of each search information included in the recall pool dictionary, and then select the corresponding search information whose similarity reaches the similarity threshold as the candidate recommendation information.

[0088] In the above step S103, the electronic device may add the candidate recommendation information to the existing candidate information set, and determine the feature vector corresponding to each candidate information in the candidate information set. The existing candidate information set is a set including multiple candidate information, and the candidate information may include search information recorded in the user's historical search behavior process, etc., which can be used as candidate recommendation information. The number of candidate information can be set according to actual needs, for example, it can be hundreds, dozens, etc., and is not specifically limited here.

[0089] Since the feature vector corresponding to the above-mentioned alternative recommendation information has a high similarity with the target feature vector, the alternative recommendation information can also be used as candidate recommendation information. Then the electronic device can add the alternative recommendation information to the existing candidate information set, so that the candidate set includes the existing candidate information and the newly added alternative recommendation information.

[0090] Next, the electronic device may determine a feature vector corresponding to each candidate information in the candidate information set, so that the electronic device may subsequently select appropriate recommended search information from the candidate information.

[0091] After determining the feature vector corresponding to each candidate information, the electronic device can calculate the similarity between the feature vector corresponding to each candidate information and the target feature vector, that is, perform the above step S104. As an implementation method, the similarity between two feature vectors in the embodiment of the present invention can be represented by cosine distance, Euclidean distance, Manhattan distance, Chebyshev distance, etc., which is not specifically limited here.

[0092] As an implementation mode, for a target feature vector a and a feature vector b corresponding to any candidate information included in the candidate information set, the cosine distance between the target feature vector a and the feature vector b is:

[0093]

[0094] in,<a,b> is the inner product between the target feature vector a and the feature vector b, || is the length of the target feature vector a, || is the length of the feature vector b, and θ is the angle between the target feature vector a and the feature vector b. When cosθ is closer to 1, the directions are closer, the distance between the two feature vectors is smaller, and the similarity between the search information corresponding to the two feature vectors is higher; when cosθ is closer to -1, the directions are different, the distance between the two feature vectors is larger, and the similarity between the search information corresponding to the two feature vectors is lower.

[0095] The electronic device may determine the target candidate information from the candidate information set based on the calculated similarity as the recommended search information, and recommend the recommended search information to the user in the above step S105. In one embodiment, the electronic device may sort the calculated similarities to obtain a sorting result, and then determine a preset number of candidate information at the top of the sorting result as the target candidate information.

[0096] For example, the electronic device determines that the similarities between the feature vectors corresponding to candidate information 1, candidate information 2, candidate information 3, candidate information 4, and candidate information 5 and the target feature vector are 0.75, 0.55, 0.99, 0.80, and 0.86, respectively. The electronic device can sort the candidate information in descending order based on the similarity, and the sorting results are shown in the following table:

[0097] Serial number Similarity Sorting results 1 0.99 Candidate information 3 2 0.86 Candidate information 5 3 0.80 Candidate information 4 4 0.70 Candidate information 1 5 0.55 Candidate information 2

[0098] Then, in one implementation, the electronic device may recommend the candidate information 3 with the highest similarity as the recommended search information to the user based on the ranking results of the similarities shown in the above table.

[0099] Through the above-mentioned search information recommendation method, since it is not necessary to determine the recalled search information based on the correspondence between the search information and its recalled search information, even if the target search information is new search information input by the user, the electronic device can determine the recommended search information from the candidate set, successfully complete the recall of the search information, and improve the recommendation effect of the search information.

[0100] As an implementation manner of an embodiment of the present invention, the step of determining the feature vector corresponding to each candidate information in the candidate information set may include:

[0101] For each candidate information included in the candidate information set, if the candidate information exists in the recall pool dictionary, the feature vector corresponding to the candidate information in the recall pool dictionary is determined as the feature vector corresponding to the candidate information; if the candidate information does not exist in the recall pool dictionary, the candidate information is input into a pre-trained feature vector model to obtain the feature vector corresponding to the candidate information output by the feature vector model.

[0102] The candidate information may or may not exist in the recall pool dictionary, so the electronic device can determine whether each candidate information exists in the recall pool dictionary. In the case where the candidate information exists in the recall pool dictionary, since the recall pool dictionary already includes the correspondence between the search information and the feature vector, the electronic device can directly determine the feature vector corresponding to the candidate information in the recall pool dictionary as the feature vector corresponding to the candidate information, so that there is no need to calculate the feature vector corresponding to the candidate information, which can improve processing efficiency.

[0103] In the case where the candidate information does not exist in the recall pool dictionary, the electronic device can input the candidate information into a pre-trained feature vector model to obtain a feature vector corresponding to the candidate information output by the feature vector model. The feature vector model can be a deep learning model pre-trained based on the search information sample, for example, a convolutional neural network.

[0104] The search information sample may include various supervisory information, for example, user historical search behavior information, knowledge graph information related to the search information, etc., which are not specifically limited here, as long as a feature vector that can accurately represent the search information characteristics can be obtained through training. The training method of the feature vector model can be a gradient descent algorithm, a stochastic gradient descent algorithm, etc., which are not specifically limited here.

[0105] After the electronic device inputs the candidate information into the pre-trained feature vector model, the feature vector model can process the candidate information based on its own model parameters and finally output the feature vector corresponding to the candidate information.

[0106] It can be seen that in this embodiment, when the candidate information exists in the recall pool dictionary, the electronic device can directly use the feature vector corresponding to the candidate information in the recall pool dictionary as the feature vector corresponding to the candidate information, which is convenient and quick; when the candidate information does not exist in the recall pool dictionary, the electronic device can input it into the feature vector model, thereby obtaining the feature vector corresponding to the candidate information output by the feature vector model. In either case, an accurate feature vector can be obtained.

[0107] As an implementation method of the embodiment of the present invention, Figure 2 As shown, the step of determining the target feature vector corresponding to the target search information based on the word segmentation features of the target search information may include:

[0108] S201, performing word segmentation processing on the target search information to obtain a word segmentation feature of the target search information as a first word segmentation feature;

[0109] Since the word segmentation feature of the search information can well identify the features of the search information, the electronic device can perform word segmentation processing on the target search information, and then obtain the word segmentation feature of the target search information as the first word segmentation feature. The word segmentation feature may include word segmentation token, word segmentation part of speech, character and other features, which are not specifically limited here.

[0110] S202: Input the first word segmentation feature into a pre-trained feature vector model to obtain a feature vector output by the feature vector model based on the first word segmentation feature as a target feature vector corresponding to the target search information.

[0111] After obtaining the first analysis feature, the electronic device can input the first word segmentation feature into the pre-trained feature vector model, and the feature vector model can process the first analysis feature based on its own model parameters and output a feature vector. The electronic device can also obtain the feature vector output by the feature vector model and use it as the target feature vector corresponding to the target search information.

[0112] It can be seen that in this embodiment, the electronic device can perform word segmentation processing on the target search information, obtain the word segmentation feature of the target search information as the first word segmentation feature, and then input the first word segmentation feature into the pre-trained feature vector model to obtain the feature vector output by the feature vector model as the target feature vector corresponding to the target search information. In this way, the target feature vector corresponding to the target search information can be accurately determined to ensure that appropriate recommended search information can be determined later, thereby improving the search information recommendation effect.

[0113] As an implementation method of the embodiment of the present invention, Figure 3 As shown, the method for establishing the recall pool dictionary may include:

[0114] S301, obtaining a plurality of search information collected in advance;

[0115] The electronic device may obtain a plurality of pre-collected search information, wherein the plurality of search information may be search information in a user's historical search behavior, or may be search information set according to various types of search needs, etc., which is not specifically limited here.

[0116] In one implementation, in order to improve the recall quality, the electronic device may also perform screening, cleaning, and other processing on the plurality of pre-collected search information to retain search information with better quality for establishing a recall pool dictionary.

[0117] S302, performing word segmentation processing on each search information to obtain a word segmentation feature of each search information as a second word segmentation feature;

[0118] Next, the electronic device can perform word segmentation processing on each search information to obtain the word segmentation feature of each search information as the second word segmentation feature. The word segmentation feature may include word segmentation token, word segmentation part of speech, character and other features, which are not specifically limited here. The specific method of word segmentation processing is not specifically limited here, as long as the word segmentation feature of the search information can be obtained.

[0119] S303, inputting the second word segmentation feature into a pre-trained feature vector model, and obtaining a feature vector output by the feature vector model based on the second word segmentation feature as a feature vector corresponding to the search information;

[0120] After determining the second word segmentation feature of the search information, the electronic device can input the second word segmentation feature into the pre-trained feature vector model, and the feature vector model can process the second word segmentation feature based on its own model parameters, and then output the feature vector corresponding to the search information.

[0121] S304, recording the search information and its corresponding feature vector to obtain a recall pool dictionary.

[0122] After obtaining the feature vectors corresponding to each search information, the electronic device can record the search information and its corresponding feature vectors, and thus obtain a recall pool dictionary. In one implementation, the recall pool dictionary can be recorded in a table, for example, as shown in the following table:

[0123] Serial number Search Information Eigenvector 1 Search Information A Eigenvector a 2 Search Information B Eigenvector b 3 Search Information C Eigenvector c … … … n Search InformationN Eigenvector n

[0124] It can be seen that in this embodiment, the electronic device can pre-establish a recall pool dictionary through the above-mentioned establishment method, and can obtain a recall pool dictionary including the correspondence between search information and feature vectors, which is convenient for finding alternative recommended information when the user searches for information and ensures the search information recommendation effect.

[0125] As an implementation method of the embodiment of the present invention, Figure 4 As shown, the above method may also include:

[0126] S401, updating the feature vector model;

[0127] In order to improve the accuracy of the feature vector output by the feature vector model, the feature vector model can be updated. In one embodiment, the feature vector model can be updated once every preset time interval, for example, every one day, every three days, every five days, etc., which is not specifically limited here. In another embodiment, the feature vector model can be continuously updated, which is also reasonable.

[0128] Specifically, new search information samples may be obtained, and then the feature vector model may be trained based on the new search information samples, and the model parameters of the feature vector model may be updated, so that the feature vector output by the feature vector model becomes more and more accurate.

[0129] S402: Update the recall pool dictionary based on the updated feature vector model.

[0130] Since the feature vectors included in the recall pool dictionary are determined based on the feature vector model, in order to ensure the accuracy of the feature vectors included in the recall pool dictionary, after the feature vector model is updated, the recall pool dictionary may be updated based on the updated feature vector model.

[0131] During the process of updating the recall pool dictionary, the electronic device can input the search information included in the recall pool dictionary into the updated feature vector model, obtain the feature vector output by the updated feature vector model, and then use the feature vector output by the updated feature vector model to replace the feature vector corresponding to the search information included in the recall pool dictionary, thereby achieving the purpose of updating the recall pool dictionary.

[0132] In actual search application scenarios, the feature vector model can be trained and updated offline, the recall pool dictionary can be established and updated offline, and then the feature vector model and the recall pool dictionary can be uploaded to the online search device at regular intervals. For example, the feature vector model and the recall pool dictionary can be uploaded to the online search device every day. In this way, the online search device can recommend search information based on the feature vector model and the recall pool dictionary to meet the user's search needs and achieve a good drainage effect.

[0133] It can be seen that in this embodiment, the above method can also update the feature vector model, and update the recall pool dictionary based on the updated feature vector model. In this way, the accuracy of the feature vector model and the recall pool dictionary can be gradually improved, thereby improving the recommendation effect of the search information.

[0134] As an implementation method of the embodiment of the present invention, Figure 5 As shown, the step of determining the target candidate information from the candidate information set based on the similarity may include:

[0135] S501, inputting each candidate information in the candidate information set and its corresponding similarity into a pre-trained ranking model, and obtaining a score corresponding to each candidate information output by the ranking model based on the similarity;

[0136] In order to accurately and quickly determine the target candidate information, a ranking model can be pre-trained to score each candidate information. The ranking model can be a convolutional neural network, a recurrent neural network, etc., which is not specifically limited here.

[0137] Since the similarity corresponding to the candidate information can represent the degree of association between the candidate information and the target search information, the electronic device can input each candidate information and its corresponding similarity into the sorting model, that is, the similarity corresponding to the candidate information can be input into the sorting model as a one-dimensional feature, and then the sorting model can output the score corresponding to each candidate information based on the similarity. The score can represent the feasibility of the candidate information as the recommended search information.

[0138] The present invention does not specifically limit the training process of the ranking model. Training samples can be obtained according to actual needs and application scenarios, and the ranking model can be trained using a gradient descent algorithm or the like, so as to obtain a ranking model capable of outputting a score. The training samples may include candidate information samples and the similarity between the candidate information and the target search sample, etc., which are not specifically limited here.

[0139] S502: Determine target candidate information from the candidate information set according to the score corresponding to each candidate information.

[0140] After obtaining the score corresponding to each candidate information, the electronic device can determine the target candidate information from the candidate information set according to the score corresponding to each candidate information. In one embodiment, the electronic device can sort the candidate information according to the score corresponding to each candidate information, and then select a preset number of candidate information with higher scores from the candidate information according to the sorting result as the target candidate information.

[0141] For example, the candidate information set includes candidate information 1-candidate information 10, and the scores of candidate information 1-candidate information 10 output by the sorting model are 15, 27, 29, 95, 87, 67, 5, 78, 36, and 88, respectively. Then the electronic device can sort candidate information 1 in descending order of scores, and obtain the sorting results of candidate information 4, candidate information 10, candidate information 5, candidate information 8, candidate information 6, candidate information 9, candidate information 2, candidate information 1, and candidate information 7. Assuming that the preset number is 3, the electronic device can select the first three candidate information in the sorting result as the target candidate information, namely candidate information 4, candidate information 10, and candidate information 5.

[0142] It can be seen that in this embodiment, the electronic device can input each candidate information in the candidate information set and its corresponding similarity into the pre-trained sorting model to obtain the score corresponding to each candidate information output by the sorting model based on the similarity, and then determine the target candidate information from the candidate information set according to the score corresponding to each candidate information. In this way, similarity is input into the sorting model as a one-dimensional feature, so that the sorting model can output an accurate score, and the electronic device can determine the accurate target candidate information based on the score, thereby ensuring the recommendation effect of the search information.

[0143] Corresponding to the above-mentioned method for recommending search information, an embodiment of the present invention further provides a device for recommending search information. The following is an introduction to the device for recommending search information provided by an embodiment of the present invention.

[0144] like Figure 6 As shown, a device for recommending search information includes:

[0145] A first vector determination module 610 is used to obtain target search information input by a user, and determine a target feature vector corresponding to the target search information based on the word segmentation features of the target search information;

[0146] The candidate recommendation information determination module 620 is used to determine the candidate recommendation information from the search information included in the recall pool dictionary according to the similarity between the target feature vector and the feature vector of the search information included in the recall pool dictionary obtained in advance by the recall pool establishment module;

[0147] A second vector determination module 630, configured to add the candidate recommendation information to an existing candidate information set, and determine a feature vector corresponding to each candidate information in the candidate information set;

[0148] A recommended search information determination module 640 is used to calculate the similarity between the feature vector corresponding to each candidate information and the target feature vector, and determine the target candidate information from the candidate information set based on the similarity as the recommended search information;

[0149] The search information recommendation module 650 is used to recommend the recommended search information to the user.

[0150] It can be seen that in the scheme provided by the embodiment of the present invention, the electronic device can obtain the target search information input by the user, and determine the target feature vector corresponding to the target search information based on the word segmentation features of the target search information, determine the alternative recommended information from the search information included in the recall pool dictionary according to the similarity between the target feature vector and the feature vector of the search information included in the pre-acquired recall pool dictionary, add the alternative recommended information to the existing candidate information set, and determine the target feature vector corresponding to each candidate information in the candidate information set, calculate the similarity between the feature vector corresponding to each candidate information and the target feature vector, and determine the target candidate information from the candidate information set based on the similarity as the recommended search information, and then recommend the recommended search information to the user. Since the electronic device adds the alternative recommended information determined based on the similarity between feature vectors to the candidate information set, the candidate set includes multiple candidate information with high similarity to the target search information. Furthermore, the electronic device determines the target candidate information based on the similarity between the feature vector corresponding to each candidate information and the target feature vector, and can further select the target candidate information with high similarity to the target feature vector from the candidate information set as the recommended search information. In this way, since there is no need to determine the recalled search information based on the correspondence between the search information and its recalled search information, even if the target search information is new search information input by the user, the electronic device can determine the recommended search information from the candidate set, successfully complete the recall of the search information, and improve the recommendation effect of the search information.

[0151] As an implementation of the embodiment of the present invention, the second vector determination module 630 may include:

[0152] a first vector determination unit, configured to determine, for each candidate information included in the candidate information set, a feature vector corresponding to the candidate information in the recall pool dictionary as a feature vector corresponding to the candidate information if the candidate information exists in the recall pool dictionary;

[0153] The second vector determination unit is used to input the candidate information into a pre-trained feature vector model when the candidate information does not exist in the recall pool dictionary, and obtain a feature vector corresponding to the candidate information output by the feature vector model.

[0154] As an implementation method of the embodiment of the present invention, Figure 7 As shown, the first vector determination module 610 may include:

[0155] A first word segmentation processing unit 611 is used to perform word segmentation processing on the target search information to obtain a word segmentation feature of the target search information as a first word segmentation feature;

[0156] The vector determination unit 612 is used to input the first word segmentation feature into a pre-trained feature vector model to obtain a feature vector output by the feature vector model based on the first word segmentation feature as a target feature vector corresponding to the target search information.

[0157] As an implementation method of the embodiment of the present invention, Figure 8 As shown, the recall pool establishment module may include:

[0158] The information acquisition unit 801 is used to acquire a plurality of pre-collected search information;

[0159] The second word segmentation processing unit 802 is used to perform word segmentation processing on each search information to obtain a word segmentation feature of each search information as a second word segmentation feature;

[0160] A third vector determination unit 803 is configured to input the second word segmentation feature into a pre-trained feature vector model, and obtain a feature vector output by the feature vector model based on the second word segmentation feature as a feature vector corresponding to the search information;

[0161] The recall pool establishing unit 804 is used to record the search information and its corresponding feature vector to obtain a recall pool dictionary.

[0162] As an implementation of the embodiment of the present invention, the above device may further include:

[0163] A first model updating module, used for updating the feature vector model;

[0164] The second model updating module is used to update the recall pool dictionary based on the updated feature vector model.

[0165] As an implementation of an embodiment of the present invention, the recommended search information determination module 640 may include:

[0166] A score determination unit, used for inputting each candidate information in the candidate information set and its corresponding similarity into a pre-trained sorting model, and obtaining a score corresponding to each candidate information output by the sorting model based on the similarity;

[0167] The candidate information determining unit is used to determine target candidate information from the candidate information set according to the score corresponding to each candidate information.

[0168] The embodiment of the present invention further provides an electronic device, such as Fig. 9 As shown, it includes a processor 901, a communication interface 902, a memory 903 and a communication bus 904, wherein the processor 901, the communication interface 902, and the memory 903 communicate with each other through the communication bus 904.

[0169] Memory 903, used for storing computer programs;

[0170] The processor 901 is configured to implement the steps of the recommended method for searching information described in any of the above embodiments when executing the program stored in the memory 903 .

[0171] It can be seen that in the scheme provided by the embodiment of the present invention, the electronic device can obtain the target search information input by the user, and determine the target feature vector corresponding to the target search information based on the word segmentation features of the target search information, determine the alternative recommended information from the search information included in the recall pool dictionary according to the similarity between the target feature vector and the feature vector of the search information included in the pre-acquired recall pool dictionary, add the alternative recommended information to the existing candidate information set, and determine the target feature vector corresponding to each candidate information in the candidate information set, calculate the similarity between the feature vector corresponding to each candidate information and the target feature vector, and determine the target candidate information from the candidate information set based on the similarity as the recommended search information, and then recommend the recommended search information to the user. Since the electronic device adds the alternative recommended information determined based on the similarity between feature vectors to the candidate information set, the candidate set includes multiple candidate information with high similarity to the target search information. Furthermore, the electronic device determines the target candidate information based on the similarity between the feature vector corresponding to each candidate information and the target feature vector, and can further select the target candidate information with high similarity to the target feature vector from the candidate information set as the recommended search information. In this way, since there is no need to determine the recalled search information based on the correspondence between the search information and its recalled search information, even if the target search information is new search information input by the user, the electronic device can determine the recommended search information from the candidate set, successfully complete the recall of the search information, and improve the recommendation effect of the search information.

[0172] The communication bus mentioned in the above electronic device can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The communication bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, only one thick line is used in the figure, but it does not mean that there is only one bus or one type of bus.

[0173] The communication interface is used for communication between the above electronic device and other devices.

[0174] The memory may include a random access memory (RAM) or a non-volatile memory, such as at least one disk memory. Optionally, the memory may also be at least one cache device located away from the aforementioned processor.

[0175] The above-mentioned processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0176] In another embodiment provided by the present invention, a computer-readable cache medium is provided, in which instructions are cached. When the computer-readable cache medium is executed on a computer, the computer executes the method for recommending search information described in any of the above embodiments.

[0177] In another embodiment of the present invention, a computer program product including instructions is provided. When the computer program product is executed on a computer, the computer executes the method for recommending search information described in any one of the above embodiments.

[0178] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented by software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in the embodiment of the present invention is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be cached in a computer-readable cache medium, or transmitted from one computer-readable cache medium to another computer-readable cache medium. For example, the computer instructions can be transmitted from one website site, computer, server or data center to another website site, computer, server or data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.). The computer-readable cache medium can be any available medium that a computer can access or a data cache device such as a server or data center that includes one or more available media integrated. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state hard disk Solid State Disk (SSD)), etc.

[0179] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0180] Each embodiment in this specification is described in a related manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device, electronic device, computer-readable storage medium, and computer program product embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiments.

[0181] The above description is only a preferred embodiment of the present invention and is not intended to limit the protection scope of the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention are included in the protection scope of the present invention.

Claims

1. A method for recommending search information, characterized in that: The method comprises: Acquire target search information input by a user, and determine a target feature vector corresponding to the target search information based on the word segmentation features of the target search information; Determining candidate recommendation information from the search information included in the recall pool dictionary according to the similarity between the target feature vector and the feature vector of the search information included in the recall pool dictionary obtained in advance; adding the candidate recommendation information to an existing candidate information set, and determining a feature vector corresponding to each candidate information in the candidate information set, wherein the existing candidate information set includes search information recorded in a user's historical search behavior process, and after the candidate recommendation information is added to the existing candidate information set, the candidate information set includes the existing candidate information and the newly added candidate recommendation information; Calculating the similarity between the feature vector corresponding to each candidate information and the target feature vector, and determining the target candidate information from the candidate information set based on the similarity between the feature vector corresponding to the candidate information and the target feature vector as the recommended search information; recommending the recommended search information to the user; The step of determining the feature vector corresponding to each candidate information in the candidate information set includes: For each candidate information included in the candidate information set, if the candidate information exists in the recall pool dictionary, determine the feature vector corresponding to the candidate information in the recall pool dictionary as the feature vector corresponding to the candidate information; In the case that the candidate information does not exist in the recall pool dictionary, the candidate information is input into a pre-trained feature vector model to obtain a feature vector corresponding to the candidate information output by the feature vector model.

2. The method according to claim 1, characterized in that The step of determining a target feature vector corresponding to the target search information based on the word segmentation features of the target search information comprises: Performing word segmentation processing on the target search information to obtain a word segmentation feature of the target search information as a first word segmentation feature; The first word segmentation feature is input into a pre-trained feature vector model to obtain a feature vector output by the feature vector model based on the first word segmentation feature as a target feature vector corresponding to the target search information.

3. The method according to any one of claims 1 to 2, characterized in that: The method for establishing the recall pool dictionary includes: Obtaining multiple pre-collected search information; Perform word segmentation processing on each search information to obtain a word segmentation feature of each search information as a second word segmentation feature; Inputting the second word segmentation feature into a pre-trained feature vector model, obtaining a feature vector output by the feature vector model based on the second word segmentation feature as a feature vector corresponding to the search information; The search information and its corresponding feature vector are recorded accordingly to obtain a recall pool dictionary.

4. The method according to claim 3, characterized in that The method further comprises: updating the feature vector model; Based on the updated feature vector model, the recall pool dictionary is updated.

5. The method according to any one of claims 1 to 2, characterized in that: The step of determining target candidate information from the candidate information set based on the similarity between the feature vector corresponding to the candidate information and the target feature vector comprises: Input each candidate information in the candidate information set and its corresponding similarity into a pre-trained ranking model, and obtain a score corresponding to each candidate information output by the ranking model based on the similarity between the feature vector corresponding to the candidate information and the target feature vector; According to the score corresponding to each candidate information, target candidate information is determined from the candidate information set.

6. A device for recommending search information, characterized in that: The device comprises: A first vector determination module, configured to obtain target search information input by a user, and determine a target feature vector corresponding to the target search information based on the word segmentation features of the target search information; a candidate recommendation information determination module, configured to determine candidate recommendation information from the search information included in the recall pool dictionary according to the similarity between the target feature vector and the feature vector of the search information included in the recall pool dictionary obtained in advance by the recall pool establishment module; a second vector determination module, configured to add the candidate recommendation information to an existing candidate information set, and determine a feature vector corresponding to each candidate information in the candidate information set, wherein the existing candidate information set includes search information recorded in the user's historical search behavior process, and after the candidate recommendation information is added to the existing candidate information set, the candidate information set includes the existing candidate information and the newly added candidate recommendation information; a recommended search information determination module, configured to calculate the similarity between the feature vector corresponding to each candidate information and the target feature vector, and determine the target candidate information from the candidate information set based on the similarity between the feature vector corresponding to the candidate information and the target feature vector, as the recommended search information; A search information recommendation module, used to recommend the recommended search information to the user; The second vector determination module includes: a first vector determination unit, configured to determine, for each candidate information included in the candidate information set, a feature vector corresponding to the candidate information in the recall pool dictionary as a feature vector corresponding to the candidate information if the candidate information exists in the recall pool dictionary; The second vector determination unit is used to input the candidate information into a pre-trained feature vector model when the candidate information does not exist in the recall pool dictionary, and obtain a feature vector corresponding to the candidate information output by the feature vector model.

7. The device according to claim 6, characterized in that The first vector determination module comprises: A first word segmentation processing unit, configured to perform word segmentation processing on the target search information to obtain a word segmentation feature of the target search information as a first word segmentation feature; A vector determination unit is used to input the first word segmentation feature into a pre-trained feature vector model to obtain a feature vector output by the feature vector model based on the first word segmentation feature as a target feature vector corresponding to the target search information.

8. The device according to any one of claims 6 to 7, characterized in that: The recall pool establishment module includes: An information acquisition unit, used to acquire a plurality of pre-collected search information; A second word segmentation processing unit, configured to perform word segmentation processing on each search information, and obtain a word segmentation feature of each search information as a second word segmentation feature; A third vector determination unit is used to input the second word segmentation feature into a pre-trained feature vector model to obtain a feature vector output by the feature vector model based on the second word segmentation feature as a feature vector corresponding to the search information; The recall pool establishing unit is used to record the search information and its corresponding feature vector to obtain a recall pool dictionary.

9. The device according to claim 8, characterized in that The device also includes: A first model updating module, used for updating the feature vector model; The second model updating module is used to update the recall pool dictionary based on the updated feature vector model.

10. The device according to any one of claims 6 to 7, characterized in that: The recommended search information determination module includes: A score determination unit, used to input each candidate information in the candidate information set and its corresponding similarity into a pre-trained ranking model, and obtain a score corresponding to each candidate information output by the ranking model based on the similarity between the feature vector corresponding to the candidate information and the target feature vector; The candidate information determining unit is used to determine target candidate information from the candidate information set according to the score corresponding to each candidate information.

11. An electronic device, characterized in that: It includes a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other through the communication bus; Memory, used to store computer programs; A processor, for implementing the method steps described in any one of claims 1 to 5 when executing a program stored in a memory.

12. A computer-readable cache medium, characterized in that: The computer-readable cache medium caches a computer program, and when the computer program is executed by a processor, the method steps described in any one of claims 1 to 5 are implemented.

Citation Information

Patent Citations

  • Method and device for generating search term and electronic device

    CN109189990A

  • Information recommendation method and device and terminal

    CN110543598A

  • Method and system for cold start candidate recommendation

    US20210097471A1