Search recommendation method and device, electronic equipment, chip, storage medium and computer program product

Abstract

Embodiments of the present application provide a search recommendation method, a search recommendation device, an electronic device, a chip, a storage medium and a computer program product. The search recommendation method comprises: determining a plurality of search results and a plurality of first users according to first information; the first information is search information input by a user; constructing a user-search result relationship graph according to historical interaction behaviors of the plurality of first users and the plurality of search results; determining the popularity of user-search results according to the user-search result relationship graph and a first weight matrix; the first weight matrix is determined based on second information, and the first weight matrix is used to adjust the edge weight of the user-search result relationship graph; sorting the plurality of search results according to the popularity of the user-search results to obtain a first sorting result; and recommending the plurality of search results to the user according to the first sorting result.

Description

Technical Field

[0001] This application relates to the field of data service technology, specifically to a retrieval and recommendation method, a retrieval and recommendation device, an electronic device, a chip, a storage medium, and a computer program product. Background Technology

[0002] Recommendation and retrieval are complementary tools for users to obtain information. Retrieval satisfies users' proactive search needs when they have a specific purpose, while recommendation helps users discover new content of interest when they don't have a specific purpose. How to recommend popular search results to users and reduce their decision-making pressure when faced with massive amounts of information is a technical problem that needs to be solved. Summary of the Invention

[0003] This application provides a retrieval and recommendation method, a retrieval and recommendation device, an electronic device, a chip, a storage medium, and a computer program product.

[0004] The retrieval and recommendation method provided in this application includes:

[0005] Based on the first information, multiple search results and multiple first users are identified; the first information is the search information input by the user; the multiple first users are users related to the multiple search results.

[0006] Based on the historical interaction behavior between the multiple first users and the multiple search results, a user-search result relationship graph is constructed.

[0007] The popularity of user-search results is determined based on the user-search result relationship graph and the first weight matrix; the first weight matrix is ​​determined based on second information, and the first weight matrix is ​​used to adjust the edge weights of the user-search result relationship graph; the second information includes one or more of the following: the number of times the first user views the search results, the cumulative number of people who access the search results when they are accessed by the same type of first user, and the cumulative number of times the first user accesses the same type of search results;

[0008] The multiple search results are sorted according to the popularity of the user-search results to obtain a first sorting result;

[0009] Based on the first ranking result, the multiple search results are recommended to the user.

[0010] The retrieval and recommendation device provided in this application embodiment includes:

[0011] Retrieval unit: used to determine multiple retrieval results and multiple first users based on first information; the first information is retrieval information input by the user; the multiple first users are users related to the multiple retrieval results;

[0012] Construction unit: used to construct a user-search result relationship graph based on the historical interaction behavior between the multiple first users and the multiple search results;

[0013] The retrieval unit is used to determine the popularity of user-retrieval results based on the user-retrieval result relationship graph and the first weight matrix; the first weight matrix is ​​determined based on second information, and the first weight matrix is ​​used to adjust the edge weights of the user-retrieval result relationship graph; the second information includes one or more of the following: the number of times a first user views the retrieval results, the cumulative number of people who access the retrieval results when they are accessed by the same type of first user, and the cumulative number of times a first user accesses the same type of retrieval results;

[0014] Recommendation unit: used to sort the multiple search results according to the popularity of the user-search results to obtain a first sorting result; and to recommend the multiple search results to the user according to the first sorting result.

[0015] The electronic device provided in this application includes a processor and a memory. The memory is used to store computer programs, and the processor is used to call and run the computer programs stored in the memory to execute any of the retrieval recommendation methods provided in this application.

[0016] The chip provided in this application includes a processor for calling and running a computer program from a memory, causing a device equipped with the chip to execute any of the search and recommendation methods provided in this application.

[0017] The storage medium provided in this application embodiment is used to store a computer program that causes a computer to execute any of the search and recommendation methods provided in this application embodiment.

[0018] The computer program product provided in this application includes a computer program that, when executed by a processor, implements any of the search and recommendation methods provided in this application.

[0019] The retrieval and recommendation method provided in this application constructs a user-retrieval result relationship graph based on the historical interaction behavior of relevant users with multiple retrieval results. Then, the edge weights in the user-retrieval result relationship graph are adjusted using a first weight matrix to determine the popularity of the retrieval results. The first weight matrix is ​​determined based on second information, which includes one or more of the following: the number of times a first user views a retrieval result, the cumulative number of users accessing the retrieval result when it is accessed by the same type of first user, and the cumulative number of times a first user accesses the same type of retrieval result. By adjusting the edge weights in the user-retrieval result relationship graph, more attractive retrieval results can be recommended to users more accurately, making it easier for users to discover popular retrieval results, thereby reducing their pressure when facing massive amounts of information and improving decision-making efficiency. Attached Figure Description

[0020] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0021] Figure 1 Schematic diagram of the implementation process of the retrieval and recommendation method provided in the embodiments of this application Figure 1 ;

[0022] Figure 2 Schematic diagram of the implementation process of the retrieval and recommendation method provided in the embodiments of this application Figure 2 ;

[0023] Figure 3 A schematic diagram of the structure of the retrieval and recommendation device provided in the embodiments of this application;

[0024] Figure 4 A schematic structural diagram of an electronic device provided in the embodiments of this application;

[0025] Figure 5 This is a schematic structural diagram of the chip provided in the embodiments of this application. Detailed Implementation

[0026] The technical solutions of the embodiments of this application will now be described with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0027] It should be noted that, in the embodiments of this application, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, in the embodiments of this application, the character " / " generally indicates that the preceding and following related objects have an "or" relationship.

[0028] In the description of the embodiments of this application, the term "correspondence" may indicate that there is a direct or indirect correspondence between two things, or that there is an association between two things, or that there is a relationship of instruction and being instructed, configuration and being configured, etc.

[0029] To facilitate understanding of the technical solutions of the embodiments of this application, the relevant technologies of the embodiments of this application are described below. The following relevant technologies are optional solutions and can be combined with the technical solutions of the embodiments of this application in any way, and they all fall within the protection scope of the embodiments of this application.

[0030] Information retrieval (IR) refers to the representation, storage, organization, and access of information. Information recommendation refers to a system recommending useful information that a user may be interested in but cannot obtain. Recommendation and retrieval are complementary tools for users to acquire information. Retrieval satisfies users' proactive search needs when they have a specific purpose, while recommendation helps users discover new content of interest when they do not have a specific purpose. Due to the significant performance of large language models in language understanding, generation, generalization, and reasoning, information retrieval and recommendation based on large language models are becoming increasingly rich, and the ways of recommending to users are becoming increasingly diverse. The retrieval and recommendation methods in related technologies are as follows:

[0031] 1. Information retrieval technology based on large language models, designing rewriters, retrievers, reorderers and readers according to the query content.

[0032] 2. Query expansion of large language models: Generate relevant passages based on the query, expand the original query by merging the generated passages, combine the query and the generated passages to construct a new query, and use this new query for retrieval.

[0033] 3. Tag recommendation and search optimization based on collaborative filtering and text similarity: For web pages that users want to annotate, the number of times the web page has been annotated is calculated. If the number exceeds a specified threshold, collaborative filtering is used to find similar users, and web pages with higher total tag weights among similar users are recommended to the user. Otherwise, the text similarity formula is used to calculate similar web pages, and web pages with higher total tag weights are recommended to the user.

[0034] 4. Active Low-Pass Filter Signal Screening – Based on a matched perspective, an OPAM active low-pass filter design is used to effectively screen information. It allows useful frequency signals to pass while suppressing useless frequency signals, allowing signals in a specified frequency band to pass while providing sufficient attenuation to suppress signals in other frequency bands.

[0035] The related technologies face the following problems:

[0036] 1. How to recommend popular search results to users to reduce their decision-making pressure when faced with massive amounts of information.

[0037] 2. In information retrieval and recommendation systems (taking user-product as an example), data updates rapidly, and users’ attention to different types of products changes quickly. Currently, most technical solutions related to this proposal are troubled by the rapid data updates and the difficulty of processing incremental data. They are also troubled by the excessive consumption of data and computing power during model training and fine-tuning.

[0038] 3. In information retrieval and recommendation systems (taking user-product as an example), the accuracy of information retrieval and recommendation results refers to generating user interest tags based on users' historical behavior and personal information, and then making personalized recommendations based on these tags. While the accuracy of recommendation results is important, an excessive pursuit of accuracy often leads to overly simplistic and limited results, potentially causing decreased user satisfaction. Information retrieval and recommendation systems should provide users with more choices and opportunities to discover a variety of popular products. Therefore, the diversity (popularity) of information retrieval and recommendation results has emerged. In practical use, user interests are broad and diverse. The diversity (popularity) of information retrieval and recommendation results refers to the degree of demand users have for popular products that differ from their own interest tags. Products in the recommended results have weighted coefficients indicating differences from existing interests. Popular products with these differences are recommended based on these weighted coefficient values. There is a balance between the diversity (popularity) and accuracy (accuracy) of information retrieval and recommendation results. Finding this balance amidst the complex relationship between these two factors requires selecting appropriate adjustment parameters and filtering methods.

[0039] Figure 1This is a schematic diagram illustrating the implementation flow of the retrieval and recommendation method provided in the embodiments of this application, as follows: Figure 1 As shown in the embodiment of this application, a retrieval and recommendation method is provided, the method comprising the following steps:

[0040] Step 101: Based on the first information, determine multiple search results and multiple first users; the first information is the search information input by the user; the multiple first users are users related to the multiple search results.

[0041] In this embodiment of the application, the search results may be products, web pages, articles, business information, etc., or other types of search results. This embodiment of the application does not limit the search results in this way.

[0042] The information retrieval method provided in this application can be applied to multiple Natural Language Processing (NLP) tasks. The datasets used can include: search engine content retrieval recommendation datasets, business information-user retrieval recommendation datasets, and user-product retrieval recommendation datasets. The dataset selection process follows the following criteria: text language representativeness, text diversity, data availability and accessibility, and annotation quality. After standard preprocessing techniques, titles, footers, and special characters are removed, and word segmentation is applied to ensure dataset uniformity. Stemming and stop word removal are also incorporated into the preprocessing process. Removing irrelevant parts such as titles, footers, and special characters reduces the impact of noisy data on model training, which helps improve the model's accuracy and efficiency. Word segmentation divides continuous text into meaningful units (such as words or phrases), providing a foundation for subsequent feature extraction. Stemming restores words of different forms to their basic forms, helping to merge similar words and reduce the dimensionality of the feature space. Stop word removal filters out words that frequently appear in the text but contribute little to the meaning, thus focusing more on key information. Maintaining the representativeness and diversity of the text language allows for the learning of a broader range of language patterns, rather than overfitting to specific samples. Reducing unnecessary data volume and feature dimensions, such as by removing stop words and applying stemming, lowers computational resource requirements and helps avoid overfitting. Uniform preprocessing steps ensure consistency within the dataset, allowing different text data to be compared and analyzed on the same benchmark.

[0043] In this embodiment of the application, based on the first information, the content input by the user is segmented into words and keywords are extracted based on the Industrial Large Language Model (LLM). Based on the keywords, search results related to the keywords are retrieved from a large dataset, as well as users who have had historical user interaction with the search results. V and I represent the set of all users and all search results retrieved based on the Industrial Large Language Model, respectively.

[0044] Step 102: Construct a user-search result relationship graph based on the historical interaction behavior between the multiple first users and the multiple search results.

[0045] In this embodiment of the application, historical interaction behaviors include browsing, accessing, purchasing, and collecting, and may also include other types of interaction behaviors. This embodiment of the application does not limit these types of interaction behaviors.

[0046] The goal of constructing a relationship graph between users and search results and calculating the popularity of users and search results is to find diverse and differentiated search results that are different from the user's own interest tags and recommend them to the user. Popularity is used to increase the diversity of search results and prevent the recommended results from being limited to the user's own interest tags, which would lead to a single type of search result recommendation and user visual fatigue.

[0047] Step 103: Determine the popularity of user-search results based on the user-search result relationship graph and the first weight matrix; the first weight matrix is ​​determined based on second information, and the first weight matrix is ​​used to adjust the edge weights of the user-search result relationship graph; the second information includes one or more of the following: the number of times the first user views the search results, the cumulative number of people who access the search results when they are accessed by the same type of first user, and the cumulative number of times the first user accesses the same type of search results.

[0048] In this embodiment, the popularity of user-search results can be normalized by using asymmetric normalization based on the user-search result relationship graph to obtain initial edge weights. Then, the initial edge weights are adjusted by the first weight matrix to determine the final popularity of user-search results.

[0049] Asymmetric normalization allows for more granular adjustment of preference weights for each user and the popularity of each search result, enabling more personalized recommendations. This approach helps discover niche but high-quality search results, meeting diverse user needs.

[0050] By adjusting the initial edge weights using the first weight matrix, more attractive search results can be recommended to users more accurately, making it easier for them to discover popular search results, thereby reducing their stress when faced with massive amounts of information and improving decision-making efficiency.

[0051] Based on this, in an optional embodiment of this application, determining the popularity of user-search results based on the user-search result relationship graph and the first weight matrix includes:

[0052] Based on the user-retrieval result relationship graph, the initial edge weights of the user-retrieval result relationship graph are obtained through asymmetric normalization.

[0053] The popularity of the user-search results is determined by adjusting the initial edge weights of the user-search result relationship graph using the first weight matrix.

[0054] In this embodiment of the application, the popularity of user-search results can be standardized in the following manner.

[0055]

[0056] Where, R∈R |v|x|I| R is a rating matrix, which represents the user's ratings of various search results based on their historical behavior. It is a standardized rating matrix, which represents the quantified values ​​of the ratings in the rating matrix in the range of 0 to 1. V and I represent the sets of all users and all search results in the information retrieval and recommendation system obtained in step 101, respectively. D V =diag(R1),D I =diag(1 T R) are the search result-search result adjacency matrix and the user-user adjacency matrix, respectively, representing similar search results and similar users.

[0057] In this embodiment, α∈[0,1] is a hyperparameter controlling the standardization of user-retrieval results. Increasing α enhances the popularity of user-retrieval results, meaning that the recommended results include more diverse and differentiated retrieval results that differ from the user's own interest tags. However, in large datasets, it is not only necessary to find retrieval results that differ from the user's own interest tags, but also to calculate the weights of these retrieval results that the user may be interested in. Retrieval results with higher weights are more likely to be of interest to the user, and these results should be prioritized and displayed to the user. The following steps describe how to calculate the weights.

[0058] For example, User-search result popularity

[0059]

[0060] In this embodiment of the application, the first weight matrix w is used to... To make adjustments, we designed an edge weight adjustment scheme for a graph showing the relationship between users and search results. The edge weight represents the weight of the relationship between each pair of users and search results. The goal is to find search results that might be of interest to users, but are different from their own interest tags, and then sort them, displaying the results with higher weights first. The adjusted calculation is as follows:

[0061]

[0062] Where i∈I and j∈V represent any user and search result in I and V respectively, w ij It's about weights, control. The distribution is normalized using three variables, the first being a. ij This refers to the number of times the j-th user views the i-th search result, corresponding to the number of times the first user views the search result in the second information, where the second is b. ij That is, the cumulative number of visitors to the i-th search result when accessed by users of the same type j, corresponding to the cumulative number of visitors to the search results in the second information when accessed by users of the same type first, and the third variable c. ij This represents the cumulative number of times the j-th user accesses the same type of search results (i), corresponding to the cumulative number of times the first user accesses the same type of search results in the second information. Finally, the softmax function is used to convert the edge weights into a probability distribution. The edge weight values ​​are quantized in the range of 0 to 1. Edge weights with higher probabilities indicate that the user is more likely to be interested in that type of search result, and such search results are recommended to the user, making it more likely that the user will be satisfied with the results.

[0063] In this embodiment, the same type of user can refer to users with the same number of visits to a specific search result or users with the same number of views on a specific search result. In practical applications, users can be classified according to the actual situation, and this embodiment does not limit this. It is understood that the same type of search result can also be classified according to the actual situation. Taking user-product as an example, it can be classified according to the product categories of the shopping website.

[0064] Step 104: Sort the multiple search results according to the popularity of the user-search results to obtain the first sorting result.

[0065] Based on the popularity of user-search results, the search results are sorted according to popularity, with search results of higher popularity appearing first and search results of lower popularity appearing last.

[0066] Step 105: Based on the first sorting result, recommend the multiple search results to the user.

[0067] Based on the first ranking result, multiple search results are recommended to the user. This can recommend more attractive search results to the user, making it easier for the user to discover popular search results, thereby reducing the pressure on them when faced with massive amounts of information and improving decision-making efficiency.

[0068] In this embodiment of the application, the user's own interest tags can also be adjusted according to the user's feedback. For example, the user's interest tags can be adjusted according to the user's behavior, such as the number of times the user views or visits a specific search result, to obtain the user's adjusted interest tags.

[0069] Based on this, in an optional embodiment of this application, the method further includes:

[0070] Obtain user feedback information;

[0071] Based on the feedback information, the user's interest tags are adjusted to obtain the user's adjusted interest tags.

[0072] In this embodiment, in addition to ensuring the popularity of the recommended results, accuracy also needs to be considered, recommending search results that are more satisfactory to the user. Multiple search results can be filtered using the user's interest tags to obtain multiple search results that the user is interested in, and then the search results that the user is interested in are recommended to the user based on the first ranking result.

[0073] Based on this, in an optional embodiment of this application, before recommending the multiple search results to the user, the method further includes:

[0074] The multiple search results are filtered based on the user's interest tags;

[0075] Get multiple search results that the user is interested in;

[0076] The step of recommending the multiple search results to the user based on the first ranking result includes:

[0077] Based on the first ranking result and the multiple search results that the user is interested in, the multiple search results that the user is interested in are recommended to the user.

[0078] In this embodiment, a polynomial low-pass graph filter is designed to filter out the search results corresponding to the user's interest tags. Based on the user's interest tags, the search results that the user is interested in are filtered out. Here, the user's interest tags can be the interest tags before adjustment, or the interest tags after adjustment based on the user's feedback after the user's previous search recommendation.

[0079] In this embodiment, since the Graph Laplace matrix L is a real symmetric matrix, it can be orthogonally diagonalized, i.e., there exist an orthogonal matrix U and a diagonal matrix Λ such that L = UΛU T In this decomposition, the elements on the diagonal of Λ are the eigenvalues ​​of L, and the columns of U are the corresponding eigenvectors. This embodiment of the application utilizes this property of the graph Laplacian matrix L to design a graph filter and calculate the accuracy of the user-retrieval results.

[0080] Will The normalized Laplace matrix is ​​expressed as:

[0081]

[0082] because It is a symmetric positive semi-definite matrix, therefore it has an orthogonal eigendecomposition.

[0083]

[0084] in because It is an orthogonal matrix, which can be obtained through The matrix polynomial is used to calculate the eigenvalues, which represent the weight coefficients of the search results that the user is interested in from their own interest tags.

[0085] The matrix factorization polynomial graphical filter can be expressed as:

[0086]

[0087] Among them, a k Here, K represents the coefficients of the matrix polynomial, and K is the maximum order of the basis of the matrix polynomial. Based on the following theorem, which provides a closed-form solution to the frequency response function of a polynomial graph filter.

[0088] Theorem: The matrix polynomial H is a graph The graphical filter has the following frequency response function:

[0089]

[0090] The polynomial graphical filter is designed as follows:

[0091]

[0092] Where, r v Is R∈R |v|x|I| The u-th row represents the interest tag matrix of user v. yes A polynomial graph filter is used to find the search results corresponding to user v's interest tags and calculate the user-search result accuracy, that is, to recommend corresponding search results based on user interest tags.

[0093] Based on this, in an optional embodiment of this application, the filtering operation on the multiple search results according to the user's interest tags includes:

[0094] Based on the user's interest tags, the multiple search results are filtered using a polynomial low-pass graph filter.

[0095] In this embodiment of the application, simply finding the target search result is far from enough; it is also necessary to calculate the weight coefficient of the search result. The larger the coefficient, the more likely the corresponding search result should be displayed to the user. The following process describes how to calculate the coefficient:

[0096] The coefficients of the polynomial low-pass graph filter are calculated using an approximate fitting method. Larger coefficients indicate a higher priority for displaying the corresponding search result in the recommendation results, and the magnitude of the coefficients affects the order in which search results are displayed. While ideal low-pass filters have good linear fitting and a good ability to identify the relationship between user interest tags and search results, in practical use, ideal low-pass filters are almost impossible to achieve; most are nonlinear fitting low-pass filters. Therefore, this embodiment uses an approximate fitting method to calculate the coefficients of the polynomial low-pass graph filter. The combination function of the ideal low-pass filter can be expressed as:

[0097]

[0098] Here, β∈[0,1] is a hyperparameter, and increasing β will enhance the accuracy of user-search results. It is an approximate ideal low-pass filter with a cutoff frequency of τ. Taking the cutoff frequency τ = 0.1, we select the approximate coefficient λ ≤ π in formula (7), i.e., the weight value K = 3. K represents the maximum order, indicating that using a third-order polynomial function to approximate the coefficients of the polynomial graph filter yields the optimal polynomial low-pass graph filter. In the current scenario, this is the closest to the ideal low-pass filter, which helps improve the accuracy of the relationship between user interest tags and search results. Based on this, the coefficients calculated in the following process can better identify the display order of search results in the recommendation results. Therefore, the coefficients of the polynomial graph filter can be approximated as follows:

[0099]

[0100] Based on the polynomial graph filter with adjusted coefficients according to formulas (8) and (10), the accuracy of the search results for user v can be calculated, that is, the user-search result recommendation accuracy is:

[0101]

[0102] in, These are the coefficients of the polynomial graph filter after approximate calculation. The larger the coefficient, the higher the position of the search results corresponding to the user's interest tag, and the higher the priority of the ranking.

[0103] The search results that the user is interested in are sorted according to the accuracy of the user-search results to obtain a second sorting result; based on the second sorting result, the search results that the user is interested in are recommended to the user.

[0104] Based on this, in an optional embodiment of this application, when recommending multiple search results that the user is interested in, the method further includes:

[0105] The coefficients of the polynomial low-pass graph filter are calculated by approximate fitting, and the accuracy of the user-retrieval results is determined based on the coefficients of the low-pass graph filter.

[0106] Based on the accuracy of the user-search results, multiple search results of interest to the user are sorted to obtain a second sorting result;

[0107] Based on the second ranking result, multiple search results that the user is interested in are recommended to the user.

[0108] In this embodiment of the application, in order to balance the strength of the popularity and accuracy of user-search results, a joint adjustment balancing factor is designed to jointly adjust the strength of the popularity and accuracy of user-search results, as shown in the following formula:

[0109]

[0110] Where S∈R |v|x|I| It is the prediction of the recommended results that user v may be interested in, that is, the prediction of recommending multiple search results to the user; yes The first v singular vectors, Let be the degree matrix of I search results. ε is a hyperparameter, which is a balancing factor that adjusts the popularity and accuracy of user-search results. ε balances hyperparameters α and β, thereby balancing the strength of popularity and accuracy. Increasing the value of the balancing factor ε will increase the popularity of user-search results and decrease accuracy, while decreasing the value of the balancing factor ε will increase the accuracy of user-search results and decrease popularity.

[0111] Based on this, in an optional embodiment of this application, before recommending the multiple search results that the user is interested in to the user, the method further includes:

[0112] The third ranking result is obtained by balancing the popularity and accuracy of the user-retrieval results through joint adjustment of balancing factors.

[0113] Based on the third ranking result, multiple search results that the user is interested in are recommended to the user.

[0114] refer to Figure 2 , Figure 2 A schematic diagram of the implementation process of the retrieval and recommendation method provided in the embodiments of this application. Figure 2 ,like Figure 2 As shown, this embodiment takes a product recommendation system as an example and includes the following steps:

[0115] Step 201: Based on the industrial large language model, retrieve all user sets and all product sets.

[0116] Based on the initial information input by the user, all product sets are retrieved, and then based on the users who interacted with the products, all user sets are obtained.

[0117] Step 202: Calculate the popularity of user-product recommendations.

[0118] Construct a user-product relationship graph, design edge weights, adjust hyperparameters, and adjust edge weights through the first weight matrix.

[0119] Step 203: Calculate the accuracy of user-product recommendations.

[0120] The product weights are determined by filtering the products that users are interested in using a polynomial low-pass filter algorithm and then using the coefficients of the polynomial low-pass graph filter.

[0121] Step 204: Jointly regulate the balance factor.

[0122] By designing a joint adjustment balancing factor, the overall strength between user-product popularity and accuracy information is balanced.

[0123] Step 205: Recommend products to users.

[0124] By combining the popularity of user-product recommendations, the accuracy of user-product recommendations, and a joint adjustment balancing factor, products are recommended to users.

[0125] The retrieval and recommendation method provided in this application constructs a user-product relationship graph. Its purpose is to find diverse and differentiated products that differ from the user's own interest tags and recommend them to the user. The edges in the relationship graph represent the relationship weights between users and products, and the product display order is arranged according to the magnitude of the relationship weights. Then, the hyperparameters of user-product standardization are adjusted, and asymmetric normalization R is used to normalize the popularity of user-products. Next, a set of edge weights, i.e., relationship weights, are designed for the user-product relationship graph. Popular products with differences from the user's own interest tags are recommended based on these edge weights. An activation function is used to convert the edge weights into a probability distribution. Edge weights with higher probabilities indicate a higher probability that the user is interested in that type of product, and recommending such products to the user results in higher user satisfaction. Based on user feedback, the user's existing interest tags can be adjusted, thereby achieving the ability to adjust the user-product popularity by adjusting the edge weights. Next, user-product accuracy is calculated. A polynomial low-pass graph filter is designed using an approximate fitting approach to obtain an approximately ideal and low-complexity hierarchy. The coefficients of the polynomial low-pass graph filter are then calculated. Products filtered through this filter represent those within the user's interest range and are displayed to the user in order of their coefficient values. This allows adjusting the coefficients to regulate user-product accuracy. Finally, a joint adjustment balance factor is designed. Increasing the balance factor increases user-product popularity but decreases accuracy, while decreasing it increases accuracy but decreases popularity. Ultimately, this achieves the ability to recommend desired and interested products to users.

[0126] The retrieval and recommendation method provided in this application not only considers the adjustment of edge weights between pairs of relationships in the user-product relationship graph during the construction of the user-product relationship graph, but also considers the parameter adjustment of the standardized hyperparameters in the user-product relationship graph for the entire user-product structural relationship. It processes incremental data in real time and is highly sensitive to changes in new data. It adopts a three-core computing process combined with micro-control and macro-control to solve the problem of finding a balance between the diversity (popularity) and accuracy (precision) of recommendation results, and to balance the contradiction between the rapidly increasing complexity of data processing and the high data quality requirements of personalized and popular recommendations.

[0127] This application also provides a retrieval and recommendation device, for reference. Figure 3 , Figure 3 This is a schematic diagram of the retrieval and recommendation device provided in an embodiment of this application. The retrieval and recommendation device in this embodiment includes:

[0128] Retrieval unit: used to determine multiple retrieval results and multiple first users based on first information; the first information is retrieval information input by the user; the multiple first users are users related to the multiple retrieval results;

[0129] Construction unit: used to construct a user-search result relationship graph based on the historical interaction behavior between the multiple first users and the multiple search results;

[0130] The retrieval unit is used to determine the popularity of user-retrieval results based on the user-retrieval result relationship graph and the first weight matrix; the first weight matrix is ​​determined based on second information, and the first weight matrix is ​​used to adjust the edge weights of the user-retrieval result relationship graph; the second information includes one or more of the following: the number of times a first user views the retrieval results, the cumulative number of people who access the retrieval results when they are accessed by the same type of first user, and the cumulative number of times a first user accesses the same type of retrieval results;

[0131] Recommendation unit: used to sort the multiple search results according to the popularity of the user-search results to obtain a first sorting result; and to recommend the multiple search results to the user according to the first sorting result.

[0132] In this embodiment of the application, the retrieval unit is used to: obtain the initial edge weights of the user-retrieval result relationship graph through asymmetric normalization based on the user-retrieval result relationship graph; and adjust the initial edge weights of the user-retrieval result relationship graph through the first weight matrix to determine the popularity of the user-retrieval result.

[0133] In this embodiment of the application, the recommendation unit is used to filter the multiple search results according to the user's interest tags to obtain multiple search results that the user is interested in; the step of recommending the multiple search results to the user according to the first ranking result includes: recommending the multiple search results that the user is interested in to the user according to the first ranking result and the multiple search results that the user is interested in.

[0134] In this embodiment of the application, the recommendation unit is used to filter the multiple search results using a multinomial low-pass graph filter based on the user's interest tags.

[0135] In this embodiment of the application, the recommendation unit is configured to: calculate the coefficients of a polynomial low-pass graph filter by approximate fitting; determine the accuracy of the user-retrieval result based on the low-pass graph filter coefficients; sort multiple retrieval results of interest to the user based on the accuracy of the user-retrieval result to obtain a second sorting result; and recommend multiple retrieval results of interest to the user based on the second sorting result.

[0136] In this embodiment of the application, the recommendation unit is used to: balance the intensity of the popularity and accuracy of the user-search results by jointly adjusting the balance factor to obtain a third ranking result; and recommend multiple search results that the user is interested in to the user based on the third ranking result.

[0137] In this embodiment of the application, the retrieval unit is used to: obtain user feedback information; and adjust the user's interest tags according to the feedback information to obtain the user's adjusted interest tags.

[0138] Those skilled in the art should understand that Figure 3 The functions of each unit in the retrieval and recommendation device shown can be understood by referring to the relevant description of the aforementioned method. Figure 3 The functions of each unit in the retrieval and recommendation device shown can be implemented by a program running on a processor or by specific logic circuits.

[0139] Figure 4 This is a schematic structural diagram of an electronic device 400 provided in an embodiment of this application. Figure 4 The illustrated electronic device 400 includes a processor 410, which can call and run computer programs from memory to implement the methods in the embodiments of this application.

[0140] Optionally, such as Figure 4 As shown, the electronic device 400 may further include a memory 420. The processor 410 can retrieve and run computer programs from the memory 420 to implement the methods described in the embodiments of this application.

[0141] The memory 420 can be a separate device independent of the processor 410, or it can be integrated into the processor 410.

[0142] Optionally, such as Figure 4 As shown, the electronic device 400 may also include a transceiver 430, which the processor 410 can control to communicate with other devices. Specifically, it can send information or data to other devices or receive information or data sent by other devices.

[0143] The transceiver 430 may include a transmitter and a receiver. The transceiver 430 may further include an antenna, and the number of antennas may be one or more.

[0144] The electronic device 400 may specifically be a retrieval and recommendation device in the embodiments of this application, and the electronic device 400 may implement the corresponding processes implemented by the retrieval and recommendation device in the various methods of the embodiments of this application. For the sake of brevity, it will not be described in detail here.

[0145] For example, embodiments of this application also provide a computer program product, including a computer program that can be executed by the processor 410 of the communication device 400 to perform the steps described in any of the foregoing methods.

[0146] Figure 5 This is a schematic structural diagram of the chip according to an embodiment of this application. Figure 5 The chip 500 shown includes a processor 510, which can call and run computer programs from memory to implement the methods in the embodiments of this application.

[0147] Optionally, such as Figure 5 As shown, chip 500 may further include memory 520. Processor 510 can retrieve and run computer programs from memory 520 to implement the methods described in this embodiment.

[0148] The memory 520 can be a separate device independent of the processor 510, or it can be integrated into the processor 510.

[0149] Optionally, the chip 500 may also include an input interface 530. The processor 510 can control the input interface 530 to communicate with other devices or chips; specifically, it can acquire information or data sent by other devices or chips.

[0150] Optionally, the chip 500 may also include an output interface 540. The processor 510 can control the output interface 540 to communicate with other devices or chips, specifically, to output information or data to other devices or chips.

[0151] The chip can be applied to the electronic device 400 in the embodiments of this application, and the chip can implement the corresponding processes implemented by the electronic device 400 in the various methods of the embodiments of this application. For the sake of brevity, it will not be described in detail here.

[0152] It should be understood that the chip mentioned in the embodiments of this application may also be referred to as a system-on-a-chip, system chip, chip system, or system-on-a-chip, etc.

[0153] It should be understood that the processor in the embodiments of this application may be an integrated circuit chip with signal processing capabilities. In implementation, the steps of the above method embodiments can be completed by integrated logic circuits in the processor's hardware or by instructions in software form. The processor described above can be a general-purpose processor, 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, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this application. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this application can be directly embodied in the execution of a hardware decoding processor, or executed by a combination of hardware and software units in the decoding processor. The software units can be located in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. The storage medium is located in memory, and the processor reads information from the memory and, in conjunction with its hardware, completes the steps of the above method.

[0154] It is understood that the memory in the embodiments of this application can be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. The non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as Static Random Access Memory (SRAM), Dynamic Random Access Memory (DRAM), Synchronous DRAM (SDRAM), Double Data Rate SDRAM (DDR SDRAM), Enhanced Synchronous DRAM (ESDRAM), Synchlink DRAM (SLDRAM), and Direct Rambus RAM (DR RAM). It should be noted that the memory used in the systems and methods described herein is intended to include, but is not limited to, these and any other suitable types of memory.

[0155] It should be understood that the above-described memory is exemplary and not a limiting description. For example, the memory in the embodiments of this application may also be static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link dynamic random access memory (SLDRAM), and direct memory bus RAM (DR RAM), etc. That is to say, the memory in the embodiments of this application is intended to include, but is not limited to, these and any other suitable types of memory.

[0156] This application also provides a storage medium for storing a computer program. This storage medium can be applied to the electronic device 400 in this application embodiment, and the computer program causes the computer to execute the corresponding processes implemented by the electronic device 400 in the various methods of this application embodiment; for brevity, further details are omitted here.

[0157] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0158] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0159] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0160] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0161] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0162] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or electronic device 400, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0163] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A retrieval and recommendation method, characterized in that, include: Based on the initial information, multiple search results and multiple first users are identified; The first information is the search information input by the user; The plurality of first users are users related to the plurality of search results; Based on the historical interaction behavior between the multiple first users and the multiple search results, a user-search result relationship graph is constructed. Based on the user-search result relationship graph and the first weight matrix, the popularity of the user-search results is determined; The first weight matrix is ​​determined based on the second information. The first weight matrix is ​​used to adjust the edge weights of the user-retrieval result relationship graph. The second information includes one or more of the following: the number of times the first user views the retrieval results, the cumulative number of people who access the retrieval results when they are accessed by the same type of first user, and the cumulative number of times the first user accesses the same type of retrieval results. The multiple search results are sorted according to the popularity of the user-search results to obtain a first sorting result; Based on the first ranking result, the multiple search results are recommended to the user.

2. The method according to claim 1, characterized in that, The step of determining the popularity of user-search results based on the user-search result relationship graph and the first weight matrix includes: Based on the user-retrieval result relationship graph, the initial edge weights of the user-retrieval result relationship graph are obtained through asymmetric normalization. The popularity of the user-search results is determined by adjusting the initial edge weights of the user-search result relationship graph using the first weight matrix.

3. The method according to claim 2, characterized in that, Before recommending the multiple search results to the user, the following is also included: The multiple search results are filtered based on the user's interest tags; Get multiple search results that the user is interested in; The step of recommending the multiple search results to the user based on the first ranking result includes: Based on the first ranking result and the multiple search results that the user is interested in, the multiple search results that the user is interested in are recommended to the user.

4. The method according to claim 3, characterized in that, The step of filtering the multiple search results based on the user's interest tags includes: Based on the user's interest tags, the multiple search results are filtered using a polynomial low-pass graph filter.

5. The method according to claim 4, characterized in that, Recommending multiple search results that the user is interested in also includes: The coefficients of the polynomial low-pass graph filter are calculated by approximate fitting, and the accuracy of the user-retrieval results is determined based on the coefficients of the low-pass graph filter. Based on the accuracy of the user-search results, multiple search results of interest to the user are sorted to obtain a second sorting result; Based on the second ranking result, multiple search results that the user is interested in are recommended to the user.

6. The method according to claim 5, characterized in that, Before recommending the multiple search results that the user is interested in, the process also includes: The third ranking result is obtained by balancing the popularity and accuracy of the user-retrieval results through joint adjustment of balancing factors. Based on the third ranking result, multiple search results that the user is interested in are recommended to the user.

7. The method according to any one of claims 1 to 6, characterized in that, Also includes: Obtain user feedback information; Based on the feedback information, the user's interest tags are adjusted to obtain the user's adjusted interest tags.

8. A retrieval and recommendation device, characterized in that, include: Retrieval unit: used to determine multiple retrieval results and multiple first users based on the first information; The first information is the search information input by the user; The plurality of first users are users related to the plurality of search results; Construction unit: used to construct a user-search result relationship graph based on the historical interaction behavior between the multiple first users and the multiple search results; The retrieval unit is used to determine the popularity of user-retrieval results based on the user-retrieval result relationship graph and the first weight matrix. The first weight matrix is ​​determined based on the second information. The first weight matrix is ​​used to adjust the edge weights of the user-retrieval result relationship graph. The second information includes one or more of the following: the number of times the first user views the retrieval results, the cumulative number of people who access the retrieval results when they are accessed by the same type of first user, and the cumulative number of times the first user accesses the same type of retrieval results. Recommendation unit: used to sort the multiple search results according to the popularity of the user-search results to obtain a first sorting result; and to recommend the multiple search results to the user according to the first sorting result.

9. An electronic device, characterized in that, include: A processor and a memory, the memory for storing a computer program, the processor for calling and running the computer program stored in the memory to perform the retrieval recommendation method as described in any one of claims 1 to 7.

10. A chip, characterized in that, include: A processor for retrieving and running a computer program from memory, causing a device equipped with the chip to perform the retrieval recommendation method according to any one of claims 1 to 7.

11. A storage medium, characterized in that, Used to store computer programs that cause a computer to perform the retrieval recommendation method as described in any one of claims 1 to 7.

12. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the retrieval and recommendation method as described in any one of claims 1 to 7.

Images