Search result recommendation method and device, electronic device and readable storage medium

Abstract

The present invention discloses a search result recommendation method and device, an electronic device, and a readable storage medium. The method comprises: inputting a target search term into a recommendation model, wherein the recommendation model is pre-trained in a preset model using a training sample set; and displaying the recommended points of interest output by the recommendation model in a preset application interface; wherein the training samples in the training sample set include search term features of the search term, first point of interest features corresponding to the search term, and second point of interest features corresponding to similar search terms, and similar search terms all have the same points of interest. The present invention solves the technical problem that, when there are fewer natural results corresponding to the query term in the related art, the search terms and points of interest recommended for supplementary recommendations are heterogeneous, resulting in the inability to recall more search results.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a search result recommendation method and device, an electronic device, and a readable storage medium. Background Art

[0002] As a supplement to natural results, supplementary recommendations play a crucial role in improving conversion efficiency and enriching the user experience. When natural results are abundant, supplementary recommendations provide users with more diverse and generalized results. When natural results are scarce, supplementary recommendations can provide candidate points based on the user's historical records to avoid a poor search experience due to a lack of results. Graph embedding models are naturally well-suited for supplementary recommendation scenarios because they fully leverage the propagation capabilities of graphs and can learn more diverse results than other methods (such as matrix co-factorization and FM).

[0003] Currently, there are two main types of graph embedding models: DeepWalk and graph neural networks. DeepWalk uses random walks (different methods have different walk strategies) to sequentially sample the graph to recover the topological relationships between nodes and their neighbors. They then use methods like word2vec to learn low-dimensional representations for each node, such as node2vec, EGES, and LINE. These methods have been successfully applied in the industry (e.g., by Airbnb and Alibaba). However, while DeepWalk focuses on learning the interactive dependencies between nodes, it lacks a comprehensive theoretical framework for learning node attribute information, making it impossible to infer unknown nodes using existing models or data. Graph neural networks utilize the topological relationships between nodes and their neighbors to perform aggregation (convolution) operations. For example, GraphSAGE uses an aggregation function to continuously aggregate neighbor information based on the number of hops.

[0004] In the process of implementing the present invention, the applicant discovered that there are at least the following technical problems in the related art.

[0005] 1. Existing training methods based on homogeneous graphs are not directly applicable to search scenarios. In search scenarios, queries are required to recall Points of Interest (PoIs). However, queries and Points of Interest (PoIs) are heterogeneous and cannot be directly learned and trained as homogeneous graphs. In other words, it is difficult to learn vectors of two dimensions simultaneously.

[0006] 2. Methods such as EGES that integrate side information learn different features of the same Point of Interest and cannot be generalized to scenarios with heterogeneous nodes.

[0007] It can be seen that no effective solution has been proposed in the related art for the above-mentioned problems. Summary of the Invention

[0008] The embodiments of the present invention provide a search result recommendation method and device, an electronic device and a readable storage medium, so as to at least solve the technical problem that when there are few natural results corresponding to the query terms in the related technology, the search terms and points of interest recommended for supplementary recommendations are heterogeneous, resulting in the inability to recall more search results.

[0009] According to one aspect of an embodiment of the present invention, a search result recommendation method is provided, comprising: inputting a target search term into a recommendation model, wherein the recommendation model is pre-trained in a preset model through a training sample set; and displaying recommended points of interest output by the recommendation model in a preset application interface; wherein the training samples in the training sample set include search term features of the search term, first point of interest features corresponding to the search term, and second point of interest features corresponding to similar search terms, and the search terms and similar search terms all have the same points of interest.

[0010] Furthermore, before inputting the target search term into the recommendation model, it also includes: obtaining similar search terms corresponding to the search term based on user behavior records; constructing a training sample corresponding to the search term based on the search term, the first point of interest corresponding to the search term, and the second point of interest corresponding to the similar search term.

[0011] Furthermore, similar search terms corresponding to the search term are obtained based on user behavior records, including: when the search term includes multiple search intentions, splitting the search term into multiple sub-search terms, wherein each sub-search term corresponds to a search intention; and obtaining similar search terms corresponding to the multiple sub-search terms.

[0012] Furthermore, based on the search term, the first point of interest corresponding to the search term, and the second point of interest corresponding to the similar search term, a training sample corresponding to the search term is constructed, including: obtaining a first query graph corresponding to the search term, and obtaining a second query graph corresponding to the similar search term, wherein the first query graph includes the search term feature and multiple first point of interest features, and the second query graph includes the similar search term feature and the second point of interest feature; merging the first query graph and the second query graph to obtain a meta-path graph, wherein the meta-path graph includes the multiple first point of interest features and the multiple second point of interest features; sampling the point of interest features in the meta-path graph according to a preset meta-path wandering strategy to obtain at least one target point of interest feature; and constructing the training sample based on the at least one target point of interest feature and the search term feature.

[0013] Furthermore, merging the first query graph and the second query graph to obtain the training sample includes: establishing a complementary connection between the search term feature and the similar search term feature according to the user behavior record.

[0014] According to another aspect of an embodiment of the present invention, a search result recommendation device is also provided, including: an input unit for inputting a target search term into a recommendation model, wherein the recommendation model is pre-trained in a preset model through a training sample set; a display unit for displaying the recommended points of interest output by the recommendation model in a preset application interface; wherein the training samples in the training sample set include search term features of the search term, first point of interest features corresponding to the search term, and second point of interest features corresponding to similar search terms, and the search terms and similar search terms have the same points of interest.

[0015] Furthermore, the device also includes: an acquisition unit, used to obtain similar search terms corresponding to the search term based on user behavior records; a processing unit, used to construct a training sample corresponding to the search term based on the search term, the first point of interest corresponding to the search term, and the second point of interest corresponding to the similar search term.

[0016] Furthermore, the acquisition unit includes: a splitting module for splitting the search term into multiple sub-search terms when the search term includes multiple search intentions, wherein each sub-search term corresponds to a search intention; a first acquisition module for acquiring similar search terms corresponding to the multiple sub-search terms.

[0017] Furthermore, the processing unit includes: a second acquisition module, used to obtain a first query graph corresponding to the search term, and to obtain a second query graph corresponding to the similar search term, wherein the first query graph includes the search term feature and multiple first point of interest features, and the second query graph includes the similar search term feature and the second point of interest feature; a merging module, used to merge the first query graph and the second query graph to obtain a meta-path graph, wherein the meta-path graph includes the multiple first point of interest features and the multiple second point of interest features; a sampling module, used to sample point of interest features in the meta-path graph according to a preset meta-path wandering strategy to obtain at least one target point of interest feature; a processing module, used to construct the training sample based on the at least one target point of interest feature and the search term feature.

[0018] Furthermore, the merging module includes: a merging submodule, configured to establish a complementary connection between the search term feature and the similar search term feature according to the user behavior record.

[0019] According to another aspect of an embodiment of the present invention, an electronic device is also provided, including a processor, a memory, and a program or instruction stored in the memory and executable on the processor, wherein the program or instruction, when executed by the processor, implements the steps of the search result recommendation method described above.

[0020] According to another aspect of an embodiment of the present invention, a readable storage medium is further provided, on which a program or instruction is stored. When the program or instruction is executed by a processor, the steps of the search result recommendation method described above are implemented.

[0021] In an embodiment of the present invention, a target search term is input into a recommendation model, wherein the recommendation model is pre-trained in a preset model through a training sample set; recommended points of interest output by the recommendation model are displayed in a preset application interface; wherein the training samples in the training sample set include search term features of the search term, first point of interest features corresponding to the search term, and second point of interest features corresponding to similar search terms, and similar search terms have the same points of interest, thereby achieving the technical effect of expanding the recall results of the search term by supplementing the points of interest corresponding to the search term in the case where the points of interest corresponding to the search term are insufficient, thereby solving the technical problem that the search terms and points of interest recommended for the supplementary recommendations are heterogeneous when there are fewer natural results corresponding to the query term in the related technology, resulting in the inability to recall more search results. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of this application. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0023] Figure 1 is a flowchart of an optional search result recommendation method according to an embodiment of the present invention;

[0024] Figure 2 is a schematic diagram of splitting a search word according to an embodiment of the present invention;

[0025] Figure 3 is a schematic diagram of a node wandering strategy according to an embodiment of the present invention;

[0026] Figure 4 is a schematic diagram of a complementary connection according to an embodiment of the present invention;

[0027] Figure 5 This is a schematic structural diagram of an optional search result recommendation device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0028] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of the present invention.

[0029] It should be noted that the terms "first", "second", etc. in the description and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that the numbers used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0030] Example 1

[0031] According to an embodiment of the present invention, a search result recommendation method is provided. Figure 1 As shown, the method includes:

[0032] S102, inputting the target search term into a recommendation model, wherein the recommendation model is pre-trained in a preset model using a training sample set;

[0033] S104, displaying the recommended points of interest output by the recommendation model in a preset application interface;

[0034] The training samples in the training sample set include search term features of the search term, first point of interest features corresponding to the search term, and second point of interest features corresponding to similar search terms. Similar search terms all have the same point of interest.

[0035] In this embodiment, a user enters a target search term in the interface of a pre-set application. The pre-set application uploads the target search term to a local server or a cloud server, which then feeds the target search term into a recommendation model. The recommendation model is pre-trained using a set of training samples and is used to retrieve corresponding points of interest based on the user's input target search term. The points of interest retrieved by the recommendation model correspond to the target search term.

[0036] The Point of Interest (POI) described in this embodiment can be any object that can be queried by users on a network platform, such as geographical locations, scenic spots, services, products, etc. The search terms and target search terms described in this embodiment can be query terms entered by users through the search page and / or selected query conditions, or can also be query terms and / or query conditions generated by the platform based on user behavior.

[0037] In specific implementation, the query input for the current query can be obtained through the query entry of the platform or a preset application. Further, the platform or the preset application searches and queries according to the search terms of the obtained query input, and recalls at least one Point of Interest (POI) that matches the search terms, as well as the user behavior records corresponding to the at least one POI, such as user selection records. For example, in a food ordering application, when a user enters the search term "steak", the platform will recall multiple merchants selling steak, as well as the dishes of "steak" sold by each merchant. For each merchant, the platform will also recall the user behavior records corresponding to the search term "steak". Each merchant pair can be considered as a recommended POI.

[0038] For example, if the Point of Interest "La Niuniu Hot Pot Restaurant (Qingnian Road Branch)" exists in the Points of Interest recalled by both the search term "La Niuniu Hot Pot Restaurant" and the search term "Shanyangyang Hot Pot Restaurant", then the search terms "La Niuniu Hot Pot Restaurant" and "Shanyangyang Hot Pot Restaurant" are considered as similar search terms to each other.

[0039] In the specific implementation of this embodiment, it is first necessary to train the recommendation model.

[0040] In some embodiments of the present application, according to the user's query click data on the POI, a training sample set is constructed, and each training sample in this training sample set includes: the search term for the target ROI, the recommended ROI.

[0041] First, obtain the user behavior data stored in the network platform or the preset server. Usually, each query behavior of a user on the platform will generate a user behavior record, and the user behavior records stored in the platform or the server at least include the following information: POI identifier, search term, whether the POI is clicked, etc.

[0042] Then, construct training samples based on the user behavior records. In some embodiments of the present application, by processing the above user behavior records, candidate training samples can be obtained. Each candidate training sample includes information such as POI identifier, POI semantics, POI features, search term, and the POI selected by the user. In some embodiments of the present application, each training sample is represented as a multi-tuple <POI identifier, POI semantics, POI features, search term features, POI selected by the user>. Among them, there are multiple POIs included in this training sample.

[0043] Afterwards, the candidate samples are screened, and the candidate samples without click behavior are filtered out. The remaining candidate samples are used as training samples to form a training sample set.

[0044] In this embodiment, the training sample set includes a certain number of training samples, each of which includes a search term feature, a first point of interest feature corresponding to the search term, and a second point of interest feature corresponding to a similar search term. In this embodiment, the first point of interest is directly recalled by the search term, and the second point of interest is directly recalled by a similar search term. The first and second points of interest overlap, meaning that the search term and the similar search term both have the same point of interest.

[0045] Training samples are constructed based on the search term features, the search term and the points of interest features corresponding to similar search terms, and the points of interest recalled by the search term are expanded. The preset model is then trained until the model converges to obtain the corresponding recommendation model.

[0046] In this embodiment, because the search term and the recalled POIs are heterogeneous, we obtain the search term embedding, i.e., the search term feature; the first POI embedding, i.e., the first POI feature; and the second POI embedding, i.e., the second POI feature. Training samples are constructed based on the search term embedding, the first POI embedding, and the second POI embedding.

[0047] In this embodiment, the target search term is input into the recommendation model in the server to obtain the recommended interest points output by the recommendation model. The server then returns the corresponding recommended interest points to the preset application and displays the recommended interest nodes in the preset application interface of the preset application.

[0048] It should be noted that, through this embodiment, the target search term is input into the recommendation model, wherein the recommendation model is pre-trained in a preset model through a training sample set; the recommended points of interest output by the recommendation model are displayed in a preset application interface; wherein the training samples in the training sample set include search term features of the search term, first point of interest features corresponding to the search term, and second point of interest features corresponding to similar search terms, and similar search terms have the same points of interest, so that when the points of interest corresponding to the search term are insufficient, the points of interest of similar search terms that have the same points of interest as the search term are supplemented, thereby achieving the technical effect of expanding the recall results of the search term, and further solving the technical problem that when the natural results corresponding to the query term in the related technology are relatively few, the search terms and points of interest recommended for the remainder are heterogeneous, resulting in the inability to recall more search results.

[0049] Optionally, in this embodiment, before inputting the target search term into the recommendation model, it also includes but is not limited to: obtaining similar search terms corresponding to the search term based on user behavior records; constructing a training sample corresponding to the search term based on the search term, the first point of interest corresponding to the search term, and the second point of interest corresponding to the similar search term.

[0050] In a specific application scenario, before obtaining the recommended points of interest corresponding to the target search term through the recommendation model, similar search terms corresponding to the search term are obtained based on the user behavior record. For example, in the user behavior record, if the point of interest selected or clicked by the user in the process of searching for a certain search term is the point of interest of other search terms, then the search term corresponding to the selected point of interest is considered to be a similar search term corresponding to the search term entered by the user.

[0051] Then, a search term feature of the search term input by the user, a first point of interest feature of a first point of interest of the search term input by the user, and a second point of interest feature of a second point of interest similar to the search term are obtained to construct a training sample corresponding to the search term.

[0052] Optionally, in this embodiment, similar search terms corresponding to the search term are obtained based on the user behavior records, including but not limited to: when the search term includes multiple search intentions, splitting the search term into multiple sub-search terms, wherein each sub-search term corresponds to a search intention; obtaining similar search terms corresponding to the multiple sub-search terms.

[0053] In real-world applications, a large number of search queries have multiple meanings and encompass multiple intents. However, POIs provide single-value information. For example, a user searching for "hot pot" might click on "Chongqing hot pot" or "Old Beijing Copper Pot Shabu-Shabu Meat" on the platform. Therefore, in the POI-query structure, the query "hot pot" is connected to both the POI "Chongqing hot pot" and the POI "Old Beijing hot pot." The goal of the recommendation model is to minimize the cosine distance between adjacent vectors, so the "hot pot" vector must be similar to both the "Chongqing hot pot" vector and the "Old Beijing hot pot" vector. This results in the query vector failing to learn very accurate information and overgeneralizing.

[0054] Therefore, in the embodiments of this application, for multi-intent queries, the query is hierarchically split based on the intent information, splitting it into N sub-dimensions, each corresponding to a sub-search term query. This way, the sub-queries in each dimension are single-intent and easier to learn together with POIs. When constructing the POI-query, the sub-queries in each dimension are associated with the POIs in their respective dimensions. When sampling training samples, the queries are also randomly stratified and sampled, and only POIs corresponding to single-dimensional queries are collected as positive samples.

[0055] In one example, Figure 2 As shown in the figure, the query corresponds to multiple intents, each intent corresponds to a POI, and the query is split into query1, query2, and query3, and each sub-query corresponds to a POI.

[0056] Optionally, in this embodiment, a training sample corresponding to the search term is constructed based on the search term, the first point of interest corresponding to the search term, and the second point of interest corresponding to the similar search term, including but not limited to: obtaining a first query graph corresponding to the search term, and obtaining a second query graph corresponding to the similar search term, wherein the first query graph includes a search term feature and multiple first point of interest features, and the second query graph includes similar search term features and second point of interest features; merging the first query graph and the second query graph to obtain a meta-path graph, wherein the meta-path graph includes multiple first point of interest features and multiple second point of interest features; sampling the point of interest features in the meta-path graph according to a preset meta-path wandering strategy to obtain at least one target point of interest feature; and constructing a training sample based on at least one target point of interest feature and the search term feature.

[0057] Specifically, in this embodiment, a first query graph of a search term and a second query graph of a similar search term are obtained. The query graph of the search term is composed of the query term and the point of interest POI corresponding to the query term, wherein the first query graph includes the search term feature and multiple first point of interest features, and the second query graph includes similar search term features and second point of interest features. In the case where there is the same POI between two search terms, the first query graph and the second query graph are merged to obtain the original path graph, and the meta-path graph includes the search term feature, similar search term feature, multiple first point of interest features and multiple second point of interest features. Then, according to the preset meta-path wandering strategy, wandering sampling is performed in the meta-path graph to obtain at least one target point of interest feature, and then a training sample is constructed based on at least one target point of interest feature, the search term feature and the user behavior record.

[0058] In specific application scenarios, GraphSAGE directly aggregates central nodes based on neighbor nodes, reflecting the structural nature of the model. However, for Poi's complementary recommendation business, homogeneity is more important than structure. In an example, Figure 3As shown in the figure, for the node query McDonald's, GraphSAGE makes the embedding of McDonald's similar to the embeddings of Poi A, Poi B, Poi C, and Poi D based on the structural nature of the model. However, for Poi recommendation, Poi F and Poi I should be closer to the embedding of McDonald's (making the embedding of McDonald's more similar to the embeddings of Pizza Hut-related stores).

[0059] To address this issue, this embodiment introduces parameters p and q. Parameter p controls the probability of fasting to visit a vertex that has just been visited, while q controls the tendency to wander. If q > 1, the wandering tendency model is structural; if q < 1, the wandering tendency model is homogeneous. The value of q is appropriately increased during the complementary recommendation process. This allows the method to better adhere to the homogeneity of the learning network compared to methods such as GraphSAGE, meeting the actual needs of the complementary recommendation business.

[0060] In the process of constructing the meta-path graph, in this embodiment, the meta-path graph g is first constructed based on the Query-Poi pair of the natural result, and then the meta-path graph g is improved based on the Query-Poi of the complementary recommendation. If a certain Poi appears in both the natural result Query A and the complementary result Query B, an entity connection is added between Query A and Query B. By adding this entity connection, the Poi information generated by users based on the complementary recommendation module can be sampled during the node's wandering process, thereby improving the generalization ability of the node.

[0061] Optionally, in this embodiment, merging the first query graph and the second query graph to obtain training samples includes but is not limited to: establishing complementary connections between search term features and similar search term features according to user behavior records.

[0062] In one example, Figure 4 As shown, McDonald's Huajiadi store has a connection with McDonald's (Query) in the natural results and an entity relationship with Pizza Hut (Query) in the supplementary recommendations. Therefore, a supplementary connection is added between McDonald's (Query) and Pizza Hut (Query). In GraphSAGE, the embedding of McDonald's (Query) is similar to "McDonald's Guangshun North Street Store," "McDonald's Wangjing Store," and "McDonald's Huajiadi Store." Based on the supplementary recommendations, the embedding of McDonald's (Query) learns information related to the supplementary recommendations, bringing it closer to POIs such as "Pizza Hut Zhonghui Road," thereby improving the generalization ability of the query node.

[0063] As a preferred solution in this embodiment, it also includes: determining negative samples corresponding to the training samples, and adding the negative samples to the training sample set, wherein the similarities between the negative samples and the training samples include at least one of the following: category characteristics of the point of interest, feature distance of the point of interest, and text of the point of interest.

[0064] In a specific application scenario, taking the preset model as the Pinsage model as an example, in the Pinsage model, a HardExample is defined as a node that is similar to q to some extent, but is irrelevant to i, given a positive sample training pair (q, i). In the article, the selection of Hard Example is calculated based on the personalized PageRank score of q. In the complementary recommendation business, since it contains relatively rich attribute information, Hard Example cannot be limited to the topological relationship between nodes (personalized PageRank score). In the recommendation of points of interest, this embodiment redefines the existing Hard Example samples based on category characteristics: looking for POIs that are relatively close to node q in category characteristics, distance, and text as negative samples.

[0065] As a preferred solution in this embodiment, it also includes: when the search term is longer than a preset text length, determining a short text search term corresponding to the search term; and establishing a complementary connection between the acquired search term feature and the short text search term feature.

[0066] In specific application scenarios, search terms with text lengths exceeding a preset text length are considered long-tail queries. When a user enters a long-tail query, the system determines the corresponding short-text search term and rewrites the long-tail query into a short-text query, recalling the embeddings of POIs related to the short-text query. During the graphing process, the long-tail query and the rewritten short-text query are connected, thereby improving the generalization capability of the long-tail query node.

[0067] Through this embodiment, the target search term is input into the recommendation model, wherein the recommendation model is pre-trained in a preset model through a training sample set; the recommended points of interest output by the recommendation model are displayed in a preset application interface; wherein the training samples in the training sample set include search term features of the search term, first point of interest features corresponding to the search term, and second point of interest features corresponding to similar search terms, and similar search terms have the same points of interest, so that when the points of interest corresponding to the search term are insufficient, the points of interest of similar search terms that have the same points of interest as the search term are supplemented, thereby achieving the technical effect of expanding the recall results of the search term, and further solving the technical problem that when the natural results corresponding to the query term in the related technology are relatively few, the search terms and points of interest recommended for the remainder are heterogeneous, resulting in the inability to recall more search results.

[0068] It should be noted that for the aforementioned method embodiments, for simplicity of description, they are all expressed as a series of action combinations. However, those skilled in the art should be aware that the present invention is not limited by the order of the actions described, because according to the present invention, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in this specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present invention.

[0069] Through the description of the above embodiments, those skilled in the art can clearly understand that the method according to the above embodiment can be implemented by means of software plus the necessary general hardware platform, and of course it can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes a number of instructions for enabling a terminal device (which can be a mobile phone, computer, server, or network device, etc.) to execute the methods described in each embodiment of the present invention.

[0070] Example 2

[0071] According to an embodiment of the present invention, a search result recommendation device for implementing the above search result recommendation method is also provided. Figure 5 As shown, the device includes:

[0072] 1) An input unit 50, configured to input a target search term into a recommendation model, wherein the recommendation model is pre-trained in a preset model using a training sample set;

[0073] 2) a display unit 52, configured to display the recommended points of interest output by the recommendation model in a preset application interface;

[0074] The training samples in the training sample set include search term features of the search term, first point of interest features corresponding to the search term, and second point of interest features corresponding to similar search terms, and the search terms and similar search terms all have the same point of interest.

[0075] Optionally, in this embodiment, the device further includes:

[0076] 1) an acquisition unit, configured to acquire similar search terms corresponding to the search term based on user behavior records;

[0077] 2) A processing unit, configured to construct a training sample corresponding to the search term based on the search term, the first point of interest corresponding to the search term, and the second point of interest corresponding to the similar search term.

[0078] Optionally, in this embodiment, the acquiring unit includes:

[0079] 1) a splitting module, configured to split the search term into multiple sub-search terms when the search term includes multiple search intents, wherein each sub-search term corresponds to a search intent;

[0080] 2) A first acquisition module is used to acquire similar search terms corresponding to the multiple sub-search terms.

[0081] Optionally, in this embodiment, the processing unit includes:

[0082] 1) A second acquisition module, configured to acquire a first query graph corresponding to the search term, and a second query graph corresponding to the similar search term, wherein the first query graph includes the search term feature and a plurality of first POI features, and the second query graph includes the similar search term feature and the second POI feature;

[0083] 2) a merging module, configured to merge the first query graph and the second query graph to obtain a meta-path graph, wherein the meta-path graph includes the plurality of first point-of-interest features and the plurality of second point-of-interest features;

[0084] 3) a sampling module, configured to sample features of points of interest in the meta-path graph according to a preset meta-path walking strategy to obtain features of at least one target point of interest;

[0085] 4) A processing module, configured to construct the training sample according to the at least one target interest point feature and the search term feature.

[0086] Optionally, in this embodiment, the merging module includes:

[0087] 1) A merging submodule, configured to establish a complementary connection between the search term feature and the similar search term feature according to the user behavior record.

[0088] The search result recommendation device proposed in this embodiment achieves the technical effect of expanding the recall results of the search term by supplementing the interest points of similar search terms that have the same interest points as the search term when the interest points corresponding to the search term are insufficient, thereby solving the technical problem that in the related technology, when there are few natural results corresponding to the query term, the supplementary recommended search terms and interest points are heterogeneous, resulting in the inability to recall more search results.

[0089] Example 3

[0090] According to an embodiment of the present invention, an electronic device is also provided, including a processor, a memory, and a program or instruction stored in the memory and executable on the processor. When the program or instruction is executed by the processor, the steps of the search result recommendation method described in Example 1 are implemented.

[0091] Optionally, in this embodiment, the memory is configured to store program codes for executing the following steps:

[0092] S1, inputting the target search term into the recommendation model, wherein the recommendation model is pre-trained in a preset model using a training sample set;

[0093] S2, displaying the recommended points of interest output by the recommendation model in a preset application interface;

[0094] The training samples in the training sample set include search term features of the search term, first point of interest features corresponding to the search term, and second point of interest features corresponding to similar search terms, and the search terms and similar search terms all have the same point of interest.

[0095] Optionally, the specific examples in this embodiment may refer to the examples described in the above embodiment 1, and this embodiment will not be described in detail here.

[0096] Example 4

[0097] An embodiment of the present invention further provides a readable storage medium, on which a program or instruction is stored. When the program or instruction is executed by a processor, the steps of the search result recommendation method as described in Example 1 are implemented.

[0098] Optionally, in this embodiment, the storage medium is configured to store program codes for executing the following steps:

[0099] S1, inputting the target search term into the recommendation model, wherein the recommendation model is pre-trained in a preset model using a training sample set;

[0100] S2, displaying the recommended points of interest output by the recommendation model in a preset application interface;

[0101] The training samples in the training sample set include search term features of the search term, first point of interest features corresponding to the search term, and second point of interest features corresponding to similar search terms, and the search terms and similar search terms all have the same point of interest.

[0102] Optionally, the storage medium is further configured to store program codes for executing the steps included in the method in the above embodiment 1, which will not be described in detail in this embodiment.

[0103] Optionally, in this embodiment, the above-mentioned storage medium may include but is not limited to: a USB flash drive, a read-only memory (ROM), a random access memory (RAM), a mobile hard disk, a magnetic disk or an optical disk, and other media that can store program codes.

[0104] Optionally, the specific examples in this embodiment may refer to the examples described in the above embodiment 1, and this embodiment will not be described in detail here.

[0105] The serial numbers of the above embodiments of the present invention are for description only and do not represent the advantages or disadvantages of the embodiments.

[0106] If the integrated units in the above embodiments are implemented in the form of software functional units and sold or used as independent products, they can be stored in the above-mentioned computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the existing technology, or all or part 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 a number of instructions for causing one or more computer devices (such as personal computers, servers, or network devices) to execute all or part of the steps of the methods described in various embodiments of the present invention.

[0107] In the above embodiments of the present invention, the description of each embodiment has its own focus. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0108] In the several embodiments provided in this application, it should be understood that the disclosed client can be implemented in other ways. Among them, the device embodiments described above are merely illustrative. For example, the division of the units is merely a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, and can be electrical or other forms.

[0109] The units described as separate components may or may not be physically separate, and 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 these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0110] In addition, the functional units in the various embodiments of the present invention may be integrated into a single processing unit, each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0111] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as within the scope of protection of the present invention.

Claims

1. A search result recommendation method, characterized in that: include: Inputting the target search term into the recommendation model, wherein the recommendation model is pre-trained in a preset model using a training sample set; Displaying the recommended points of interest output by the recommendation model in a preset application interface, wherein the search terms and the recalled points of interest are heterogeneous, and the points of interest (POIs) are any of geographical locations, scenic spots, and services; The training samples in the training sample set include search term features of a search term, first point of interest features corresponding to the search term, and second point of interest features corresponding to similar search terms, wherein the search terms and the similar search terms all have the same point of interest. Before inputting the target search term into the recommendation model, the training sample set further includes: Obtain similar search terms corresponding to the search terms based on user behavior records; constructing a training sample corresponding to the search term based on the search term, the first point of interest corresponding to the search term, and the second point of interest corresponding to the similar search term; Constructing a training sample corresponding to the search term according to the search term, the first interest point corresponding to the search term, and the second interest point corresponding to the similar search term, including: Obtaining a first query graph corresponding to the search term, and obtaining a second query graph corresponding to the similar search term, wherein the first query graph includes the search term feature and a plurality of first point of interest features, and the second query graph includes the similar search term feature and the second point of interest feature; Merging the first query graph and the second query graph to obtain a meta-path graph, wherein the meta-path graph includes a plurality of first point-of-interest features and a plurality of second point-of-interest features; Sampling the features of the points of interest in the meta-path graph according to a preset meta-path walking strategy to obtain at least one target point of interest feature; The training sample is constructed according to the at least one target interest point feature and the search term feature.

2. The method according to claim 1, characterized in that Obtaining similar search terms corresponding to the search term based on user behavior records, including: In the case where the search term includes multiple search intents, splitting the search term into multiple sub-search terms, wherein each sub-search term corresponds to one search intent; Obtain similar search terms corresponding to the multiple sub-search terms.

3. The method according to claim 1, characterized in that Merging the first query graph and the second query graph to obtain the training sample includes: A complementary connection is established between the search term feature and the similar search term feature according to the user behavior record.

4. A search result recommendation device, characterized in that: include: An input unit, configured to input a target search term into a recommendation model, wherein the recommendation model is pre-trained in a preset model using a training sample set; A display unit, configured to display the recommended points of interest output by the recommendation model in a preset application interface, wherein the search terms and the recalled points of interest are heterogeneous, and the points of interest (POIs) are any of geographical locations, scenic spots, and services; The training samples in the training sample set include a search term feature of a search term, a first point of interest feature corresponding to the search term, and a second point of interest feature corresponding to a similar search term, and the search term and the similar search term both have the same point of interest; The device further comprises: An acquisition unit, configured to acquire similar search terms corresponding to the search terms based on user behavior records; a processing unit, configured to construct a training sample corresponding to the search term based on the search term, the first point of interest corresponding to the search term, and the second point of interest corresponding to the similar search term; The processing unit includes: a second acquisition module, configured to acquire a first query graph corresponding to the search term, and acquire a second query graph corresponding to the similar search term, wherein the first query graph includes the search term feature and a plurality of first POI features, and the second query graph includes the similar search term feature and the second POI feature; a merging module, configured to merge the first query graph and the second query graph to obtain a meta-path graph, wherein the meta-path graph includes a plurality of first point-of-interest features and a plurality of second point-of-interest features; A sampling module, configured to sample features of points of interest in the meta-path graph according to a preset meta-path walking strategy to obtain features of at least one target point of interest; A processing module is used to construct the training sample according to the at least one target interest point feature and the search term feature.

5. The device according to claim 4, characterized in that The acquisition unit includes: a splitting module, configured to split the search term into multiple sub-search terms when the search term includes multiple search intents, wherein each sub-search term corresponds to one search intent; The first acquisition module is configured to acquire similar search terms corresponding to the multiple sub-search terms.

6. The device according to claim 4, characterized in that The merging module includes: The merging submodule is used to establish a complementary connection between the search term feature and the similar search term feature according to the user behavior record.

7. An electronic device, characterized in that: The method comprises a processor, a memory, and a program or instruction stored in the memory and executable on the processor. When the program or instruction is executed by the processor, the steps of the search result recommendation method according to claims 1 to 3 are implemented.

8. A readable storage medium, characterized in that: The readable storage medium stores a program or instruction, and when the program or instruction is executed by the processor, the steps of the search result recommendation method according to claims 1-3 are implemented.

Images