Palmtop bank personalized search recommendation method and system, electronic equipment and storage medium

Abstract

The invention provides a handheld bank personalized search recommendation method and system, electronic equipment and a storage medium, and is suitable for a handheld bank APP to obtain a pre-constructed collaborative knowledge graph; obtaining a historical user query, and extracting a historical query suggestion set related to the historical user query from the collaborative knowledge graph; forming a plurality of training triads according to each user in the user set, the historical user query and a historical query suggestion set related to the historical user query, and training a to-be-trained personalized search recommendation model according to a scoring function learned by a training set constructed by each training triad to obtain the personalized search recommendation model; and when a query currently input by a target user in the search session of the handheld bank APP is detected, calculating the recommendation degree between the query and each entity in the handheld bank APP by using the personalized search recommendation model, screening out to-be-recommended entities from the entities according to the recommendation degree of each entity, and outputting each to-be-recommended entity.

Description

Technical Field

[0001] The present invention relates to the field of financial technology, and more specifically to a method, system, electronic device and storage medium for personalized search and recommendation of mobile banking. Background Art

[0002] With the rapid development of mobile Internet, mobile banking apps have become the main tool for people to manage their finances. The search and recommendation system of mobile banking can provide relevant search results based on the query information entered by users, making it easier for users to find the service information they need, thereby reducing the cost of manual business communication. At the same time, it can also recommend products or services that users may be interested in based on their personal preferences, thereby bringing traffic to other related products of the bank. However, affected by many factors such as the user's cognitive level and personal habits, the queries submitted by users are short and irregular, making it difficult for the search and recommendation system to infer their real needs and intentions from the user's short keywords, resulting in low efficiency and accuracy in obtaining recommendation results, leading to poor user experience.

[0003] Therefore, how to provide a search recommendation method for mobile banking to improve the efficiency and accuracy of search recommendations, better meet user needs, and improve user experience is a problem that urgently needs to be solved in this application. Summary of the invention

[0004] In view of this, the present invention provides a method, system, electronic device and storage medium for personalized search and recommendation of mobile banking, aiming to improve search and recommendation efficiency and accuracy, better meet user needs and improve user experience.

[0005] The first aspect of the present application provides a mobile banking personalized search and recommendation method, which is applicable to a mobile banking APP, and the method includes:

[0006] Obtain a pre-built collaborative knowledge graph, wherein the collaborative knowledge graph is composed of a user-item interaction graph and a knowledge graph, wherein the user-item interaction graph is composed of a user set, an item set, and a connection relationship between each user and each item; the collaborative knowledge graph includes a plurality of triples, each of which describes a relationship from a head entity to a tail entity, and each entity in the collaborative knowledge graph is aligned with each item;

[0007] Obtaining historical user queries, and extracting a set of historical query suggestions related to the historical user queries from the collaborative knowledge graph;

[0008] According to each user in the user set, the historical user query and a set of historical query suggestions related to the historical user query, a plurality of training triples are formed, and the personalized search recommendation model to be trained is trained according to the scoring function learned from the training set constructed by each of the training triples to obtain a personalized search recommendation model;

[0009] When a query currently input by the target user in the search session of the mobile banking APP is detected, the personalized search recommendation model is used to calculate the recommendation degree between the query and each entity in the mobile banking APP, and based on the recommendation degree of each displayed entity, the entities to be recommended are screened out from each of the entities, and each of the entities to be recommended is output.

[0010] Optionally, the acquiring of historical user queries and extracting a set of historical query suggestions related to the historical user queries from the collaborative knowledge graph includes:

[0011] Filtering a set of candidate entities related to the historical user query from each entity in the collaborative knowledge graph; wherein the set of candidate entities includes multiple candidate entities related to the historical user query;

[0012] For each of the candidate entities, multiple candidate related entities related to the candidate entity are obtained, and a candidate related entity set is constructed according to the multiple candidate related entities related to each of the candidate entities;

[0013] The candidate entity set and the candidate related entity set are merged to obtain a historical query suggestion set related to the historical user query.

[0014] Optionally, the step of filtering out a set of candidate entities related to the historical user query from the entities in the collaborative knowledge graph includes:

[0015] A graph mining algorithm based on a random walk model calculates the minimum walk time for each entity in the collaborative knowledge graph to reach the historical user query;

[0016] Determine a plurality of candidate entities from each of the entities according to the minimum travel time between the historical user query and each of the entities;

[0017] Calculating the correlation between the historical user query and each of the entities, and selecting a plurality of candidate entities from each of the entities according to the correlation between the historical user query and each of the entities;

[0018] Generate a corresponding candidate entity set according to each of the candidate entities.

[0019] Optionally, for each candidate entity, obtaining multiple candidate related entities related to the candidate entity, and constructing a candidate related entity set according to the multiple candidate related entities related to each candidate entity, includes:

[0020] For each candidate entity, multiple candidate related entities related to the candidate entity are obtained; wherein the multiple candidate related entities are the previous entities that co-occur multiple times with the candidate entity in the same search session and have the highest number of co-occurrences;

[0021] The relevance between the historical user query and each of the candidate related entities is calculated, and a candidate related entity set is screened out from the candidate related entities according to the relevance of each of the candidate related entities.

[0022] Optionally, forming a plurality of training triples according to each user in the user set, the historical user queries, and a set of historical query suggestions related to the historical user queries, and training the personalized search recommendation model to be trained according to a scoring function learned from a training set constructed from each of the training triples to obtain a personalized search recommendation model, including:

[0023] According to each user in the user set, the historical user query and a set of historical query suggestions related to the historical user query, a plurality of training triples are formed, wherein each of the training triples is composed of the user, the historical user query and a recommended entity; the set of historical query suggestions includes a plurality of recommended entities;

[0024] For each of the training triples, obtaining the total number of clicks issued by the users in the training triples to the recommended entity when searching for the historical user query, and calculating the interest level of the training triples according to the users, the historical user queries, the recommended entities and the corresponding total number of clicks in the training triples;

[0025] A corresponding training set is constructed according to each of the training triples and their interest levels, and a personalized search recommendation model to be trained is trained based on the scoring function learned from the training set constructed from each of the training triples, the relevance of each candidate entity in the candidate entity set to the historical user query, the relevance of each candidate related entity in the candidate related entity set to the historical user query, and the interest level of each recommended entity to the historical user query to obtain a personalized search recommendation model.

[0026] Optionally, after obtaining the personalized search recommendation model, the method further includes:

[0027] Verifying the personalized search recommendation model;

[0028] If the verification of the personalized search recommendation model passes, the step of using the personalized search recommendation model to calculate the recommendation degree between the query and each entity in the mobile banking APP when detecting the query currently entered by the target user in the search session of the mobile banking APP, and filtering out the entities to be recommended from each of the entities according to the recommendation degree of each of the entities, and outputting each of the entities to be recommended is executed.

[0029] A second aspect of the present application provides a mobile banking personalized search and recommendation system, which is applicable to a mobile banking APP, and the system includes:

[0030] A first acquisition unit is used to acquire a pre-constructed collaborative knowledge graph, wherein the collaborative knowledge graph is composed of a user-item binary interaction graph and a knowledge graph, and the user-item binary interaction graph is composed of a user set, an item set, and a connection relationship between each user and each item; the collaborative knowledge graph includes a plurality of triples, each of which describes a relationship from a head entity to a tail entity, and each entity in the collaborative knowledge graph is aligned with each item;

[0031] An extraction unit, configured to obtain historical user queries and extract a set of historical query suggestions related to the historical user queries from the collaborative knowledge graph;

[0032] A training unit, configured to form a plurality of training triples according to each user in the user set, the historical user queries, and a set of historical query suggestions related to the historical user queries, and to train a personalized search recommendation model to be trained according to a scoring function learned from a training set constructed from each of the training triples, so as to obtain a personalized search recommendation model;

[0033] The recommendation unit is used to, when detecting a query currently input by a target user in the search session of the mobile banking APP, calculate the recommendation degree between the query and each entity in the mobile banking APP using the personalized search recommendation model, filter out entities to be recommended from each of the entities according to the recommendation degree of each of the entities, and output each of the entities to be recommended.

[0034] Optionally, the extraction unit includes:

[0035] A screening unit, configured to screen out a candidate historical query suggestion set related to the historical user query from each entity in the collaborative knowledge graph; wherein the historical query suggestion set includes a plurality of candidate entities related to the historical user query;

[0036] A candidate related historical query suggestion set building unit is used to obtain, for each candidate entity, multiple candidate related entities related to the candidate entity, and build a candidate related entity set according to the multiple candidate related entities related to each candidate entity;

[0037] A merging unit is used to merge the candidate entity set with the candidate related entity set to obtain a historical query suggestion set related to the historical user query.

[0038] The third aspect of the present application provides an electronic device, comprising: a processor and a memory, wherein the processor and the memory are connected via a communication bus; wherein the processor is used to call and execute a program stored in the memory; and the memory is used to store a program, wherein the program is used to implement the personalized search and recommendation method for mobile banking provided in the first aspect of the present application.

[0039] A fourth aspect of the present application provides a computer-readable storage medium, in which computer-executable instructions are stored, and the computer-executable instructions are used to execute the mobile banking personalized search and recommendation method provided in the first aspect of the present application.

[0040] The present invention provides a method, system, electronic device and storage medium for personalized search recommendation on mobile banking, which obtains a pre-built collaborative knowledge graph, wherein the collaborative knowledge graph is composed of a user-item binary interaction graph and a knowledge graph, and the user-item binary interaction graph is composed of a user set, an item set and a connection relationship between each user and each item; the collaborative knowledge graph includes multiple triples, each triple describes a relationship from a head entity to a tail entity, and each entity in the collaborative knowledge graph is aligned with each item; historical user queries are obtained, and a historical query suggestion set related to the historical user queries is extracted from the collaborative knowledge graph Combine; form multiple training triples according to each user in the user set, historical user queries and historical query suggestion sets related to historical user queries, and train the personalized search recommendation model to be trained according to the scoring function learned from the training set constructed by each training triple, to obtain a personalized search recommendation model; when the query currently entered by the target user in the search session of the mobile banking APP is detected, the personalized search recommendation model is used to calculate the recommendation degree between the query and each entity in the mobile banking APP, and the entities to be recommended are screened from each entity according to the recommendation degree of each entity, and each entity to be recommended is output. The technical solution provided by the present application can effectively mine the relationship between more different entities by combining the pre-constructed knowledge graph and the user-item interaction graph constructed by using the search log as the data source, effectively solving the problem of data sparsity in the prior art, and making the query results more relevant and coherent. In addition, the knowledge graph is built on the semantic model, which can better understand the semantic associations between entities, thereby better understanding the user's query intentions and providing more accurate information for subsequent search recommendations; combining features of two different granularities, relevance and interest, to model user preferences for entities improves the personalization of query results, that is, using the two features of relevance and interest to construct a corresponding training set, so that the personalized search recommendation model trained using the constructed training set can more efficiently and accurately search for entities that meet the needs of target users. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0042] Figure 1 An example diagram of a search recommendation method provided by the prior art;

[0043] Figure 2A flowchart of a personalized search and recommendation method for mobile banking provided in an embodiment of the present application;

[0044] Figure 3 An example diagram of a personalized search and recommendation method for mobile banking provided in an embodiment of the present application;

[0045] Figure 4 A schematic diagram of the structure of a mobile banking personalized search and recommendation system provided by an embodiment of the present invention;

[0046] Figure 5 A schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0047] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments 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 creative work are within the scope of protection of the present invention.

[0048] In this application, relational terms such as first and second, etc. are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "comprise", "include" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the statement "comprises a ..." does not exclude the presence of other identical elements in the process, method, article or device including the element.

[0049] In order to better understand this application, the technical names involved in this application are explained below:

[0050] Search recommendation: Search recommendation is a search term understanding technology that enables interaction between users and search engines. Its main idea is to recommend search terms related to the query terms submitted by users to users by calculating the similarity of the query terms submitted by users. Its core task is to help users construct effective queries and thereby accurately describe their information needs, thereby providing users with higher quality search services.

[0051] Graph theory: Graph theory is a field in mathematics and computer science that studies graphs, an abstract structure. A graph consists of vertices (nodes) and edges (lines connecting nodes). Vertices refer to individuals in the graph and can represent entities or abstract concepts, while edges are the relationships connecting vertices. Graphs can be directed or undirected. If it is a directed graph, the edges have directionality. A graph can be represented by an adjacency matrix, which is a two-dimensional array that represents the connection relationship between vertices.

[0052] Random walk: Random walk is a model in graph theory that describes the process of starting from a node in the graph and randomly selecting adjacent nodes for movement according to certain rules. The common rule is to select adjacent nodes as the next moving position with equal probability based on the structure of the graph until the stopping condition is reached (such as a specific number of time steps, reaching a specific node, or other preset conditions). For an undirected graph, a node is randomly selected from the neighboring nodes of the current node as the next node to move. For a directed graph, the next node is selected based on the direction of the edge. Based on the random walk model, it is possible to simulate the behavior of users randomly browsing web pages on the Internet to predict the content that users may be interested in.

[0053] Knowledge graph: Knowledge graph is a technology that organizes information in a graphical structure. It aims to describe the relationship between things in the real world, store this relationship and information in a huge database in a way that computers can understand, and enable computers to understand, infer and help people get the information they want based on these relationships and information. This technology can help computers process information more intelligently, such as providing more accurate results in search engines or answering more complex questions in smart assistants. It is similar to a huge network, containing a variety of entities (such as people, places, objects, concepts, etc.) and the associations and attributes between them. These entities and relationships are represented in the form of a graph, consisting of nodes and edges. Nodes represent entities, and edges represent the relationship between entities. The construction of a knowledge graph usually involves the integration of multiple information sources, including various information from structured data to unstructured data, such as databases, network content, text documents, images, etc., connecting this information and representing it in a way that computers can understand, thereby helping computers better understand the meaning, context and associations of information.

[0054] From the above background technology, it can be seen that the traditional recommendation search method is affected by many factors such as user cognitive level and personal habits. The queries submitted by users are short and irregular, making it difficult for the search recommendation system to infer their real needs and intentions from their short keywords, resulting in low efficiency and accuracy of recommendation results, leading to poor user experience.

[0055] In the prior art, some incomplete or non-compliant log information can be removed by preprocessing the original log files. Generate a query click graph, which consists of the query (query) input by the user, the click web page (URL) corresponding to the query, and the connecting edge between the two. The weight on the edge is the click frequency. Perform non-connected subgraph detection and graph merging operations on the query click graph. Input the given original query, use the random walk algorithm to generate a candidate query suggestion list: sort the candidate query suggestion list according to the relevance of each query, give a threshold, and take the previous query as the query suggestion for the original query, such as Figure 1 shown.

[0056] However, from Figure 1 As can be seen from the figure, the search recommendation method of the query-URL graph is implemented based on query click records. However, search engines serve many users, and a single user will often only search a small part of the queries and click on a very small part of the recommended results. This will inevitably lead to serious data sparsity in the click data of most users, thus affecting the training effect of the model and the search efficiency and accuracy of the final model. In addition, the query click graph ignores the user information in the query log, making it difficult to implement search recommendations based on user interests, resulting in a poor user experience.

[0057] Therefore, the present application provides a method, system, electronic device and storage medium for personalized search recommendation of mobile banking. By combining the pre-built knowledge graph and the user-item interaction graph built by using the search log as the data source, it can effectively mine the relationship between more different entities, effectively solve the problem of sparse data rows in the prior art, and make the query results more relevant and coherent. In addition, the knowledge graph is built on the semantic model, which can better understand the semantic association between entities, so as to better understand the intention of the user's query and provide more accurate information for subsequent search recommendations; the features of two different granularities, relevance and interest, are combined to model the user's preference for entities, which improves the personalization of the query results, that is, the corresponding training set is constructed using the two features of relevance and interest, so that the personalized search recommendation model trained by using the constructed training set can more efficiently and accurately search for various entities that meet the needs of the target user.

[0058] See also Figure 2 , shows a flow chart of a method for personalized search and recommendation of mobile banking provided in an embodiment of the present application, which is applicable to a mobile banking APP, and specifically comprises the following steps:

[0059] S201: Obtain a pre-built collaborative knowledge graph, wherein the collaborative knowledge graph is composed of a user-item binary interaction graph and a knowledge graph, wherein the user-item binary interaction graph is composed of a user set, an item set, and a connection relationship between each user and each item; the knowledge graph includes multiple triples, each triple describes a relationship from a head entity to a tail entity, and each entity in the knowledge graph is aligned with each item.

[0060] In the embodiment of the present application, historical user-item interaction record data (search log) can be obtained, and a corresponding user-item interaction graph can be generated according to the historical user-item interaction record data, as shown in formula (1). The user-item binary interaction graph is composed of a user set, an item set, and a connection relationship between each user and each item. The connection relationship is that there is an interaction between the user and the item, or there is no interaction between the user and the item. The historical user-item interaction record data includes multiple historical user-item interaction records. For example, the interaction in the historical user-item interaction record can be purchase, click, etc., which is not limited in the embodiment of the present application.

[0061] g 1 :{(u,y ui ,i)|u∈U,i∈I}(1)

[0062] Among them, U and I represent the user set and item set respectively, u represents the user, i represents the item, and y ui represents the connection relationship between user u and item i; if the connection relationship y ui is equal to 1, it means that there is an existing interaction between user u and item i, otherwise y ui is equal to 0.

[0063] In the embodiment of the present application, in addition to the historical user-item interaction record data, auxiliary information related to the item can be further obtained, such as item attributes and external knowledge. Generally, such auxiliary data consists of real-world entities and the relationships between them. An item can be described by auxiliary information related to the item. For example, a financial product (item) can be described by its service industry, product form, and return risk (auxiliary information). Therefore, by obtaining the auxiliary information of the item and using the auxiliary information of the item to construct a corresponding knowledge graph, the side information can be organized in the form of a knowledge graph.

[0064] It should be noted that the knowledge graph is a directed graph consisting of subject-attribute-object triple facts. Each triple describes the relationship r from the head entity h to the tail entity t, and R contains two relationships in the positive direction and the negative direction. In addition, a set of item-entity alignments A:{(i,e)|i∈I,e∈ε} can be defined, where (i,e) means that item i can be aligned with entity e in the knowledge graph. It can be seen that the corresponding knowledge graph g can be constructed by defining the alignment relationship between items and entities and obtaining auxiliary information related to items. 2 :{(h,r,t)|h,t∈ε,r∈R}.

[0065] In the embodiment of the present application, after constructing the user-item binary interaction graph and the knowledge graph, the user-item binary interaction graph and the knowledge graph can be used to encode user behavior and item information into a unified relationship graph. Specifically, the user-item binary interaction graph can be aligned and integrated with the item-entity in the knowledge graph to obtain the corresponding collaborative knowledge graph, as shown in formula (2).

[0066] g={(h,r,t)|h,t∈ε′,r∈R′}(2)

[0067] Among them, ε′=ε∪U, R′=R∪{Interact}, Interact={y ui |u∈U,i∈I}. As can be seen from formula (2), the collaborative knowledge graph also includes multiple triples, each of which also describes the relationship r from the head entity h to the tail entity t. Unlike the knowledge graph, the range from the head entity h to the tail entity t in the collaborative knowledge graph is ε′=ε∪U, and the range of the relationship r from the head entity h to the tail entity t is ε′=ε∪U.

[0068] It should be noted that a sequence of entities and relationships from user u to item i can also be defined as a path for the collaborative knowledge graph, as shown in formula (3).

[0069]

[0070] Among them, e 1 =u,e L =i(e l ,r l ,e l+1 ), L-1 represents the number of triplets in the path.

[0071] S202: Obtain historical user queries, and extract a set of historical query suggestions related to the queries from the collaborative knowledge graph.

[0072] In the specific process of executing step S202, after obtaining the collaborative knowledge graph, historical user queries can be further obtained. First, a candidate historical query suggestion set related to the historical user queries can be screened out from each entity in the collaborative knowledge graph; wherein the historical query suggestion set includes multiple candidate entities related to the historical user queries; for each candidate entity, multiple candidate related entities related to the candidate entity are obtained, and the multiple candidate related entities related to each candidate entity are combined into a candidate related historical query suggestion set; the candidate entity set is merged with the candidate related entity set to obtain a historical query suggestion set related to the historical user queries.

[0073] In some embodiments, a set of candidate entities related to historical user queries can be screened out from various entities in the collaborative knowledge graph through a graph mining algorithm of a random walk model.

[0074] Optionally, the process of filtering out a set of candidate entities related to historical user queries from various entities in the collaborative knowledge graph can be as follows: a graph mining algorithm based on a random walk model calculates the minimum walk time for each entity in the collaborative knowledge graph to reach historical user queries; based on the minimum walk time between historical user queries and each entity, multiple candidate entities are determined from each entity, and a corresponding candidate entity set is generated based on each candidate entity.

[0075] Specifically, G = (V, E) can be used to represent the collaborative knowledge graph connected graph of the collaborative knowledge graph, which contains n nodes and m edges, where n nodes are n entities in the collaborative knowledge graph; random walks start from node v 0 Initially, assume that we reach node v at step l t , and 1 / d(v t ) probability of selecting the node’s neighbor for the next move, then (v t :t=0,1,...) is a Markov chain. The initial node v 0 It may be fixed, but it can also be randomly selected, with a probability distribution of P 0 If P t The representation of can be shown as formula (4). The probability matrix is ​​used to represent the probability of each selection in the Markov chain, as shown in formula (5); the representation of the elements in the probability matrix can be shown as formula (6).

[0076] P t (i) = prob(v t =i)(4)

[0077] M=(p ij ) i,j∈V (5)

[0078]

[0079] Among them, p ij is the probability of selecting node i’s neighbor node j for the next walk.

[0080] Combining formula (4)-formula (6), the first arrival time in the random walk model can be defined as: Assuming that the initial query (historical user query) q is given, then starting from any node i (i≠q), the probability of selecting node i’s neighbor node j for the next walk is selected, and the node j for the next walk is selected at the node, so as to calculate the minimum walk time for node i to reach the historical user query q for the first time after several steps of random walk. and Satisfies formula (7), that is, when t = 0, the initial is 0. Set the threshold t, then from the initial Start by calculating the minimum travel time from node i to historical user query q The specific calculation method is shown in formula (8), where: The smaller it is, the more relevant the node i is to the historical user query q in the collaborative knowledge graph; then the minimum walk time corresponding to each entity is sorted from small to large, and the first N entities in the obtained sequence are taken as candidate entities related to the historical user query q, and the set of candidate entities is taken as the required candidate entity set.

[0081]

[0082] In some embodiments, after obtaining the candidate entity set, the candidate entity set can be recorded as K(e q ), where e q is a candidate entity; the correlation score between the historical user query q and each entity e is calculated by formula (9); each entity whose correlation score with the historical user query q is 1 is taken as a candidate entity of the historical user query q, and each determined candidate entity is added to the candidate entity set required above, so as to obtain a candidate entity set related to the historical user query q.

[0083]

[0084] Among them, P k (e|q) is the correlation score between the historical user query q and each entity e.

[0085] Furthermore, since the entity and relationship information in the collaborative knowledge graph may not be complete, the candidate related entity set of the historical user query q can be supplemented based on the search session co-occurrence. qThe top N most frequent co-occurrences in the same search session s Entities are extracted, and each extracted entity is used as a candidate related entity S(e) for the historical user query q. q ), so as to further screen out multiple candidate related entities from the extracted candidate related entities, and finally form a candidate related entity set with the screened candidate related entities. The main reason for adopting this method is that some users who use search engines will search for different queries multiple times in the same search session. The search logs accumulated by these search behaviors of users are an effective data source for mining entities and their relationships, because this information can help discover some entities that have potential associations with historical user queries q. Given a historical user query q and a candidate related entity S(e q ), point mutual information is used to estimate the correlation (correlation) between the two. The specific calculation formula is shown in formula (10). After calculating the historical user query q and each candidate related entity S (e q ), each candidate related entity whose correlation is greater than a preset correlation threshold can be formed into a candidate related entity set.

[0086]

[0087] In the above formula, P s (e|q) is the historical user query q and the candidate related entity S(e q );cnt(e,e q ) is entity e and candidate entity e q is the number of search sessions that have co-occurred, and cnt(e) is the number of search sessions containing entity e.

[0088] Finally, the entities obtained by the above two methods are merged to obtain a historical query suggestion set R(q) related to q, as shown in formula (11), where the historical query suggestion set includes multiple recommendation instances.

[0089] R(q)≈K(e q )∪S(e q )(11)

[0090] S203: According to each user in the user set, the historical user query and the historical query suggestion set related to the historical user query, a plurality of training triples are formed, and the personalized search recommendation model to be trained is trained according to the scoring function learned from the training set constructed by each training triple, so as to obtain the personalized search recommendation model.

[0091] In an embodiment of the present application, after obtaining a set of historical query suggestions related to a historical user query q, multiple training triples can be further formed based on each user in the user set, historical user queries, and a set of historical query suggestions related to the historical user queries, and the degree of interest corresponding to each training triple is calculated. Finally, a corresponding training set is constructed based on each training triple and its degree of interest, and the personalized search recommendation model to be trained is trained based on the scoring function learned from the training set constructed from each training triple to obtain a personalized search recommendation model.

[0092] Optionally, a plurality of training triples are formed according to each user in the user set, historical user queries, and a set of historical query suggestions related to the historical user queries, and the process of learning a scoring function of the personalized search recommendation model according to the training set constructed by each training triple can be specifically as follows: a plurality of training triples are formed according to each user in the user set, historical user queries, and a set of historical query suggestions related to the historical user queries, wherein each training triple is composed of a user, a historical user query, and a recommended entity; the historical query suggestion set includes multiple recommended entities; for each training triple, the user's search history in the training triple is obtained. The total number of clicks issued on the recommended entity at the time, and the interest of the training triplet is calculated according to the user, historical user query, recommended entity and the corresponding total number of clicks in the training triplet; a corresponding training set is constructed according to each training triplet and its interest, and the personalized search recommendation model to be trained is trained according to the scoring function learned from the training set constructed by each training triplet, the relevance of each candidate entity in the candidate entity set and the historical user query, the relevance of each candidate related entity in the candidate related entity set and the historical user query, and the interest of each recommended entity and the historical user query to obtain a personalized search recommendation model.

[0093] In actual applications, the user's click behavior is used as a signal to model whether the user is interested in the currently recommended entity when searching for a query. The main reason is that click behavior is an important observation signal that can characterize interest. Intuitively, if a user clicks on a recommended entity when searching for a query, it can be reasonably assumed that the user is interested in the recommended entity currently clicked, and thus clicks on the recommended entity to further learn more related information. Therefore, the aggregated click information obtained from large-scale search logs can be used to model the user's interest in entities. That is, compared with entities that do not attract any clicks, when the user enters the current query, those entities that attract users to click once or more are more likely to arouse the user's interest.

[0094] Specifically, according to each user in the user set, historical user queries and historical query suggestion sets related to historical user queries, multiple training triples are formed, where each training triple is composed of a user, a historical user query and a recommended entity; the historical query suggestion set includes multiple recommended entities; the learning objectives required for generating a training ranking model (a personalized search recommendation model to be trained) from the "query-entity" click data are used, that is, for each training triple, the user u i In the search history user query q j When recommending entity e k The total number of clicks issued, and based on user u i In the search history user query q j When recommending entity e k The total number of clicks issued, calculated for user u i In the search history user query q j When recommending entity e k The interest degree of the training triplet is obtained, and the specific calculation method is shown in formula (12).

[0095]

[0096] In the above formula, click(u i ,q j ,e k ) is user u i In the search query q j When recommending entity e k Total number of clicks issued, y ijk To train the ternary (u i ,q j ,e k )’s interest.

[0097] In the training ranking model, the feature function is constructed using two features: relevance and interest. Relevance is used to ensure that two given candidate entities e q The correlation between entity e and interest is used to estimate the user u's interest in the recommended entity e when the historical user query q k Given a training set H consisting of training triples and their corresponding interest degrees, i ,q j ,e k >,y ijk )}, the training goal is to learn a scoring function f of the ranking model from H, which is used to accurately classify a given training triplet <u i ,q j ,e k >Interest score f(u i ,q j,e k ). At the same time, you can also get user u i Search historical user queries under different conditions j When recommending entity e k The total number of clicks and impressions issued, and based on user u i Search historical user queries under different conditions j When recommending entity e k The total number of clicks and impressions issued is used to calculate the interest of each recommended entity and historical user queries at different granularities.

[0098] It should be noted that f 1 is the correlation (degree of relevance) between two entities in the collaborative knowledge graph, f 1 =P k (e|q);f 2 is the relevance of two entities in the search session, f 2 =P s (e|q).f 3 、f 4 、f 5 Reflects the interest of recommended entities of different granularities and historical user queries. The calculation method is as follows:

[0099]

[0100] In the above formula, click(·) and impression(·) are the recommended entities e k The total number of clicks and impressions under different conditions.

[0101] In some embodiments, after obtaining the scoring function of the personalized search recommendation model, the obtained scoring function, the relevance of each candidate entity in the candidate entity set to the historical user query, the relevance of each candidate related entity in the candidate related entity set to the historical user query, and the interest of each recommended entity to the historical user query at different granularities can be further used to train the personalized search recommendation model to obtain the corresponding personalized search recommendation model.

[0102] Specifically, based on the above-mentioned feature function (scoring function), a gradient boosted decision tree (GBDT) is used to train the sorting model, and the grid search method is used to determine the optimal hyperparameters of the sorting model within a given parameter range: number of trees, learning rate, and tree depth. The click data of the palm banking search engine is used to construct a knowledge graph in the financial field, and the search log is used as experimental data. The sorting model is learned by minimizing the objective function on the training set. Specifically, given a training set H consisting of training triples and their corresponding interest levels, H = {(<u i ,q j ,e k >,y ijk )}, the training goal is to learn a scoring function f from H to accurately classify a given training triplet <u i ,q j ,e k >Interest level f(u i ,q j ,e k ) is estimated, and finally the objective function of the minimization ranking model is adjusted by using the scoring function f, the relevance of each candidate entity in the candidate entity set to the historical user query, the relevance of each candidate related entity in the candidate related entity set to the historical user query, and the interest of each recommended entity to the historical user query, and the corresponding mapping function is obtained, as shown in formula (15), that is, the personalized search recommendation model is obtained. Among them, the method of minimizing the objective function of the ranking model by using the scoring function f is shown in formula (14).

[0103]

[0104] Among them, Loss(H) is the cross entropy loss function.

[0105] f':U×Q×E→S(15)

[0106] Among them, U is the user set consisting of all users currently registered on the mobile banking APP, Q is the set consisting of all queries, E is the set of historical query suggestions related to all queries in Q, and S is the recommendation degree; the function f(u,q,e) describes the interest of user u∈U in entity e∈E when searching for q∈Q.

[0107] Furthermore, in an embodiment of the present application, after obtaining the personalized search recommendation model, the personalized search recommendation model is verified to ensure the search recommendation efficiency and accuracy of the obtained personalized search recommendation model, thereby ensuring the efficiency and accuracy of the entities recommended to the target users by the subsequent use of the personalized search recommendation model.

[0108] Specifically, the algorithm can be evaluated by combining offline evaluation with online evaluation, that is, verifying the obtained personalized search recommendation model. The goal of offline evaluation is to evaluate the set of historical query suggestions related to the historical user query q input by user u, using the discounted cumulative gain (DCG) indicator. DCG is an evaluation indicator used to measure the ranking quality in information retrieval and is widely used to evaluate the relevance of search results. The specific evaluation method is shown in formula (16).

[0109]

[0110] Where p is the given sort position, It is the binary relevance (1 is relevant, 0 is irrelevant) of the entity at ranking position i in the ranking results of each recommended entity of the historical user query q input by user u. i in this formula represents the ranking position.

[0111] Offline evaluation can only evaluate the quality of the ranking results, but cannot evaluate the real user's participation in the recommendation results given by the personalized search recommendation model. In order to quantify and compare the user participation of the personalized search recommendation model, a reasonable way is to compare the click-through rate (CTR) of the recommended entities of different recommendation results. CTR is an effective evaluation indicator for evaluating the effectiveness of online services. For example, it can be used to evaluate the effectiveness of recommendation systems, search advertising, and web search results page functions. The specific evaluation method is shown in formula (17).

[0112]

[0113] Among them, R′(q) is the historical query suggestion set of q∈T, click(q,e) is the total number of times user u clicks on entity e, and impression(q,e) is the total number of pages that have displayed entity e when the user searches for q.

[0114] It should be noted that, when both the evaluation index and the effective evaluation index are passed, it can be considered that the verification of the obtained personalized search recommendation model has passed; if it fails, the personalized search recommendation model to be trained can be retrained.

[0115] S204: When a query currently input by the target user in the search session of the mobile banking APP is detected, the recommendation degree between the query and each entity in the mobile banking APP is calculated using the personalized search recommendation model, and based on the recommendation degree of each entity, the entities to be recommended are screened out from each entity, and each entity to be recommended is output.

[0116] In the embodiment of the present application, the pre-built knowledge graph and the user-item interaction graph built using the search log are combined as the data source, and the corresponding historical query suggestion set is screened from the data source (collaborative knowledge graph); the feature function (scoring function) is constructed using the two features of relevance and interest, and the relevance is used to ensure that two given candidate entities e q The correlation (relevance) between entity e and interest is used to estimate the user u's interest in the recommended entity e when the historical user query q k interest; using the obtained scoring function, the relevance of each candidate entity in the candidate entity set and the historical user query, the relevance of each candidate related entity in the candidate related entity set and the historical user query, and the interest of each recommended entity and the historical user query, the personalized search recommendation model to be trained is trained to obtain a personalized search recommendation model; after obtaining the personalized search recommendation model, it is possible to detect in real time whether the mobile banking APP currently has a query entered by the target user in the search session of the mobile banking APP. When the query is detected, the query can be input into the personalized search recommendation model, so that the personalized search model forms corresponding triples with the target user and the query respectively and each instance in the mobile banking APP, and uses its own mapping function to score the recommendation degree according to each obtained triple to obtain the recommendation degree of each instance; sorting each instance from large to small according to the recommendation degree of each instance, taking the first N entities in the obtained sorting sequence as the entities to be recommended, and finally sorting each entity to be recommended from large to small according to the recommendation degree, generating a recommendation set (recommendation sequence), and outputting the recommendation set, such as Figure 3 shown.

[0117] The present application provides a personalized search and recommendation method for mobile banking. By combining a pre-constructed knowledge graph and a user-item interaction graph constructed using search logs as a data source, it can effectively mine the relationships between more different entities, effectively solve the problem of sparse data rows in the prior art, and make the query results more relevant and coherent. In addition, the knowledge graph is built on a semantic model, which can better understand the semantic associations between entities, thereby better understanding the intent of user queries and providing more accurate information for subsequent search recommendations; combining features of two different granularities, relevance and interest, to model user preferences for entities, improves the personalization of query results, that is, using the two features of relevance and interest to construct a corresponding training set, so that the personalized search recommendation model trained using the constructed training set can more efficiently and accurately search for various entities that meet the needs of target users.

[0118] Based on the above-mentioned mobile banking personalized search and recommendation method provided in the embodiment of the present application, correspondingly, the embodiment of the present application also provides a mobile banking personalized search and recommendation system, such as Figure 4 As shown, the personalized search recommendation based on mobile banking includes:

[0119] The first acquisition unit 41 is used to acquire a pre-constructed collaborative knowledge graph, wherein the collaborative knowledge graph is composed of a user-item binary interaction graph and a knowledge graph, and the user-item binary interaction graph is composed of a user set, an item set, and a connection relationship between each user and each item; the collaborative knowledge graph includes multiple triples, each triple describes a relationship from a head entity to a tail entity, and each entity in the collaborative knowledge graph is aligned with each item;

[0120] An extraction unit 42, used to obtain historical user queries and extract a set of historical query suggestions related to the historical user queries from the collaborative knowledge graph;

[0121] The training unit 43 is used to form a plurality of training triples according to each user in the user set, the historical user query and the historical query suggestion set related to the historical user query, and train the personalized search recommendation model to be trained according to the scoring function learned from the training set constructed by each training triple, so as to obtain the personalized search recommendation model;

[0122] The recommendation unit 44 is used to, when detecting a query currently input by the target user in the search session of the mobile banking APP, calculate the recommendation degree between the query and each entity in the mobile banking APP using the personalized search recommendation model, filter out entities to be recommended from each entity according to the recommendation degree of each entity, and output each entity to be recommended.

[0123] The specific principles and execution processes of each unit in the personalized search and recommendation system for mobile banking disclosed in the above-mentioned embodiment of the present application are the same as those of the personalized search and recommendation method for mobile banking disclosed in the above-mentioned embodiment of the present application. Please refer to the corresponding parts of the personalized search and recommendation method for mobile banking disclosed in the above-mentioned embodiment of the present application, and no further details will be given here.

[0124] The present application provides a personalized search and recommendation method for mobile banking. By combining a pre-constructed knowledge graph and a user-item interaction graph constructed using search logs as a data source, it can effectively mine the relationships between more different entities, effectively solve the problem of sparse data rows in the prior art, and make the query results more relevant and coherent. In addition, the knowledge graph is built on a semantic model, which can better understand the semantic associations between entities, thereby better understanding the intent of user queries and providing more accurate information for subsequent search recommendations; combining features of two different granularities, relevance and interest, to model user preferences for entities, improves the personalization of query results, that is, using the two features of relevance and interest to construct a corresponding training set, so that the personalized search recommendation model trained using the constructed training set can more efficiently and accurately search for various entities that meet the needs of target users.

[0125] Extraction unit, comprising:

[0126] A screening unit, used to screen out a candidate entity set related to the historical user query from each entity in the collaborative knowledge graph; wherein the candidate entity set includes multiple candidate entities related to the historical user query;

[0127] A candidate related entity set construction unit is used to obtain, for each candidate entity, multiple candidate related entities related to the candidate entity, and construct a candidate related entity set according to the multiple candidate related entities related to each candidate entity;

[0128] The merging unit is used to merge the candidate entity set with the candidate related entity set to obtain a historical query suggestion set related to the historical user query.

[0129] Optional, screening unit, including:

[0130] The minimum walk time calculation unit is used to calculate the minimum walk time for each entity in the collaborative knowledge graph to reach the historical user query based on the graph mining algorithm of the random walk model;

[0131] The candidate entity set generating unit is used to determine a plurality of candidate entities from each entity according to the historical user query and the travel time of each entity, and generate a corresponding candidate entity set according to each candidate entity.

[0132] Optional training modules include:

[0133] A training triplet composition unit, used to compose a plurality of training triples according to each user in the user set, a historical user query, and a historical query suggestion set related to the historical user query, wherein each training triplet consists of a user, a historical user query, and a recommended entity; and the historical query suggestion set includes a plurality of recommended entities;

[0134] An interest calculation unit is used to obtain, for each training triplet, the total number of clicks issued by the user in the training triplet to the recommended entity when searching for the historical user query, and calculate the interest of the training triplet according to the user, the historical user query, the recommended entity and the corresponding total number of clicks in the training triplet;

[0135] The training subunit is used to construct a corresponding training set according to each training triple and its interest, and train the personalized search recommendation model to be trained according to the scoring function learned from the training set constructed by each training triple, the relevance of each candidate entity in the candidate entity set to the historical user query, the relevance of each candidate related entity in the candidate related entity set to the historical user query, and the interest of each recommended entity to the historical user query, so as to obtain a personalized search recommendation model.

[0136] Furthermore, the embodiment of the present application provides a personalized search and recommendation system for mobile banking, further comprising:

[0137] A verification unit, used to verify the personalized search recommendation model;

[0138] The execution unit is used to execute the recommendation unit if the verification of the personalized search recommendation model passes.

[0139] The present application also provides a storage medium, in which program instructions are stored. When the program instructions are loaded and executed by a processor, any one of the above-mentioned unsupervised cross-modal review recommendation method embodiments is implemented.

[0140] The present application also provides an electronic device, such as Figure 5 As shown, the device includes a processor 501 and a memory 502; the memory stores program instructions; the processor calls the program instructions in the memory to execute any of the above-mentioned unsupervised cross-modal review recommendation method embodiments.

[0141] The processor in this article can be a CPU of a terminal, or an MCU integrated in the terminal, or a combination of a CPU and an MCU. Moreover, the processor includes a kernel, which calls a corresponding program from a memory, and one or more kernels can be set.

[0142] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM, and the memory includes at least one memory chip.

[0143] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system or system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment. The system and system embodiments described above are merely schematic, in which the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0144] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0145] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

[0146] The above are only preferred embodiments 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 the scope of protection of the present invention.

Claims

1. A personalized search and recommendation method for mobile banking, characterized in that: Applicable to the mobile banking APP, the method includes: Obtain a pre-built collaborative knowledge graph, wherein the collaborative knowledge graph is composed of a user-item interaction graph and a knowledge graph, wherein the user-item interaction graph is composed of a user set, an item set, and a connection relationship between each user and each item; the collaborative knowledge graph includes a plurality of triples, each of which describes a relationship from a head entity to a tail entity, and each entity in the collaborative knowledge graph is aligned with each item; Obtaining historical user queries, and extracting a set of historical query suggestions related to the historical user queries from the collaborative knowledge graph; According to each user in the user set, the historical user query and a set of historical query suggestions related to the historical user query, a plurality of training triples are formed, and the personalized search recommendation model to be trained is trained according to the scoring function learned from the training set constructed by each of the training triples to obtain a personalized search recommendation model; When a query currently input by the target user in the search session of the mobile banking APP is detected, the personalized search recommendation model is used to calculate the recommendation degree between the query and each entity in the mobile banking APP, and based on the recommendation degree of each displayed entity, the entities to be recommended are screened out from each of the entities, and each of the entities to be recommended is output.

2. The method according to claim 1, characterized in that The obtaining of historical user queries and extracting a set of historical query suggestions related to the historical user queries from the collaborative knowledge graph includes: Filtering a set of candidate entities related to the historical user query from each entity in the collaborative knowledge graph; wherein the set of candidate entities includes multiple candidate entities related to the historical user query; For each of the candidate entities, multiple candidate related entities related to the candidate entity are obtained, and a candidate related entity set is constructed according to the multiple candidate related entities related to each of the candidate entities; The candidate entity set and the candidate related entity set are merged to obtain a historical query suggestion set related to the historical user query.

3. The method according to claim 2, characterized in that The step of selecting a candidate entity set related to the historical user query from each entity in the collaborative knowledge graph includes: A graph mining algorithm based on a random walk model calculates the minimum walk time for each entity in the collaborative knowledge graph to reach the historical user query; Determine a plurality of candidate entities from each of the entities according to the minimum travel time between the historical user query and each of the entities; Calculating the correlation between the historical user query and each of the entities, and selecting a plurality of candidate entities from each of the entities according to the correlation between the historical user query and each of the entities; Generate a corresponding candidate entity set according to each of the candidate entities.

4. The method according to claim 3, characterized in that The step of acquiring, for each candidate entity, a plurality of candidate related entities related to the candidate entity, and constructing a candidate related entity set according to the plurality of candidate related entities related to each candidate entity includes: For each candidate entity, multiple candidate related entities related to the candidate entity are obtained; wherein the multiple candidate related entities are the top N entities that co-occur multiple times with the candidate entity in the same search session and have the highest number of co-occurrences. s Entity; The relevance between the historical user query and each of the candidate related entities is calculated, and a candidate related entity set is screened out from the candidate related entities according to the relevance of each of the candidate related entities.

5. The method according to claim 4, characterized in that The method comprises forming a plurality of training triples according to each user in the user set, the historical user query and a historical query suggestion set related to the historical user query, and training the personalized search recommendation model to be trained according to the scoring function learned from the training set constructed by each training triple to obtain the personalized search recommendation model, including: According to each user in the user set, the historical user query and a set of historical query suggestions related to the historical user query, a plurality of training triples are formed, wherein each of the training triples is composed of the user, the historical user query and a recommended entity; the set of historical query suggestions includes a plurality of recommended entities; For each of the training triples, obtaining the total number of clicks issued by the users in the training triples to the recommended entity when searching for the historical user query, and calculating the interest level of the training triples according to the users, the historical user queries, the recommended entities and the corresponding total number of clicks in the training triples; A corresponding training set is constructed according to each of the training triples and their interest levels, and a personalized search recommendation model to be trained is trained based on the scoring function learned from the training set constructed from each of the training triples, the relevance of each candidate entity in the candidate entity set to the historical user query, the relevance of each candidate related entity in the candidate related entity set to the historical user query, and the interest level of each recommended entity to the historical user query to obtain a personalized search recommendation model.

6. The method according to claim 1, characterized in that After obtaining the personalized search recommendation model, the method further includes: Verifying the personalized search recommendation model; If the verification of the personalized search recommendation model passes, the step of using the personalized search recommendation model to calculate the recommendation degree between the query and each entity in the mobile banking APP when detecting the query currently entered by the target user in the search session of the mobile banking APP, and filtering out the entities to be recommended from each of the entities according to the recommendation degree of each of the entities, and outputting each of the entities to be recommended is executed.

7. A mobile banking personalized search recommendation system, characterized in that: Applicable to mobile banking APP, the system includes: A first acquisition unit is used to acquire a pre-constructed collaborative knowledge graph, wherein the collaborative knowledge graph is composed of a user-item binary interaction graph and a knowledge graph, and the user-item binary interaction graph is composed of a user set, an item set, and a connection relationship between each user and each item; the collaborative knowledge graph includes a plurality of triples, each of which describes a relationship from a head entity to a tail entity, and each entity in the collaborative knowledge graph is aligned with each item; An extraction unit, configured to obtain historical user queries and extract a set of historical query suggestions related to the historical user queries from the collaborative knowledge graph; A training unit, configured to form a plurality of training triples according to each user in the user set, the historical user queries, and a set of historical query suggestions related to the historical user queries, and to train a personalized search recommendation model to be trained according to a scoring function learned from a training set constructed from each of the training triples, so as to obtain a personalized search recommendation model; The recommendation unit is used to, when detecting a query currently input by a target user in the search session of the mobile banking APP, calculate the recommendation degree between the query and each entity in the mobile banking APP using the personalized search recommendation model, filter out entities to be recommended from each of the entities according to the recommendation degree of each of the entities, and output each of the entities to be recommended.

8. The system according to claim 7, characterized in that The extraction unit comprises: A screening unit, configured to screen out a candidate historical query suggestion set related to the historical user query from each entity in the collaborative knowledge graph; wherein the historical query suggestion set includes a plurality of candidate entities related to the historical user query; A candidate related historical query suggestion set building unit is used to obtain, for each candidate entity, multiple candidate related entities related to the candidate entity, and build a candidate related entity set according to the multiple candidate related entities related to each candidate entity; A merging unit is used to merge the candidate entity set with the candidate related entity set to obtain a historical query suggestion set related to the historical user query.

9. An electronic device, characterized in that: include: A processor and a memory, wherein the processor and the memory are connected via a communication bus; wherein the processor is used to call and execute a program stored in the memory; The memory is used to store a program, and the program is used to implement the mobile banking personalized search and recommendation method as described in any one of claims 1-6.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, and the computer-executable instructions are used to execute the mobile banking personalized search and recommendation method according to any one of claims 1 to 6.

Images