Intelligent recommendation method and system based on knowledge graph

Abstract

The invention relates to the technical field of information, in particular to an intelligent recommendation method and system based on a knowledge graph. The method comprises the steps of collecting multiple pieces of target information, constructing multiple triples based on the target information, creating a knowledge graph by combining the multiple triples, then creating a recommendation model, importing triple data into the recommendation model, then obtaining user information and user search records, and inputting the user information and the user search records into the recommendation model. The method comprises the following steps: establishing a relation between user information and a triple by a recommendation model to obtain a trained recommendation model, finally collecting user real-time information, inputting the user real-time information into the trained recommendation model, and obtaining a user recommendation result output by the trained recommendation model. According to the method, the recommendation target most suitable for the user is obtained by establishing the user information, the user search record and the knowledge sub-path of the knowledge graph and combining the time decay edge of the knowledge sub-path, so that the accuracy of the recommendation target is improved.

Description

Technical Field

[0001] The present invention relates to the field of information technology, and in particular to an intelligent recommendation method and system based on a knowledge graph. Background Art

[0002] A knowledge graph is called knowledge domain visualization or knowledge domain mapping map in the library and information field. It is a series of various graphs showing the development process and structural relationships of knowledge. Visualization technology is used to describe knowledge resources and their carriers, and to mine, analyze, construct, draw, and display knowledge and the interconnections between them.

[0003] Chinese Patent with publication number CN118585654A discloses a node recommendation method, system, storage medium, and processor based on a knowledge graph. Through real-time analysis based on the IT asset knowledge topology map combined with the health status, potential security risks can be detected early, helping organizations take corresponding preventive and repair measures to reduce the risk of information systems being attacked or data being leaked. However, in the prior art, the relevant information of users is not considered, resulting in it being difficult to establish a connection between the recommended results and user information, thus making it difficult to meet the search needs of customers. Summary of the Invention

[0004] The object of the present invention is to address the problems in the background art and propose an intelligent recommendation method and system based on a knowledge graph.

[0005] The technical solution of the present invention:

[0006] On the one hand, the present application provides an intelligent recommendation method based on a knowledge graph, including:

[0007] Collect multiple target information, construct multiple triples based on the target information, and create a knowledge graph by combining the multiple triples;

[0008] Create a recommendation model and import the triple data into the recommendation model;

[0009] Obtain user information and user search records, input the user information and user search records into the recommendation model, so that the recommendation model establishes a connection between the user information and the triples, and obtain the trained recommendation model;

[0010] Collect user real-time information, input the user real-time information into the trained recommendation model, and obtain the user recommendation results output by the trained recommendation model.

[0011] Preferably, collecting multiple target information, constructing multiple triples based on the target information, and creating a knowledge graph by combining the multiple triples includes:

[0012] Create a target information table;

[0013] Collect the target information of multiple targets and put all the collected target information into the target information table; the target information includes entity information, relationship information, and attribute information;

[0014] Select a target from the target information table and construct a triple based on the target information of this target;

[0015] Return to select a target from the target information table until all the targets in the target information table have been selected, and obtain multiple triples;

[0016] Construct the data layer of the knowledge graph based on the set of multiple triples.

[0017] Preferably, before creating a recommendation model and importing the triple data into the recommendation model, it includes:

[0018] Obtain multiple user information and the user search records of each user; the user information includes the user's occupation, age, and health information;

[0019] Sort all the search targets in the user search records according to the search frequency from high to low, and screen out the top N search targets; the selected N search targets are recorded as high-frequency targets;

[0020] For each high-frequency target, establish the corresponding relationship between the high-frequency target and the user information of the user corresponding to this high-frequency target, and record the high-frequency target, user information, and the corresponding relationship of high-frequency-user information as an initial sample, and obtain N initial samples.

[0021] Preferably, collecting stock information, creating a recommendation model, and importing the triple data into the recommendation model, includes:

[0022] Create a recommendation model;

[0023] Divide all the initial samples into a training set and a test set according to a random ratio;

[0024] Input the initial samples in the test set into the recommendation model, so that the recommendation model continuously establishes the coupling relationship between the high-frequency target and the triple, and obtains the trained recommendation model; the trained recommendation model has the ability to automatically obtain the target with a high recommendation value corresponding to the input user information;

[0025] Input the test set into the trained recommendation model to verify whether the trained recommendation model is trained successfully.

[0026] Preferably, inputting the test set into the recommendation model, so that the recommendation model continuously establishes the coupling relationship between the high-frequency target and the triple, and obtains the trained recommendation model, includes:

[0027] Import triple data into the recommendation model;

[0028] Randomly select an initial sample from the training set;

[0029] Select triples from the triple data that contain entity targets corresponding to the search targets of the initial sample, and denote the selected triples as target triples;

[0030] Establish the correspondence between the target triples and user information to generate knowledge sub-paths;

[0031] Return to randomly select an initial sample from the training set until all initial samples in the training set have been selected, obtaining multiple knowledge sub-paths.

[0032] Preferably, input the test set into the recommendation model to enable the recommendation model to continuously establish the coupling relationship between high-frequency targets and triples, obtaining the trained recommendation model, further including:

[0033] Select a knowledge sub-path and obtain the user search information corresponding to the knowledge sub-path;

[0034] Calculate the time decay edge of the knowledge sub-path based on the user search information through Formula 1;

[0035]

[0036] where, P IM is the time decay edge of the knowledge sub-path, Q i is the i-th data in the knowledge sub-path, is the time decay weight corresponding to the i-th data, and K is the total number of data included in the knowledge sub-path;

[0037] Return to select a knowledge sub-path until all knowledge sub-paths have been selected, obtaining the time decay edge of each knowledge sub-path;

[0038] Sort the time decay edges of all knowledge sub-paths from largest to smallest, obtain and delete the knowledge sub-paths corresponding to the last M time decay edges;

[0039] Output the remaining N - M knowledge sub-paths.

[0040] Preferably, collect user real-time information, input the user real-time information into the trained recommendation model, and obtain the user recommendation result output by the trained recommendation model, including:

[0041] Obtain user real-time information;

[0042] Input the user real-time information into the trained recommendation model and obtain the recommendation result output by the trained recommendation model;

[0043] Convert the recommended results output by the trained recommendation model into readable text and output it.

[0044] Preferably, the recommendation model includes a heterogeneous information network and a language output layer. The triple data is mapped to low-dimensional vectors through the heterogeneous information network, and the knowledge path is converted into readable text and output through the language output layer.

[0045] On the other hand, the present application also provides an intelligent recommendation system based on a knowledge graph, including an information collection component and a processing component. The target information and user information are collected through the information collection component, and the intelligent recommendation method based on the knowledge graph described in any one of the foregoing is executed through the processing component. The processing component is communicatively connected to the information collection component. After the information collected by the information collection component is transmitted to the processing component, the processing component makes recommendations to the user according to the target information and user information.

[0046] Preferably, the processing component includes a knowledge construction layer, an inference layer, and a feedback layer. The triple data is created through the knowledge construction layer for the target information, the connection between the target information and the triple data is constructed through the inference layer, and finally the user recommendation content is output through the feedback layer.

[0047] Compared with the prior art, the above technical solution of the present invention has the following beneficial technical effects:

[0048] By collecting multiple target information, constructing multiple triples based on the target information, combining the multiple triples to create a knowledge graph, then creating a recommendation model, importing the triple data into the recommendation model, then obtaining the user information and user search records, inputting the user information and user search records into the recommendation model, enabling the recommendation model to establish the connection between the user information and the triples, obtaining the trained recommendation model, and finally collecting the user real-time information, inputting the user real-time information into the trained recommendation model, obtaining the user recommendation results output by the trained recommendation model, the present application improves the accuracy of the recommendation target by establishing the knowledge sub-path between the user information, user search records and the knowledge graph, and combining the time decay edge of the knowledge sub-path to obtain the most suitable recommendation target for the user. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 It is a schematic flow chart of an intelligent recommendation method based on a knowledge graph proposed by the present invention;

[0050] Figure 2 It is a schematic structural diagram of an intelligent recommendation system based on a knowledge graph proposed by the present invention;

[0051] Figure 3 It is a schematic data flow diagram of an intelligent recommendation system based on a knowledge graph proposed by the present invention;

[0052] Reference signs: 100, information collection component; 200, processing component; 201, knowledge construction layer;

[0053] 202, inference layer; 203, feedback layer. Detailed implementation mode

[0054] In the first embodiment, as Figure 1 shown, an intelligent recommendation method based on a knowledge graph proposed by the present invention includes:

[0055] S100, collecting multiple target information, constructing multiple triples based on the target information, and creating a knowledge graph by combining the multiple triples;

[0056] S200, creating a recommendation model and importing the triple data into the recommendation model;

[0057] S300, obtaining user information and user search records, inputting the user information and user search records into the recommendation model, so that the recommendation model establishes a connection between the user information and the triples, and obtaining a trained recommendation model;

[0058] S400, collecting real-time user information, inputting the real-time user information into the trained recommendation model, and obtaining the user recommendation result output by the trained recommendation model.

[0059] In the present invention, by collecting multiple target information, constructing multiple triples based on the target information, creating a knowledge graph by combining the multiple triples, then creating a recommendation model, importing the triple data into the recommendation model, then obtaining user information and user search records, inputting the user information and user search records into the recommendation model, so that the recommendation model establishes a connection between the user information and the triples, obtaining a trained recommendation model, and finally collecting real-time user information, inputting the real-time user information into the trained recommendation model, and obtaining the user recommendation result output by the trained recommendation model. The present application improves the accuracy of the recommendation target by establishing a knowledge sub-path between the user information, the user search record and the knowledge graph, and combining the time decay edge of the knowledge sub-path to obtain the most suitable recommendation target for the user.

[0060] In an optional embodiment, the 100 includes:

[0061] S110, creating a target information table;

[0062] S120, collecting the target information of multiple targets and putting all the collected target information into the target information table; the target information includes entity information, relationship information and attribute information;

[0063] S130, selecting a target from the target information table and constructing a triple based on the target information of the target;

[0064] S140, Return to step S130 until all the targets in the target information table are selected, obtaining multiple triples;

[0065] S150, Based on the set of multiple triples, construct the data layer of the knowledge graph.

[0066] It should be noted that a knowledge graph is a technical system that organizes and represents knowledge in the form of a graph structure. Its core feature is to reveal the relevance between data through semantic modeling of entities, relationships, and attributes. Therefore, the triples contain three elements: "entity", "relationship", and "attribute", thus constructing the basic unit in the knowledge graph. For example, for A who is a teacher in School A, the first triple constructed with A as the object can be A - employed at - School A. Similarly, for A, the second triple can be A - occupation - teacher. Then, combining the first triple and the second triple can construct the knowledge graph about A.

[0067] In an alternative embodiment, before the S200, it includes:

[0068] K100, Obtain multiple user information and the user search records of each user; the user information includes the user's occupation, age, and health information;

[0069] K110, Sort all the search targets in the user search records from high to low according to the search frequency, and screen out the top N search targets; Denote the selected N search targets as high - frequency targets;

[0070] K120, For each high - frequency target, respectively establish the corresponding relationship between the high - frequency target and the user information of the user corresponding to this high - frequency target, and denote the high - frequency target, user information, and the corresponding relationship of high - frequency - user information as an initial sample, obtaining N initial samples.

[0071] It should be noted that before training the recommendation model, by collecting the user information and user search records of multiple users, training samples are created for training the recommendation model. When creating the training samples, by screening out the high - frequency targets and establishing the corresponding relationship between the high - frequency targets and the user information, it is possible to establish knowledge sub - paths between the high - frequency targets and the existing triples in the knowledge graph during the training of the recommendation model, thereby expanding the scalability of the knowledge graph.

[0072] In an alternative embodiment, the S200 includes:

[0073] S210, Create a recommendation model;

[0074] S220, Divide all the initial samples into a training set and a test set according to a random ratio;

[0075] Specifically, the division ratio of the training set is higher than that of the test set.

[0076] S230. Input the initial samples in the test set into the recommendation model, so that the recommendation model continuously establishes the coupling relationship between the high-frequency target and the triple, and obtain the trained recommendation model; the trained recommendation model has the ability to automatically obtain the target with a high recommendation value corresponding to the input user information.

[0077] S240. Input the test set into the trained recommendation model to verify whether the trained recommendation model is trained completely.

[0078] Specifically, when verifying whether the trained recommendation model is trained completely, the response speed and / or recommendation satisfaction of the trained recommendation model can be used as the judgment criteria.

[0079] It should be noted that by sequentially inputting the training samples in the training set into the recommendation model, the recommendation model continuously learns to establish the knowledge sub-path between the high-frequency target and the user information, so that the trained knowledge graph can find the appropriate knowledge sub-path according to the imported user information and the user search information, and thus recommend the appropriate target to the user based on the knowledge sub-path.

[0080] In an optional embodiment, the S230 includes:

[0081] S231. Import the triple data into the recommendation model.

[0082] S232. Randomly select an initial sample from the training set.

[0083] S233. Select the triple containing the entity target corresponding to the search target of the initial sample from the triple data, and denote the selected triple as the target triple.

[0084] S234. Establish the corresponding relationship between the target triple and the user information, and generate the knowledge sub-path.

[0085] S235. Return to randomly select an initial sample from the training set until all the initial samples in the training set are selected, and obtain multiple knowledge sub-paths.

[0086] It should be noted that since the initial samples are created based on high-frequency targets and user information, a knowledge sub-path can be established between the triples corresponding to the high-frequency targets and user information through the recommendation model. For example, if the high-frequency target is "linear equation with two variables", for user A, since "linear equation with two variables" belongs to mathematical equations and considering user A's attribute as a teacher, a knowledge sub-path of A - linear equation with two variables - mathematical equations - teacher can be generated. As a result, when other users search for a math teacher, there is a chance to recommend user A to other users, that is, by establishing the knowledge sub-path to expand.

[0087] In an optional embodiment, S230 further includes:

[0088] S236, select a knowledge sub-path and obtain the user search information corresponding to this knowledge sub-path;

[0089] S237, calculate the time decay edge of this knowledge sub-path based on the user search information through Formula 1;

[0090]

[0091] where P IM is the time decay edge of the knowledge sub-path, Q i is the i-th data in the knowledge sub-path, is the time decay weight corresponding to the i-th data, and K is the total number of data included in the knowledge sub-path;

[0092] Specifically, the time decay weight refers to the residence time of the user on the i-th data. The longer the user's residence time, the stronger the attraction of this data to the user. Then, the time decay edge obtained in this way can reflect the attraction of the entire knowledge sub-path to the user, that is, the knowledge sub-path with a larger time decay edge has a stronger attraction to the user and is more suitable for pushing to the user;

[0093] S238, return to step S236 until all knowledge sub-paths are selected, and obtain the time decay edges of each knowledge sub-path;

[0094] S239, sort the time decay edges of all knowledge sub-paths from largest to smallest, obtain and delete the knowledge sub-paths corresponding to the last M time decay edges;

[0095] Specifically, M is less than N, and both M and N are positive integers;

[0096] S239-1, output the remaining N - M knowledge sub-paths.

[0097] It should be noted that multiple knowledge sub-paths are obtained through the foregoing steps. However, it is difficult to evaluate the advantages and disadvantages of these knowledge sub-paths. Therefore, by calculating the time decay edges of each knowledge sub-path and using the magnitude of the time decay edges as the criterion for evaluating the advantages and disadvantages of the knowledge sub-paths, the larger the time decay edge, the better the corresponding knowledge sub-path. That is, for a knowledge sub-path with a larger time decay edge, the connectivity between its high-frequency targets and the user is stronger and it is more worthy of recommendation. Therefore, the M knowledge sub-paths with the time decay edges at the end can be eliminated, thereby reducing the recommendation of targets with lower recommendation values to the user.

[0098] In an alternative embodiment, the S400 includes:

[0099] S410, obtaining real-time user information;

[0100] Specifically, the real-time user information includes user identity information and real-time user search information;

[0101] S420, inputting the real-time user information into the trained recommendation model and obtaining the recommendation result output by the trained recommendation model;

[0102] S430, converting the recommendation result output by the trained recommendation model into readable text and outputting it.

[0103] It should be noted that after receiving the real-time user information, the trained recommendation model creates knowledge sub-paths according to the real-time user information and calculates the magnitude of the time decay edge of each knowledge sub-path. Finally, the qualified knowledge sub-paths are converted into readable text and displayed to the user through the visualization module.

[0104] In an alternative embodiment, the recommendation model includes a heterogeneous information network and a language output layer. The triple data is mapped to a low-dimensional vector through the heterogeneous information network, and the knowledge path is converted into readable text and output through the language output layer.

[0105] It should be noted that the heterogeneous information network in the present application enables the recommendation model to continuously learn heterogeneous network embeddings, thereby realizing the continuous update of the recommendation model, while the language output layer can convert the knowledge sub-path into natural language to facilitate user understanding.

[0106] Such as Figure 2As shown in the figure, the present application also provides an intelligent recommendation system based on a knowledge graph, which includes an information collection component 100 and a processing component 200. The target information and user information are collected through the information collection component 100, and the intelligent recommendation method based on the knowledge graph described in any one of the embodiments in Embodiment 1 is executed through the processing component 200. The processing component 200 is communicatively connected to the information collection component 100. After the information collected by the information collection component 100 is transmitted to the processing component 200, the processing component 200 makes recommendations to the user according to the target information and the user information.

[0107] It should be noted that the target information and the user information are respectively collected through the information collection component 100, and the collected target information and user information are transmitted into the processing component 200. The processing component 200 combines the target information and the user information to filter out high-frequency targets, and combines the high-frequency targets to filter out appropriate knowledge sub-paths, and finally recommends appropriate search results to the user based on the knowledge sub-paths.

[0108] As Figure 3 shown, in an optional embodiment, the processing component 200 includes a knowledge construction layer 201, an inference layer 202, and a feedback layer 203. Triad data is created for the target information through the knowledge construction layer 201, the connection between the target information and the triad data is constructed through the inference layer 202, and finally the user recommendation content is output through the feedback layer 203.

[0109] It should be noted that after the user information and the target information reach the knowledge construction layer 201, the semantic relationships in the information text are extracted through the knowledge construction layer 201 to identify the association relationships between entities. Then, through the inference layer 202, the fusion of multi-source information is realized, and after inputting the high-frequency targets, knowledge sub-paths are established to continuously expand the scope of the knowledge graph. Finally, through the feedback layer 203, the knowledge sub-paths are transformed into visual texts to recommend appropriate search targets to the customers.

[0110] The embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the present invention is not limited thereto. Various changes can be made without departing from the spirit of the present invention within the scope of knowledge possessed by those skilled in the art to which the present invention pertains.

Claims

1. An intelligent recommendation method based on a knowledge graph, characterized in that Including: Collect multiple target information, construct multiple triples based on the target information, and create a knowledge graph by combining multiple triples; Create a recommendation model and import the triple data into the recommendation model; Obtain user information and user search records, input the user information and user search records into the recommendation model, so that the recommendation model establishes the connection between the user information and the triples, and obtain the trained recommendation model; Collect user real-time information, input the user real-time information into the trained recommendation model, and obtain the user recommendation results output by the trained recommendation model.

2. The intelligent recommendation method based on a knowledge graph according to claim 1, wherein Collect multiple target information, construct multiple triples based on the target information, and create a knowledge graph by combining multiple triples, including: Create a target information table; Collect the target information of multiple targets and put all the collected target information into the target information table; the target information includes entity information, relationship information and attribute information; Select a target from the target information table and construct a triple based on the target information of the target; Return to select a target from the target information table until all the targets in the target information table are selected, and obtain multiple triples; Construct the data layer of the knowledge graph based on the set of multiple triples.

3. The intelligent recommendation method based on a knowledge graph according to claim 2, wherein Before creating a recommendation model and importing the triple data into the recommendation model, including: Obtain multiple user information and the user search records of each user; the user information includes the user's occupation, age and health information; Sort all the search targets in the user search records from high to low according to the search frequency, and screen out the top N search targets; the selected N search targets are recorded as high-frequency targets; For each high-frequency target, establish the corresponding relationship between the high-frequency target and the user information of the user corresponding to the high-frequency target, and record the high-frequency target, user information and the corresponding relationship of high-frequency-user information as an initial sample, and obtain N initial samples.

4. An intelligent recommendation method based on a knowledge graph according to claim 3, wherein Collect stock information, create a recommendation model, and import the triple data into the recommendation model, including: Create a recommendation model; Divide all the initial samples into a training set and a test set according to a random ratio; Input the initial samples in the test set into the recommendation model, so that the recommendation model continuously establishes the coupling relationship between the high-frequency target and the triples, and obtain the trained recommendation model; the trained recommendation model has the ability to automatically obtain the target with a high recommendation value corresponding to the input user information; Input the test set into the trained recommendation model to verify whether the trained recommendation model is trained.

5. An intelligent recommendation method based on a knowledge graph according to claim 4, characterized in that, Input the test set into the recommendation model, so that the recommendation model continuously establishes the coupling relationship between the high-frequency target and the triples, and obtain the trained recommendation model, including: Import the triple data into the recommendation model; Randomly select an initial sample from the training set; Select the triples containing the entity targets corresponding to the search targets of the initial sample from the triple data, and record the selected triples as target triples; Establish the corresponding relationship between the target triple and the user information to generate a knowledge sub-path; Return to randomly select an initial sample from the training set until all the initial samples in the training set are selected, and obtain multiple knowledge sub-paths.

6. The intelligent recommendation method based on a knowledge graph according to claim 5, wherein Input the test set into the recommendation model, enabling the recommendation model to continuously establish the coupling relationship between high-frequency targets and triples, and obtaining the trained recommendation model. The method further includes: Select a knowledge sub-path and obtain the user search information corresponding to the knowledge sub-path; Calculate the time decay edge of the knowledge sub-path based on the user search information through Formula 1; Among them, P IM is the time decay edge of the knowledge sub-path, Q i is the i-th data in the knowledge sub-path, is the time decay weight corresponding to the i-th data, and K is the total number of data included in the knowledge sub-path; Return to select a knowledge sub-path until all knowledge sub-paths are selected, and obtain the time decay edge of each knowledge sub-path; Sort the time decay edges of all knowledge sub-paths from largest to smallest, and obtain and delete the knowledge sub-paths corresponding to the last M time decay edges; Output the remaining N - M knowledge sub-paths.

7. An intelligent recommendation method based on a knowledge graph according to claim 6, characterized in that, Collect user real-time information, input the user real-time information into the trained recommendation model, and obtain the user recommendation result output by the trained recommendation model. The method includes: Obtain user real-time information; Input the user real-time information into the trained recommendation model, and obtain the recommendation result output by the trained recommendation model; Convert the recommendation result output by the trained recommendation model into readable text and output it.

8. An intelligent recommendation method based on a knowledge graph according to claim 7, characterized in that The recommendation model includes a heterogeneous information network and a language output layer. The triple data is mapped to a low-dimensional vector through the heterogeneous information network, and the knowledge path is converted into readable text and output through the language output layer.

9. An intelligent recommendation system based on a knowledge graph, characterized in that, It includes: An information collection component that collects target information and user information through the information collection component; A processing component that executes the knowledge graph-based intelligent recommendation method described in any one of claims 1 to 8 through the processing component. The processing component is communicatively connected to the information collection component. After the information collected by the information collection component is transmitted to the processing component, the processing component makes recommendations to the user based on the target information and user information.

10. An intelligent recommendation system based on a knowledge graph according to claim 9, characterized in that, The processing component includes a knowledge construction layer, an inference layer, and a feedback layer. The triple data is created through the knowledge construction layer for the target information, the connection between the target information and the triple data is constructed through the inference layer, and finally the user recommendation content is output through the feedback layer.

Images