An item recommendation method, apparatus, medium, and device

Abstract

The application discloses an article recommendation method, device, medium and equipment, comprising: obtaining a first association result according to an external information resource; obtaining a second association result according to historical operation data of a user, and obtaining a third association result according to article index data; obtaining a meta-path sequence set according to the first association result, the second association result and the third association result; training a machine learning model according to the meta-path sequence set to obtain a vector embedding representation of the article; calculating a fourth association result between articles according to the vector embedding representation of the article; and determining a target article in the article according to the historical operation data of a target user and the fourth association result to recommend the target article to the target user. The application relates to the fields of natural language processing and content recommendation, and similarity between labels is constructed by introducing external information to calculate similarity between articles, so that the diversification and accuracy of article recommendation are improved.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, specifically to a method, apparatus, medium, and device for recommending items. Background Technology

[0002] Artificial Intelligence (AI) is a comprehensive technology within computer science that studies the design principles and implementation methods of various intelligent machines, enabling them to possess perception, reasoning, and decision-making capabilities. AI technology is a multidisciplinary field encompassing a wide range of areas, including natural language processing, machine learning, and deep learning. With technological advancements, AI will be applied in more fields and play an increasingly important role. Content recommendation is a crucial application area of ​​AI. Recommendation systems, as information processing systems, predict content that users may be interested in, such as news, short videos, products, movies, articles, friends, stocks, games, food, and anything else that can be recommended, and then present this content to the user. It has already been widely applied in e-commerce, video websites, social networking sites, personalized reading, and personalized advertising.

[0003] For a recommendation system to provide truly useful recommendations to users, efficient and accurate recommendation techniques are crucial; that is, the characteristics and potential of recommendation filtering techniques are paramount. In recommendation systems, recommendation filtering techniques mainly include content-based filtering and collaborative filtering. For example, collaborative filtering includes two methods: item-based collaborative filtering and user-based collaborative filtering. The former primarily obtains an item feature matrix based on the historical clicks of items, calculates the similarity between items based on the feature matrix, and then recommends other items to the user based on the user's historical clicks and the similarity between items. The latter primarily obtains a user feature matrix based on the user's historical click data, calculates the similarity between users based on the feature matrix, and then selects items clicked by these similar users but not clicked by the target user for recommendation. The former is prone to item clustering; that is, for popular items, many items may have high similarity, resulting in low coverage and diversity of recommended items. The latter suffers from data sparsity and scalability issues. Because the number of videos clicked by users is relatively small, the data is sparse, leading to low accuracy in calculating user similarity, resulting in low recommendation accuracy and weak generalization performance.

[0004] Recently, deep learning models have also been applied to recommender systems, defining them as multi-classifiers that determine which item a user will choose to consume from the system's available items in a given scenario and time. Regardless of whether it's content-based filtering, collaborative filtering, or deep learning model-based methods, there is still significant room for improvement in the diversity and accuracy of content recommendations. Summary of the Invention

[0005] To address the problems of existing technologies and improve the diversity and accuracy of content recommendations, this invention provides a method, apparatus, medium, and device for item recommendation. The technical solution is as follows:

[0006] In a first aspect, the present invention provides a method for recommending items, the method comprising:

[0007] A first association result is obtained based on external information resources, and the first association result represents the association relationship between various tags;

[0008] A second association result is obtained based on the user's historical operation data, which represents the association between the user and the item; and a third association result is obtained based on the item index data, which represents the association between the item and the item's tag.

[0009] A set of meta-path sequences is obtained based on the first association result, the second association result, and the third association result;

[0010] The machine learning model is trained based on the set of meta-path sequences to obtain the vector embedding representation of the item;

[0011] The fourth association result between the items is calculated based on the vector embedding representation of the items;

[0012] Based on the target user's historical operation data and the fourth association result, target items are identified among the items to be recommended to the target user.

[0013] Secondly, the present invention provides an item recommendation device, the device comprising:

[0014] The first calculation module is used to obtain a first association result based on external information resources, wherein the first association result represents the association relationship between various tags;

[0015] The second calculation module is used to obtain a second association result based on the user's historical operation data, the second association result representing the association relationship between the user and the item, and to obtain a third association result based on the item index data, the third association result representing the association relationship between the item and the item's tag;

[0016] A path sequence generation module is used to obtain a set of meta-path sequences based on the first association result, the second association result, and the third association result;

[0017] The vector representation module is used to train the machine learning model based on the set of meta-path sequences to obtain the vector embedding representation of the item;

[0018] The third calculation module is used to calculate the fourth association result between the items based on the vector embedding representation of the items;

[0019] The item identification module is used to identify target items from the items based on the target user's historical operation data and the fourth association result, so as to recommend them to the target user.

[0020] Thirdly, the present invention provides a computer-readable storage medium storing at least one instruction or at least one program, which is loaded and executed by a processor to implement an item recommendation method as described in the first aspect.

[0021] Fourthly, the present invention provides a computer device comprising a processor and a memory, the memory storing at least one instruction or at least one program, the at least one instruction or at least one program being loaded and executed by the processor to implement an item recommendation method as described in the first aspect.

[0022] Fifthly, the present invention provides a computer program product or computer program comprising computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the item recommendation method provided in the first aspect above.

[0023] The present invention provides a method, apparatus, device, and storage medium for recommending items, which has the following technical effects:

[0024] (1) The solution provided by the present invention introduces external information resources as auxiliary information in the recommendation system. With richer item information data, the relationship between items can be accurately described, thereby improving the relevance of item recommendations.

[0025] (2) The solution provided by the present invention constructs the association between tags based on external information resources, which can further calculate the similarity between items, improve the diversity of item recommendations, and ensure the accuracy of item recommendations.

[0026] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0027] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0028] Figure 1 This is a schematic diagram of the implementation environment of an item recommendation method provided in an embodiment of the present invention;

[0029] Figure 2 This is a flowchart illustrating an item recommendation method provided in an embodiment of the present invention;

[0030] Figure 3 This is a flowchart illustrating how an embodiment of the present invention obtains a first association result between multiple tags based on external information resources;

[0031] Figure 4 This is a schematic diagram of entity data in a knowledge graph provided by an embodiment of the present invention;

[0032] Figure 5 This is a schematic diagram of the structure of a jump character model provided in an embodiment of the present invention;

[0033] Figure 6 This is a schematic diagram of another process for obtaining the first association result between multiple tags based on external information resources, provided by an embodiment of the present invention;

[0034] Figure 7 This is a flowchart illustrating how a second and third association results are obtained based on a user's historical operation data and item index data, according to an embodiment of the present invention.

[0035] Figure 8 This is a schematic diagram of a process for training a machine learning model based on the set of meta-path sequences to obtain the vector embedding representation of the item, provided by an embodiment of the present invention.

[0036] Figure 9 This is a schematic diagram of an item recommendation device provided in an embodiment of the present invention.

[0037] Figure 10 This is a schematic diagram of the hardware structure of a device for implementing an item recommendation method provided in an embodiment of the present invention. Detailed Implementation

[0038] Artificial Intelligence (AI) is the theory, methods, technology, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to achieve optimal results. In other words, AI is a comprehensive technology within computer science that attempts to understand the essence of intelligence and produce a new kind of intelligent machine that can react in a way similar to human intelligence. AI studies the design principles and implementation methods of various intelligent machines, enabling them to have perception, reasoning, and decision-making capabilities. AI technology is a comprehensive discipline involving a wide range of fields, encompassing both hardware and software technologies. Fundamental AI technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing technology, operating / interactive systems, and mechatronics.

[0039] The solutions provided in this application involve artificial intelligence technologies such as machine learning (ML) and natural language processing (NLP).

[0040] Machine learning (ML) is a multidisciplinary field involving probability theory, statistics, approximation theory, convex analysis, and algorithm complexity theory. It specifically studies how computers can simulate or implement human learning behavior to acquire new knowledge or skills and reorganize existing knowledge structures to continuously improve their performance. Machine learning is the core of artificial intelligence and the fundamental way to endow computers with intelligence; its applications span all areas of artificial intelligence. Machine learning and deep learning typically include techniques such as artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and instructional learning.

[0041] Natural Language Processing (NLP) is an important field within computer science and artificial intelligence. It studies the theories and methods for enabling effective communication between humans and computers using natural language. NLP is a science that integrates linguistics, computer science, and mathematics. Therefore, research in this field involves natural language—the language people use in daily life—and thus it is closely related to linguistic research. NLP technologies typically include text processing, semantic understanding, machine translation, question answering, and knowledge graphs. Among these, a knowledge graph, known in library and information science as knowledge domain visualization or a knowledge domain mapping map, is a series of various graphs that display the development process and structural relationships of knowledge. It uses visualization techniques to describe knowledge resources and their carriers, mining, analyzing, constructing, drawing, and displaying knowledge and the relationships between them. Essentially, a knowledge graph is a semantic network that reveals the relationships between entities. In other words, a knowledge graph is composed of knowledge entries, each represented as an SPO (Subject-Predicate-Object) triple. The architecture of a knowledge graph mainly includes its logical structure and system architecture. Logically, a knowledge graph can be divided into two layers: a schema layer and a data layer. The data layer primarily consists of a series of facts, and knowledge is stored in units of facts. If facts are expressed using triples such as (entity 1, relation, entity 2) or (entity, attribute, attribute value), a graph database can be chosen as the storage medium. The schema layer is built upon the data layer and mainly uses an ontology library to standardize the representation of facts in the data layer. An ontology is a conceptual template for a structured knowledge base; a knowledge base formed through an ontology library not only has a strong hierarchical structure but also low redundancy.

[0042] The solutions provided in this application also relate to cloud technology, cloud computing, databases, and big data.

[0043] Cloud technology refers to a hosting technology that unifies hardware, software, and network resources within a wide area network (WAN) or local area network (LAN) to achieve data computation, storage, processing, and sharing. It can also be understood as a general term for network technologies, information technologies, integration technologies, management platform technologies, and application technologies based on cloud computing business models. These technologies can form resource pools, allowing for on-demand use and flexibility. Backend services of cloud computing systems require substantial computing and storage resources, such as video websites, image websites, and many portal websites. With the rapid development and application of the internet industry, every item may have its own identification mark in the future, requiring transmission to backend systems for logical processing. Data at different levels will be processed separately, and various industry data require robust system support; therefore, cloud technology relies on cloud computing as its foundation. Cloud computing is a computing model that distributes computing tasks across a resource pool composed of numerous computers, enabling various application systems to obtain computing power, storage space, and information services as needed. The network providing these resources is called the "cloud." From the user's perspective, resources in the "cloud" are infinitely scalable, readily available, and can be used on demand, expanded at any time, and paid for based on usage. As a provider of fundamental cloud computing capabilities, a cloud resource pool platform, often referred to as a cloud platform or Infrastructure as a Service (IaaS), is established. This platform deploys various types of virtual resources within the resource pool for external customers to choose from. The cloud resource pool primarily includes: computing devices (which can be virtualized machines containing operating systems), storage devices, and network devices.

[0044] A database, simply put, can be viewed as an electronic filing cabinet—a place to store electronic files, where users can perform operations such as adding, querying, updating, and deleting data. A "database" is a collection of data stored together in a certain way, shared by multiple users, with minimal redundancy, and independent of application programs. A Database Management System (DBMS) is a computer software system designed to manage databases, generally possessing basic functions such as storage, retrieval, security, and backup. DBMSs can be classified according to the database model they support, such as relational or XML (Extensible Markup Language); or according to the type of computer they support, such as server clusters or mobile phones; or according to the query language used, such as SQL (Structured Query Language) or XQuery; or according to performance priorities, such as maximum scale or maximum operating speed; or other classification methods. Regardless of the classification method used, some DBMSs can cross categories, for example, supporting multiple query languages ​​simultaneously.

[0045] Big data refers to data sets that cannot be captured, managed, and processed within a certain timeframe using conventional software tools. It represents massive, rapidly growing, and diverse information assets that require new processing models to achieve stronger decision-making, insightful discovery, and process optimization capabilities. With the advent of the cloud era, big data has attracted increasing attention. Big data requires specialized technologies to effectively process large amounts of data within a tolerable timeframe. Technologies suitable for big data include massively parallel processing databases, data mining, distributed file systems, distributed databases, cloud computing platforms, the internet, and scalable storage systems.

[0046] Currently, in recommendation systems, common recommendation algorithms use collaborative filtering to recommend content, generally divided into two methods: item-based collaborative filtering and user-based collaborative filtering. The former primarily obtains an item feature matrix based on the historical clicks of items, calculates the similarity between items based on this matrix, and then recommends other items to the user based on the similarity between items, according to the user's historical clicks. The latter primarily obtains a user feature matrix based on the user's historical click data, calculates the similarity between users based on this matrix, and then selects items clicked by similar users but not by the target user for recommendation. The former method is prone to item clustering; that is, for popular items, many users will click on them, potentially resulting in many items with high similarity, leading to low coverage and diversity of recommended items. The latter method suffers from data sparsity and scalability issues. Because the number of videos clicked by users is relatively small, the data is sparse, resulting in low accuracy in calculating user similarity, leading to low recommendation accuracy and weak generalization performance.

[0047] To improve the diversity and accuracy of item recommendations, embodiments of the present invention provide an item recommendation method, apparatus, medium, and device. The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout.

[0048] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or server that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.

[0049] To facilitate understanding of the technical solutions and their effects described in the embodiments of the present invention, the relevant technical terms are explained in the embodiments of the present invention:

[0050] Forward index: Also known as a forward index, such as in a video search engine where each video corresponds to a number, and the video content is represented as a series of tags to facilitate quick identification of the video content by number.

[0051] Meta path: A path in a network graph that links two objects. For example, the path "Name A (husband relationship) Name B (movie) Movie C" is a meta path that connects "Name A" and "Movie C".

[0052] Random walk, also known as a random stroll, refers to a phenomenon where past performance cannot predict future steps or directions. The core concept is that the conserved quantities carried by any random walker correspond to a diffusion transport law, approximating Brownian motion—the ideal mathematical state of Brownian motion. Currently, it is primarily applied in internet link analysis and financial stock markets. In mathematical statistical models, it is represented by a series of trajectories, where each step is random.

[0053] Collaborative filtering (CF) predicts content that a user might be interested in by analyzing the similarity between users or items and recommends that content to the user.

[0054] Word embedding: A general term for models that vectorize words. The core idea is to map each word to a dense vector in a low-dimensional space. For example, Word2vec (word vectorization) uses an N-gram model, which assumes that a word is only related to its surrounding n words and not to other words in the text. It is a group of related models used to generate word vectors. These models are shallow, two-layer neural networks used to train and reconstruct linguistic word text. The network represents words and needs to guess the input words in adjacent positions. After training, the Word2vec model can be used to map each word to a vector, which can be used to represent the relationship between words.

[0055] Skip-gram, or word-skipping algorithm, predicts the context words given a target word. The basic form of the Skip-gram model is very simple; it actually consists of two parts: building the model and obtaining embedded word vectors. The entire modeling process is similar to the idea of ​​an autoencoder, where a neural network is first built based on training data. Once trained, this model isn't used to handle new tasks; what's truly needed are the parameters learned from the training data, such as the weight matrices of the hidden layers.

[0056] It should be noted that, for the user's historical operation data, behavior data, etc. in the embodiments of this application, when the embodiments of this application are applied to specific products or technologies, the user's permission or consent is required, and the collection, use and processing of the relevant data must comply with the relevant international and regional laws, regulations and standards.

[0057] Please see Figure 2 This is a schematic diagram illustrating the implementation environment of an item recommendation method provided in this application embodiment, such as... Figure 2 As shown, the implementation environment may include at least client 01 and server 02.

[0058] Specifically, the client 01 may include devices such as smartphones, desktop computers, tablets, laptops, digital assistants, smart wearable devices, monitoring devices, and voice interaction devices. It may also include software running on the device, such as web pages provided to users by service providers, or applications provided by those service providers. Specifically, the client 01 can be used to acquire the user's historical operation data and send it to the server via the network, as well as display target items sent by the server 02 to complete item recommendations for the target user.

[0059] Specifically, the server 02 can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms. The server 02 may include a network communication unit, a processor, and a memory, etc. The terminal and the server can be directly or indirectly connected via wired or wireless communication, which is not limited herein. Specifically, the server 02 can be used to calculate the association results between tags based on external information resources, train a machine learning model based on meta-path sequences to obtain word embedding representations, and calculate and determine target items to recommend to the target user based on the target user's historical operation data. The server 02 can connect to a database via a network to obtain external information resources.

[0060] This invention can also be implemented using cloud technology. Cloud technology refers to a hosting technology that unifies hardware, software, and network resources within a wide area network (WAN) or local area network (LAN) to achieve data computation, storage, processing, and sharing. It can also be understood as a general term for network technologies, information technologies, integration technologies, management platform technologies, and application technologies based on cloud computing business models. Cloud technology requires cloud computing as its support. Cloud computing is a computing model that distributes computing tasks across a resource pool composed of a large number of computers, enabling various application systems to obtain computing power, storage space, and information services as needed. The network providing these resources is called the "cloud." Specifically, the server 02 and the database are located in the cloud. The server 02 can be a physical machine or a virtualized machine.

[0061] The following describes a method for recommending items provided by this invention. Figure 2 This is a flowchart illustrating an item recommendation method provided in an embodiment of this specification. This specification provides the operational steps of the method described in the embodiments or flowchart, but based on conventional or non-inventive labor, more or fewer operational steps may be included. The order of steps listed in the embodiments is merely one possible execution order among many and does not represent the only possible execution order. In actual system or server product execution, the method can be executed sequentially according to the embodiments or accompanying drawings, or in parallel (e.g., in a parallel processor or multi-threaded processing environment). Please refer to... Figure 2 The item recommendation method provided in the embodiments of this specification may include the following steps:

[0062] S101: Obtain a first association result based on external information resources. The first association result represents the association relationship between various tags.

[0063] Understandably, in the application field of content recommendation, the "item" in the recommendation algorithm can refer to any content object such as videos, articles, products, and news, or it can be referred to as a project. Tags are used to characterize the features of the item. For example, in a video website, the item is a video, and the tag is a key phrase that reflects the video content. It can be an abstract concept word such as: comedy, youth, or a concrete entity word such as: TV series name D, lead actor E, etc.

[0064] Understandably, for a recommender system to provide truly useful recommendations to users, efficient and accurate recommendation technology is crucial; that is, the characteristics and potential of recommendation filtering technology are paramount. In recommender systems, recommendation filtering technology mainly includes content-based filtering and collaborative filtering. Content-based filtering (CBF) is a domain-dependent algorithm that emphasizes the analysis of item attributes when generating predictions. Its main drawback is that it requires the system to have a sufficiently deep understanding of the features of the item content; in other words, CBF relies on item metadata. Before recommending to users, the system needs rich item descriptions and complete operational data. Therefore, the effectiveness of CBF depends on the availability of descriptive data, making the use of external knowledge a critical issue that CBF urgently needs to address. Another serious problem facing CBF is overly specialized content, which manifests as users only receiving recommendation results similar to items in their own profiles. Collaborative filtering (CF) technology includes two methods: item-based collaborative filtering and user-based collaborative filtering. The former method primarily uses historical click data to obtain an item feature matrix, calculates the similarity between items based on this matrix, and then recommends other items to the user based on the similarity between items, according to the user's historical clicks. The latter method primarily uses historical click data to obtain a user feature matrix, calculates the similarity between users based on this matrix, and then selects items clicked by similar users but not by the target user for recommendation. The former method is prone to item clustering, meaning that for popular items, many items may have high similarity to them, resulting in low coverage and diversity of recommended items. The latter method suffers from data sparsity and scalability issues. Due to the relatively small number of videos clicked by users, the data is sparse, leading to low accuracy in calculating user similarity, resulting in low recommendation accuracy and weak generalization performance.

[0065] In the embodiments of this specification, the item recommendation method provided by the present invention introduces external information resources to calculate the similarity between items, which can be regarded as an innovation of content-based filtering technology. At the same time, the method provided by the present invention constructs the first association result of features (tags) based on external information resources and then calculates the similarity between items to determine the target items for recommendation, which can also be regarded as a kind of collaborative filtering. Alternatively, the method provided by the present invention can be regarded as a combination of two filtering technologies.

[0066] In the embodiments of this specification, the external information resource can be structured, semi-structured, or unstructured data, collected from servers or user terminals through various methods such as data mining and web crawling. Simultaneously, big data technologies, including massively parallel processing databases, data mining, distributed file systems, distributed databases, cloud computing platforms, and scalable storage systems, can be used to store or process the external information resource. When applied to recommendation algorithms, the external information resource contains rich descriptions of item content, item attributes, and relationships between items.

[0067] In one embodiment of this specification, specifically, as shown... Figure 3 As shown, step S101 provided in this embodiment may include the following steps:

[0068] S201: Obtain entity data from the knowledge graph, wherein the entity data includes the entity's attributes and relationships.

[0069] In one feasible implementation, a knowledge graph is used as an external information resource. A knowledge graph is essentially a semantic network that describes objective things in the form of a graph. Here, "graph" refers to a graph in data structures, composed of nodes and edges, which is the true meaning of a knowledge graph. Nodes in a knowledge graph represent concepts and entities. Concepts are abstract things, while entities are concrete things, which can also be considered as distinguishable and independently existing transactions in the real world, such as names of people, places, or games. Edges represent relationships and attributes; internal characteristics of things are represented by attributes, and external connections are represented by relationships. Generally, to simplify the description of a knowledge graph, entities and concepts are collectively referred to as entities, and relationships and attributes are collectively referred to as relationships. In other words, a knowledge graph describes entities and the relationships between them. Entities can be people, places, organizations, concepts, etc., while relationships are even more diverse, including relationships between people, relationships between people and organizations, relationships between concepts and objects, and so on. For example, "Name B" is an entity in the knowledge graph, and "Name A" is another entity in the knowledge graph. "Name B" and "Name A" can be linked through a "spouse" relationship, and "Name B-spouse-Name A" is also called a triple. Because the knowledge graph contains different types of entities, it can also be regarded as a heterogeneous information network (HIN).

[0070] Understandably, the construction and application of large-scale knowledge bases require the support of various intelligent information processing technologies. Knowledge extraction techniques can extract knowledge elements such as entities, relationships, and attributes from publicly available semi-structured and unstructured data. Knowledge fusion can eliminate ambiguities between entities, relationships, attributes, and other referential terms and factual objects, forming a high-quality knowledge base. Knowledge reasoning further mines implicit knowledge based on the existing knowledge base, thereby enriching and expanding it. The comprehensive vector formed by distributed knowledge representation is of great significance for the construction, reasoning, fusion, and application of knowledge bases.

[0071] Specifically, such as Figure 4 As shown, each entity contains a set of attributes, typically including entity type, entity name, and entity popularity. The knowledge graph is composed of triples, which can be structured as head entity-relation name-tail entity. For example, in video and article recommendation applications, entities with high popularity attribute values, such as people, movies, TV series, and variety shows, are mainly extracted, and their attribute values ​​and relationships are used as entity data.

[0072] S203: Perform a random walk based on the entity data to obtain a set of entity sequences.

[0073] As can be understood, a random walk involves repeatedly and randomly selecting paths on a network to eventually form a path that traverses the network. Starting from a specific endpoint, each step of the walk involves randomly selecting an edge from the edges connected to the current node and moving along that edge to the next vertex, repeating this process continuously.

[0074] Specifically, random walks are performed on the selected knowledge graph data based on the attributes or relationships between entities to form a set of entity sequences. Random walks can be performed through first-degree relationships between entities. For example: Actor A – Appearance – TV Series A – Actor – Actor B – Partner – Actor C, then Actor A, TV Series A, Actor B, and Actor C form a sequence. Random walks can also be performed through the superordinate terms of entities (which can also be considered as the attributes of entities). For example: Actor B – Actress – Singer D – Singer – Singer E, then Actor B, Singer D, and Singer E form a sequence of entities.

[0075] In the embodiments of this specification, the length of the entity sequence is not limited and can be selected according to the actual application scenario and practical experience. When the sequence length is fixed, it is also called a truncated random walk, which will not be elaborated here.

[0076] S205: Train the word vector model based on the entity sequence set to obtain the vector embedding representation of the entity.

[0077] It is understandable that each entity sequence in the aforementioned entity sequence set can be considered a set of text content. In the field of natural language processing, statistical language models are the foundation of all natural language processing techniques and are widely used in tasks such as text processing, speech recognition, machine translation, word segmentation, part-of-speech tagging, and information retrieval. Simply put, a statistical language model is a probabilistic model used to calculate the probability of a sentence, usually built based on a corpus. When building a statistical language model based on neural networks, the text is converted into numerical tensors, i.e., word vectorization. There are two main methods of word vectorization: one is one-hot encoding, which cannot reflect the correlation between words when there are many words and high dimensionality; the other is word embedding, which refers to converting words in text into numerical vectors. In order to analyze words in text using standard machine learning algorithms, these words first need to be converted into numerical vectors as input. The process of word embedding is to embed a high-dimensional space with the number of words into a continuous vector space with a much lower dimension. Each word or phrase is mapped to a vector in the real number field, and the result of word embedding is the generation of word vectors. For example, such as Figure 5 As shown, the skip-gram model can include an input layer, a prediction layer, and an output layer. Simply put, the skip-gram model predicts the context words given a target word (center word). When the window size is 2, the context words are the two words to the left and right of the center word. In the embodiments of this specification, a neural network model with a single hidden layer is constructed using the skip-gram algorithm as a word vector model. The model is trained on entity sequences generated by random walks, but it is not used for the training task of predicting context; instead, it is used to learn the weight matrix of the hidden layer. In a feasible implementation, when the dimension of the weight matrix is ​​too large, negative sampling techniques can be used to reduce the computational burden of the training process.

[0078] S207: Obtain the association results between the entities based on the vector embedding representation of the entities.

[0079] Specifically, for the vector embedding representations of all entities, the similarity between each pair is calculated, and the resulting similarity matrix is ​​used as the association result between the entities. In some feasible implementations, the similarity calculation may be based on the Pearson Correlation Coefficient, Euclidean Distance, Cosine Similarity, or Tanimoto Coefficient; however, this specification does not limit the specific implementation of these methods.

[0080] S209: Map the association results between the entities based on the entity data to obtain a first association result representing the association relationship between each label.

[0081] In the embodiments of this specification, the label can be an abstract conceptual word or a concrete entity word, and can be considered as the attribute values ​​of the entity.

[0082] In one feasible implementation, entities are mapped to tags based on attributes such as name and alias, resulting in a tag-to-tag similarity matrix. This similarity matrix is ​​then used as the first association result. For example, an entity in a knowledge graph is essentially a numbered identifier. Based on entity data, the similarity matrix between entities is mapped to a specific tag similarity matrix, constructing a feature-to-feature similarity matrix. In the application scenario of article recommendation, tags are the features of the article; they can be the title, the type of article, or the content text. For example, in the association results between entities, the entity name is only represented as "person's name B." When calculating the similarity of articles, the similarity between two articles is determined by whether the name "person's name B" appears in the articles. If other attributes of the entity are added, such as the alias "Little B" for "person's name B," then when calculating the similarity of articles, it will not only determine whether "person's name B" appears in both, but also whether "Little B" appears in both. Increasing the attributes of the entity allows for a more accurate and comprehensive calculation of the similarity between items.

[0083] In another embodiment of this specification, specifically, as Figure 6 As shown, step S101 provided in the embodiments of this specification may further include the following steps:

[0084] S202: Obtain external information resources.

[0085] In one feasible implementation, raw data is obtained from a database, or data is obtained from a cloud database using cloud technology as an external information resource for the recommendation system. Information resources cover various types, including but not limited to videos, books, audio, public accounts, and news.

[0086] S204: Perform feature extraction based on the external information resources to obtain a feature matrix.

[0087] Specifically, based on the content characteristics of the items described in the external information resources, feature extraction is performed to construct a content profile of the items. Taking movies as an example, a movie may contain elements such as romance, art, action, science fiction, etc., and be rated from 1 to 5 points depending on the degree of inclusion. The higher the degree of inclusion, the higher the score. A matrix composed of several movie content features can be called a content feature matrix.

[0088] S206: By performing matrix decomposition and linear regression on the feature matrix, a first association result representing the association relationship between the various labels is obtained.

[0089] Specifically, in the aforementioned content feature matrix, a particular item may only be represented in a few feature dimensions, while remaining unrepresented in most, resulting in a sparse content feature matrix. Furthermore, some features in the content feature matrix have low relevance, and others have limited impact on machine learning. A key idea in machine learning, such as deep learning, is to enable the machine to select features itself. This can be achieved by reducing the matrix dimensionality through methods like matrix factorization or sparse linear regression, extracting effective features, and ultimately constructing the first association between labels (features).

[0090] Understandably, the first association result between the constructed tags represents the relationship between the content features of the items. In applications used for item recommendation, it can not only ensure the relevance between the recommended items and the items clicked by the user in the past, but also achieve the diversification of item recommendations, recommending items of different types to the user that meet a certain degree of content relevance, such as associating videos with similar books for recommendation.

[0091] S103: Obtain a second association result based on the user's historical operation data, the second association result representing the association relationship between the user and the item; and obtain a third association result based on the item index data, the third association result representing the association relationship between the item and the item's tag.

[0092] In one feasible implementation, specifically, such as Figure 7 As shown, step S103 provided in the embodiments of this specification may include the following steps:

[0093] S301: Obtain the user's historical operation data, which includes the user's identifier, item identifier, and operation information.

[0094] In the embodiments of this specification, historical operation data of users can be collected from the user's terminal or terminal application. This historical operation data can be explicit or implicit. Explicit data refers to data actively input by the user, such as comments, likes, shares, and downloads of items or content. Implicit data refers to the user's browsing history, reading time, viewing records, search logs, etc. Generally, the application backend creates a dataset for each user who uses the application product or visits the application site. User behavior data can be obtained through user behavior logs on the site and used as the user's historical operation data. This data can include, but is not limited to, user identifiers, item identifiers, and operation information. Items are the objects on which the user performs operations, and operation information can be specific information such as comment content, like information, and viewing time.

[0095] S303: Generate a first information matrix between the user and the item based on the user's historical operation data, and use the first information matrix as a second association result.

[0096] In the embodiments of this specification, the first information matrix represents the association between users and items. This association can be in matrix form, where rows represent user identifiers, columns represent item identifiers, and values ​​represent user click counts or viewing durations. Alternatively, it can be a weighted value calculated based on various behavioral types such as comments, likes, shares, downloads, views, and watchtimes to characterize the association between a user and an item. This embodiment of the specification does not limit the definition and calculation method of the values ​​in the matrix. In other feasible implementations, the mathematical representation of the association result is not limited to a matrix, and will not be elaborated here.

[0097] In one feasible implementation, video click logs of users within a recent period, such as 3 days, are extracted to generate a user-video matrix, where rows represent users, columns represent videos, and values ​​represent the number of times a user clicks.

[0098] S305: Obtain the forward index data of the item, the forward index data including the item's identifier and the item's tag.

[0099] Understandably, search engines create indexes after crawling data to facilitate retrieval. There are two main types of indexes: forward indexes and inverted indexes. The structure of a forward index (also called a forward index) is as follows:

[0100] "Document 1" ID > Word 1: Number of occurrences, List of occurrence locations; Word 2: Number of occurrences, List of occurrence locations; ...

[0101] The ID of "Document 2" > List of keywords appearing in this document.

[0102] In other words, in a search engine, each item corresponds to a unique number (ID), and the item content is represented as a series of tags to facilitate quick identification of the item content by ID. An inverted index does the opposite, converting the mapping from item ID to tag to a mapping from tag to item ID. Each tag corresponds to a series of items, and these items all exhibit this tag in their characteristics.

[0103] In the embodiments of this specification, the forward index data of the items is used to obtain the second information matrix of items and tags. If the search engine uses an inverted index, the second information matrix of items and tags can be obtained by performing simple operations such as matrix inverse transformation and transpose on the information matrix of tags and items.

[0104] S307: Generate a second information matrix of the item and the item's tag based on the forward index data, and use the second information matrix as the third association result.

[0105] In the embodiments of this specification, the second information matrix represents the association between items and tags, where rows represent documents (items), columns represent tags, and values ​​represent confidence levels. In the application of article recommendation, the confidence level can be based on the probability value of article tags obtained through mining and feature extraction of article content, and the probability value is used as the confidence level.

[0106] S105: Obtain a set of meta-path sequences based on the first association result, the second association result, and the third association result.

[0107] As can be understood from the above content of the embodiments of this specification, the first association result represents the association relationship between tags and can be represented by a tag-tag similarity matrix; the second association result represents the association relationship between users and items and can be represented by a user-item matrix; and the third association result represents the association relationship between items and tags and can be represented by an item-tag matrix obtained by forward indexing.

[0108] In one feasible implementation, step S105 provided in the embodiments of the present invention may specifically include the following steps:

[0109] S501: Filter the second association result and the third association result according to the exposure frequency of the item and the operation frequency of the user to obtain the target second association result and the target third association result.

[0110] In one embodiment of this specification, items with excessively high and low exposure counts are removed, and users with excessively high and low click counts are removed to obtain the second and third target association results. The number threshold can be set according to the actual application scenario and practical experience, and this embodiment of the specification does not specifically limit it.

[0111] In another embodiment of this specification, for the first association result, the second association result, and the third association result, text detection is performed on the values ​​to determine whether they are sensitive words. If they are sensitive words, the user, item, or tag is filtered out to obtain the target first association result, the target second association result, and the target third association result.

[0112] S503: Based on the preset node type of the metapath, perform a random walk on the first association result, the second association result of the target, and the third association result of the target to obtain a set of metapath sequences.

[0113] Understandably, the concept of Meta Path was proposed to address similarity search in heterogeneous networks. A MetaPath is a path containing a sequence of relationships, defined between objects of different types, where "meta" represents the type of a node. The set of MetaPath sequences obtained through random walks can also be considered as constituting a heterogeneous information network.

[0114] For example, in one embodiment of this specification, the node type of the preset metapath can be: item-user-item-tag-tag. The length and node type of the metapath can be set according to the actual application task and practical experience, and this embodiment does not impose specific limitations. A random walk is performed in the first association result, the target second association result, and the target third association result according to the order of the node types. Taking the walk order of item-user-item-tag-tag as an example, the "item-user" step can be based on the matrix obtained by transposing the user-item matrix representing the second association result or the target second association result; the "user-item" step can be based on the user-item matrix representing the second association result or the target second association result; the "item-tag" step can be based on the item-tag matrix representing the third association result or the target third association result; and the "tag-tag" step can be based on the tag-tag matrix representing the first association result.

[0115] S107: Train the machine learning model based on the set of meta-path sequences to obtain the vector embedding representation of the item.

[0116] It is understood that each group of meta-path sequences in the meta-path sequence combination can also be considered as a group of text content. As described in step S205 of the embodiments of this specification, when constructing a statistical language model based on a neural network, the text is converted into a numerical tensor, i.e., word vectorization. After the model is trained, it will not be used to process new tasks. What is really needed is the parameters learned by the model through the training data, such as the weight matrix of the hidden layer, in order to obtain the vector representation of the text.

[0117] In one feasible implementation, specifically, such as Figure 8 As shown, step S107 provided in the embodiments of this specification may include the following steps:

[0118] S701: Building machine learning models based on the jump word algorithm.

[0119] Specifically, the skip-gram algorithm is used to construct machine learning models, such as neural network models, with a single hidden layer (or hidden layer) as word vector models.

[0120] S703: Use each metapath sequence in the metapath sequence set as a training sample to train the machine learning model to obtain a word vector model. In the training samples, items are input samples, and metapath sequences with items as starting nodes are output samples.

[0121] Specifically, when training the model, the sliding window size for training the model needs to be determined first. Training sample pairs are then obtained based on the determined sliding window size, with each pair including an input sample and an output sample. The model is trained based on these training sample pairs to obtain the parameters of the hidden layers. The training objective of the model is to learn the distribution of word representation vectors, and its optimization objective is to maximize the likelihood function given the vectors of the features. Specifically, to calculate the similarity between items, the text representing the items is used as the input sample, and the meta-path sequences starting from the items in the meta-path sequence set are used as the output samples. The output layer outputs the probability of other nodes appearing before and after the item. In this embodiment, it is not necessary to obtain the final probability output value; instead, the aim is to obtain the weight matrix of the hidden layers in the middle of the word vector model. In a feasible implementation, when the dimension of the weight matrix is ​​too large, negative sampling techniques can be used to reduce the computational burden of the training process.

[0122] S705: Obtain the vector embedding representation of the item based on the output of the hidden layer in the word vector model.

[0123] Specifically, after building and training the word vector model, the vector corresponding to each item can be obtained using the trained word vector model.

[0124] Understandably, in order to calculate the similarity between items, the vector embedding representation of the features of the items is extracted from the output of the hidden layer. In other feasible implementations, such as to calculate the similarity between users, the text representing the user can be used as the input sample, the meta-path sequence starting from the user can be used as the output sample, and finally the vector embedding representation of the user's features can be extracted.

[0125] S109: Calculate the fourth association result between the items based on the vector embedding representation of the items.

[0126] In one feasible implementation, specifically, step S109 provided in the embodiments of this specification may include the following steps:

[0127] S901: Calculate the similarity matrix between the items based on the vector embedding representation of the items, and use the similarity matrix as the fourth association result.

[0128] Specifically, for the vector embedding representations of all items, similarity is calculated pairwise to obtain an item-item similarity matrix, which is then used as the fourth association result between the items. In some feasible implementations, the similarity calculation can be based on the Pearson Correlation Coefficient, Euclidean Distance, Cosine Similarity, or Tanimoto Coefficient; this specification does not limit the specific implementation.

[0129] S111: Based on the target user's historical operation data and the fourth association result, determine the target item among the items to recommend to the target user.

[0130] In one feasible implementation, specifically, step S111 provided in the embodiments of this specification may include the following steps:

[0131] S1101: Obtain the target user's historical operation data and determine the target user's historical items.

[0132] In the embodiments of this specification, historical operation data of the target user can be collected from the target user's terminal or terminal application. The historical operation data can be explicit or implicit data. Explicit data refers to data actively input by the target user, such as comments, likes, shares, and downloads of items or content. Implicit data refers to the target user's browsing history, reading time, viewing records, search logs, etc. The target user's behavioral data can be obtained by acquiring the target user's behavior logs and used as the target user's historical operation data. This data may include, but is not limited to, the target user identifier, historical item identifiers, and operation information. The operation information may include specific information such as comment content, like information, and viewing time. For example, the historical items that the target user has clicked can be determined based on the target user's historical operation data.

[0133] S1103: Based on the similarity matrix between the target user's historical items and the items, determine the target items that meet the preset similarity conditions with the target user's historical items, and recommend the target items to the target user as recommended items.

[0134] In the embodiments of this specification, the preset similarity condition may be a similarity threshold between the items and historical items in the similarity matrix between the items, or it may be the items ranked in the top N after sorting based on their similarity to historical items. This specification does not specifically limit this.

[0135] It is understood that the solutions provided in the embodiments of this specification introduce external information resources as auxiliary information in the recommendation system. By using richer item information data, the relationships between items can be accurately described, thereby improving the relevance of item recommendations. At the same time, the solutions provided in the embodiments of this specification construct the relationships between tags based on external information resources, which can further accurately calculate the similarity between items, thereby improving the diversity of item recommendations while ensuring the accuracy of item recommendations.

[0136] This invention also provides an item recommendation device, such as... Figure 9 As shown, the device may include:

[0137] The first calculation module 910 is used to obtain a first association result based on external information resources, wherein the first association result represents the association relationship between various tags.

[0138] The second calculation module 920 is used to obtain a second association result based on the user's historical operation data, the second association result representing the association relationship between the user and the item, and to obtain a third association result based on the item index data, the third association result representing the association relationship between the item and the item's tag.

[0139] The path sequence generation module 930 is used to obtain a set of meta-path sequences based on the first association result, the second association result, and the third association result.

[0140] The vector representation module 940 is used to train the machine learning model based on the set of meta-path sequences to obtain the vector embedding representation of the item.

[0141] The third calculation module 950 is used to calculate a fourth association result between the items based on the vector embedding representation of the items.

[0142] The item identification module 960 is used to identify target items among the items based on the target user's historical operation data and the fourth association result, so as to recommend them to the target user.

[0143] In one embodiment of this specification, the first computing module 910 may include:

[0144] The entity data acquisition unit 911 is used to acquire entity data in the knowledge graph, wherein the entity data includes the entity's attributes and relationships.

[0145] The random walk unit 912 is used to perform a random walk based on the entity data to obtain a set of entity sequences.

[0146] Vector representation unit 913 is used to train the word vector model based on the entity sequence set to obtain the vector embedding representation of the entity.

[0147] The entity association unit 914 is used to obtain the association result between the entities based on the vector embedding representation of the entities.

[0148] The tag association unit 915 is used to map the association results between the entities according to the entity data to obtain a first association result that represents the association relationship between each tag.

[0149] It should be noted that the apparatus provided in the above embodiments is only illustrated by the division of the above functional modules when implementing its functions. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. In addition, the apparatus and method embodiments provided in the above embodiments belong to the same concept, and the specific implementation process can be found in the method embodiments, which will not be repeated here.

[0150] This invention provides a computer device including a processor and a memory, the memory storing at least one instruction or at least one program, the at least one instruction or at least one program being loaded and executed by the processor to implement an item recommendation method as provided in the above method embodiments.

[0151] Figure 10 A schematic diagram of the hardware structure of a device for implementing an item recommendation method provided in an embodiment of the present invention is shown. This device may constitute or include the apparatus or system provided in the embodiments of this application. Figure 10 As shown, device 10 may include one or more processors 1002 (shown as 1002a, 1002b, ..., 1002n in the figure) 1002 (processor 1002 may include, but is not limited to, a microprocessor MCU or a programmable logic device FPGA, etc.), a memory 1004 for storing data, and a transmission device 1006 for communication functions. In addition, it may also include: a display, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of the I / O interface), a network interface, a power supply, and / or a camera. Those skilled in the art will understand that... Figure 10 The structure shown is for illustrative purposes only and does not limit the structure of the electronic device described above. For example, device 10 may also include a... Figure 10 The more or fewer components shown, or having the same Figure 10 The different configurations shown.

[0152] It should be noted that the aforementioned one or more processors 1002 and / or other data processing circuits are generally referred to herein as "data processing circuits". These data processing circuits may be embodied, in whole or in part, in software, hardware, firmware, or any other combination thereof. Furthermore, the data processing circuits may be a single, independent processing module, or may be wholly or partially integrated into any other element within device 10 (or mobile device). As involved in the embodiments of this application, the data processing circuits serve as a processor control mechanism (e.g., selection of a variable resistor termination path connected to an interface).

[0153] The memory 1004 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the method described in the embodiments of this application. The processor 1002 executes various functional applications and data processing by running the software programs and modules stored in the memory 1004, thereby realizing the above-mentioned item recommendation method. The memory 1004 may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 1004 may further include memory remotely located relative to the processor 1002, and these remote memories can be connected to the device 10 via a network. Examples of the above-mentioned networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0154] The transmission device 1006 is used to receive or send data via a network. Specific examples of the network described above may include a wireless network provided by the communication provider of device 10. In one example, the transmission device 1006 includes a network interface controller (NIC), which can connect to other network devices via a base station to communicate with the Internet. In another example, the transmission device 1006 may be a radio frequency (RF) module used for wireless communication with the Internet.

[0155] The display may be, for example, a touchscreen liquid crystal display (LCD) that allows the user to interact with the user interface of device 10 (or mobile device).

[0156] This invention also provides a computer-readable storage medium, which can be disposed in a server to store at least one instruction or at least one program related to implementing an item recommendation method in the method embodiments. The at least one instruction or the at least one program is loaded and executed by the processor to implement the item recommendation method provided in the above method embodiments.

[0157] Optionally, in this embodiment, the storage medium may be located at at least one of the multiple network servers in a computer network. Optionally, in this embodiment, the storage medium may include, but is not limited to, various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0158] This invention also provides a computer program product or computer program, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the methods provided in the various optional embodiments described above.

[0159] As can be seen from the embodiments of the article recommendation method, apparatus, medium and equipment provided by the present invention,

[0160] (1) The solution provided by the present invention introduces external information resources as auxiliary information in the recommendation system. With richer item information data, the relationship between items can be accurately described, thereby improving the relevance of item recommendations.

[0161] (2) The solution provided by the present invention constructs the association between tags based on external information resources, which can further calculate the similarity between items, improve the diversity of item recommendations, and ensure the accuracy of item recommendations.

[0162] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, specific embodiments have been described above. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps described in the claims can be performed in a different order than that shown in the embodiments and still achieve the desired result. Additionally, the processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0163] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the embodiments of apparatus, devices, and storage media are basically similar to the method embodiments, so the descriptions are relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0164] Those skilled in the art will understand that all or part of the steps of the above embodiments can be implemented by hardware or by a program instructing related hardware. The program can be stored in a computer-readable storage medium, such as a read-only memory, a disk, or an optical disk.

[0165] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An article recommendation method characterized by comprising: The method comprises: According to the entity data of the knowledge graph in the external information resource, the association results between entities are mapped to obtain a first association result, which represents the semantic association relationship between each label; the labels are derived from the entity data; the entity data includes the attributes and relationships of the entities; According to the historical operation data of the user, a second association result is obtained, which represents the association relationship between the user and the item, and according to the item index data, a third association result is obtained, which represents the association relationship between the item and the label of the item; the label of the item indicates the content corresponding to the item itself; According to the first association result, the second association result and the third association result, a meta-path sequence set is obtained; the meta-path sequence set includes multiple meta-paths starting from the item and containing a path segment of a first label-second label; the first label is the label of the item indicated by the third association result, and the second label is the entity data indicated by the first association result which has a semantic association relationship with the first label; According to the meta-path sequence set, a machine learning model is trained to obtain a word vector model, and a vector embedding representation of the item output by the word vector model is obtained; According to the vector embedding representation of the item, a fourth association result between the items is calculated; the fourth association result represents the similarity between the items; According to the historical operation data of the target user and the fourth association result, a target item is determined among the items to recommend to the target user.

2. The article recommendation method according to claim 1, wherein The method further comprises: According to the entity data, a random walk is performed to obtain an entity sequence set; According to the entity sequence set, a word vector model is trained to obtain a vector embedding representation of the entity; According to the vector embedding representation of the entity, an association result between the entities is obtained.

3. The article recommendation method of claim 1, wherein, According to the external information resource, the first association result representing the association relationship between each label further comprises: Obtain external information resources; According to the external information resources, feature extraction is performed to obtain a feature matrix; By matrix decomposition and linear regression on the feature matrix, the first association result representing the semantic association relationship between the labels is obtained.

4. The article recommendation method according to claim 1, wherein According to the historical operation data of the user, a second association result representing the association relationship between the user and the item is obtained, and according to the item index data, a third association result representing the association relationship between the item and the label of the item is obtained, which comprises: Obtain the historical operation data of the user, which includes the identification of the user, the identification of the item and the operation information; According to the historical operation data of the user, a first information matrix of the user and the item is generated, and the first information matrix is taken as a second association result; Obtain the positive index data of the item, which includes the identification of the item and the label of the item; generate a second information matrix of the item and labels of the item according to the positive index data, and take the second information matrix as a third correlation result.

5. The article recommendation method of claim 1, wherein, The obtaining of the meta-path sequence set according to the first correlation result, the second correlation result and the third correlation result includes: filtering the second correlation result and the third correlation result according to exposure frequencies of items and operation frequencies of users to obtain a target second correlation result and a target third correlation result; performing random walk on the first correlation result, the target second correlation result and the target third correlation result according to preset node types of meta-paths to obtain a meta-path sequence set.

6. The article recommendation method of claim 1, wherein, The training of the machine learning model according to the meta-path sequence set to obtain a word vector model and the obtaining of vector embedding representations of the items output by the word vector model include: constructing a machine learning model based on a skip-gram algorithm; training the machine learning model by taking each meta-path sequence in the meta-path sequence set as a training sample, taking an item as an input sample and taking a meta-path sequence with the item as a starting node as an output sample to obtain the word vector model; and obtaining vector embedding representations of the items according to outputs of hidden layers in the word vector model.

7. The article recommendation method of claim 1, wherein, The fourth correlation result between the items is calculated according to the vector embedding representations of the items, and a target item is determined from the items according to historical operation data of a target user and the fourth correlation result to recommend the target item to the target user, including: calculating a similarity matrix between the items according to the vector embedding representations of the items, and taking the similarity matrix as a fourth correlation result; obtaining historical operation data of a target user to determine historical items of the target user; determining a target item that satisfies a preset similarity condition with the historical items of the target user from the items according to the historical items of the target user and the similarity matrix between the items, and taking the target item as a recommended item of the target user to recommend the target item to the target user. 8.An article recommendation device characterized by comprising: The apparatus includes: a first calculation module configured to map correlation results between entities according to entity data of a knowledge graph in an external information resource to obtain a first correlation result, the first correlation result representing semantic correlation relationships between labels, the labels being derived from the entity data, and the entity data including attributes and relationships of the entities; a second calculation module configured to obtain a second correlation result according to historical operation data of a user, the second correlation result representing correlation relationships between the user and items, and a third correlation result according to item index data, the third correlation result representing correlation relationships between the items and labels of the items, the labels of the items indicating contents of the items themselves. The path sequence generation module is configured to obtain a meta-path sequence set according to the first association result, the second association result and the third association result; the meta-path sequence set includes a plurality of meta-paths each having a path segment starting from the item and containing a first label-second label; the first label is a label of the item indicated by the third association result, and the second label is an entity data having a semantic association relationship with the first label indicated by the first association result; The vector representation module is configured to train a word vector model according to the meta-path sequence set, obtain a word vector model, and acquire a vector embedding representation of the item output by the word vector model; The third calculation module is configured to calculate a fourth association result between the items according to the vector embedding representation of the item; the fourth association result represents a similarity between the items; The item determination module is configured to determine a target item from the items according to the historical operation data of a target user and the fourth association result, and recommend the target item to the target user.

9. The apparatus of claim 8, wherein, The first calculation module includes: The random walk unit is configured to perform random walk according to the entity data to obtain an entity sequence set; The vector representation unit is configured to train a word vector model according to the entity sequence set to obtain a vector embedding representation of the entity; The entity association unit is configured to obtain an association result between the entities according to the vector embedding representation of the entity.

10. The apparatus of claim 8, wherein, The first calculation module is configured to acquire external information resources, perform feature extraction on the external information resources to obtain a feature matrix, and perform matrix decomposition and linear regression on the feature matrix to obtain the first association result representing a semantic association relationship between labels.

11. The apparatus of claim 8, wherein, The second calculation module is configured to acquire historical operation data of a user, the historical operation data of the user including an identifier of the user, an identifier of an item and operation information; generate a first information matrix of the user and the item according to the historical operation data of the user, take the first information matrix as a second association result; acquire positive index data of the item, the positive index data including an identifier of the item and a label of the item; generate a second information matrix of the item and the label of the item according to the positive index data, and take the second information matrix as a third association result.

12. The apparatus of claim 8, wherein, The path sequence generation module is configured to filter the second association result and the third association result according to an exposure frequency of the item and an operation frequency of the user to obtain a target second association result and a target third association result; and perform random walk on the first association result, the target second association result and the target third association result according to a preset node type of a meta-path to obtain a meta-path sequence set.

13. The apparatus of claim 8, wherein, The vector representation module is configured to construct a machine learning model based on a skip-gram algorithm, train the machine learning model with each meta-path sequence in the set of meta-path sequences as a training sample, and obtain the word vector model, the item in the training sample being an input sample, and the meta-path sequence with the item as a starting node being an output sample, and obtain the vector embedding representation of the item according to an output of a hidden layer in the word vector model.

14. The apparatus of claim 8, wherein, The item determination module is configured to calculate a similarity matrix between the items according to the vector embedding representation of the items, take the similarity matrix as a fourth association result, obtain historical operation data of a target user, determine historical items of the target user, and determine, according to the historical items of the target user and the similarity matrix between the items, a target item that meets a preset similarity condition with the historical items of the target user from the items, and take the target item as a recommended item of the target user for recommendation to the target user.

15. A computer-readable storage medium, characterized in that, The computer readable storage medium stores at least one instruction or at least one program, and the at least one instruction or the at least one program is loaded and executed by the processor to implement the item recommendation method in any one of claims 1 to 7.

16. A computer device, comprising: The computer device includes a processor and a memory, and the memory stores at least one instruction or at least one program, and the at least one instruction or the at least one program is loaded and executed by the processor to implement the item recommendation method in any one of claims 1 to 7.

Images