Article recommendation method and device, electronic equipment and storage medium

Abstract

The invention relates to an article recommendation method and device, electronic equipment and a storage medium. Reasoning an association relationship set among the historical articles based on the article features of the historical articles of the historical interaction of the target object; in the incidence relation set, selecting a plurality of candidate historical articles to replace the candidate historical articles with masks, performing context analysis on the basis of article features of other historical articles associated with each mask, selecting an estimated historical article corresponding to each mask from the candidate historical articles, and storing the estimated historical articles in the candidate historical articles; when the estimated historical article is not matched with the corresponding replaced candidate historical article, deleting the replaced candidate historical article and the article feature thereof; and based on the recommended article features of the to-be-recommended article and the target article features of the remaining target historical articles, reasoning a target association relationship between the to-be-recommended article and the target historical articles, and pushing a target association reason included in the target association relationship and the to-be-recommended article to a target object. The probability that the recommended article is subjected to interaction operation by the object can be improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer, and particularly relates to an item recommendation method and device, electronic equipment and storage medium. BACKGROUND

[0002] With the development of the Internet, an object can easily obtain a required item through the Internet, which can be an article, a video, a topic, news, a commodity, etc. In order to filter the item of interest of the object from a large amount of information, a recommendation system emerges as the times require. When the recommendation system recommends an item to the object, the recommendation system usually determines an item that the object can be interested in according to historical behavior data of the object, and selects a recommended item from the item that the object can be interested in, so as to obtain a recommendation result.

[0003] However, the recommendation result output by the recommendation system often lacks explainability, and the object cannot understand why the recommendation system pushes certain recommended items, thereby lacking trust and satisfaction of the recommended items, so that the probability of the recommended items being executed by the object is relatively small, for example, the interaction operation includes but is not limited to clicking, browsing, purchasing, sharing, etc., resulting in poor recommendation effect.

[0004] Therefore, how to improve the probability of the recommended item pushed by the recommendation system being executed by the object, so as to improve the recommendation effect, is a problem to be solved. SUMMARY

[0005] Embodiments of the present application provide an item recommendation method and device, electronic equipment and storage medium, to improve the probability of the recommended item being executed by the object, so as to improve the recommendation effect.

[0006] In one aspect, the present application provides an item recommendation method, which comprises:

[0007] Based on the item features of each historical item involved in each historical interaction operation performed by a target object, an association relationship set between the historical items is inferred; each association relationship contains at least two associated historical items and an association reason;

[0008] In the association relationship set, a plurality of candidate historical items are replaced by corresponding masks respectively, and for each mask, the following operations are performed: based on the item features of other historical items associated with the mask, context semantic analysis is performed, and based on the analysis result, a predicted historical item corresponding to the mask is selected from the plurality of candidate historical items; when the predicted historical item does not match the corresponding replaced candidate historical item, the replaced candidate historical item and the corresponding item feature are deleted;

[0009] obtain a target association relationship between the to-be-recommended item and at least one target historical item based on a recommendation item feature of the to-be-recommended item and a target item feature of each target historical item that has not been deleted, and push, as recommendation information, a target association reason included in the target association relationship to the target object in combination with the to-be-recommended item.

[0010] In one aspect, an item recommendation apparatus is provided, including:

[0011] an inference unit configured to infer a set of association relationships between the historical items based on item features of each historical item involved in each historical interaction operation performed by the target object;

[0012] a verification unit configured to replace a plurality of selected candidate historical items with corresponding masks in the set of association relationships, and for each mask, perform the following operations: performing context semantic analysis based on item features of other historical items associated with the mask, and selecting, from the plurality of candidate historical items, a predicted historical item corresponding to the mask based on an analysis result, and when the predicted historical item does not match a corresponding replaced candidate historical item, deleting the replaced candidate historical item and a corresponding item feature;

[0013] a recommendation unit configured to obtain a target association relationship between the to-be-recommended item and at least one target historical item based on a recommendation item feature of the to-be-recommended item and a target item feature of each target historical item that has not been deleted, and push, as recommendation information, a target association reason included in the target association relationship to the target object in combination with the to-be-recommended item.

[0014] Optionally, the item feature of each historical item includes an item attribute of the corresponding historical item and a historical interaction type of the corresponding historical item by the target object, and the historical items are arranged in an interaction order;

[0015] The inference unit is specifically configured to:

[0016] construct first prompt information based on the item features of the historical items;

[0017] analyze, based on the item attributes and the historical interaction types of the historical items included in the first prompt information and the interaction order of the historical items, each association relationship between the historical items and an association reason included in each association relationship, to obtain the set of association relationships between the historical items.

[0018] Optionally, the inference unit is specifically configured to:

[0019] construct the first prompt information based on the respective item features of the respective historical items and a preset set of association reasons;

[0020] using the initial large language model, based on the respective item attributes and historical interaction types of the respective historical items contained in the first prompt information, and the interaction sequence of the respective historical items, analyzing the respective association relationships existing between the respective historical items, and for each association relationship obtained by the analysis, selecting a corresponding association reason from the set of association reasons to obtain a set of association relationships between the respective historical items.

[0021] Optionally, the verification unit is specifically configured to:

[0022] based on the set of association relationships containing the respective masks and the plurality of candidate historical items, constructing second prompt information;

[0023] using the initial large language model, for each mask, performing the following operations: based on the item features of the other historical items associated with one mask contained in the second prompt information, respectively, and the semantic similarity between the respective item features of the plurality of candidate historical items, selecting a candidate historical item whose semantic similarity satisfies a set condition from the plurality of candidate historical items as a pre-estimated historical item corresponding to the one mask.

[0024] Optionally, the apparatus further comprises a training unit configured to:

[0025] whenever a pre-estimated historical item corresponding to a mask matches a corresponding replaced candidate historical item, performing accuracy verification on the association relationship in the set of association relationships to which the replaced candidate historical item belongs, and taking the association relationship that passes the verification as a sample association relationship;

[0026] in a set network layer of the initial large language model, replacing the trained parameters in the set network layer with set to-be-trained parameters to obtain a to-be-trained large language model, and based on the plurality of sample association relationships obtained, performing multiple rounds of iterative training on the to-be-trained large language model to obtain a target large language model; wherein each round of training process comprises:

[0027] inputting respective sample item features of at least two sample historical items contained in a sample association relationship into the large language model to obtain a predicted association relationship, determining a loss value based on the difference between the predicted association relationship and the sample association relationship, and adjusting the to-be-trained parameters in the set network layer based on the loss value.

[0028] Optionally, the recommended item feature comprises a recommended item attribute of the to-be-recommended item and a set interaction type; and the target item feature comprises an item attribute and a historical interaction type of a corresponding target historical item.

[0029] The recommendation unit is specifically configured to:

[0030] construct third prompt information based on the recommended item feature and each target item feature;

[0031] adopt the target large language model to perform correlation analysis on the to-be-recommended item and the target historical items based on the respective item attributes and historical interaction types of the target historical items, the recommended item attribute, the set interaction type, and the interaction sequence of the target historical items, and obtain a target correlation between the to-be-recommended item and at least one target historical item.

[0032] Optionally, when the selected multiple candidate historical items are replaced by corresponding masks in the correlation set, the verification unit is specifically configured to:

[0033] for the multiple historical items included in the correlation set, perform the following operations respectively:

[0034] determine whether to select one historical item at random according to a set probability;

[0035] when it is determined that the one historical item is selected, replace the one historical item with a corresponding mask.

[0036] Optionally, the device further comprises:

[0037] a generation unit configured to generate a corresponding association logic graph based on a latest correlation set composed of the target historical items that are not deleted; the association logic graph comprises multiple nodes, each node corresponding to a target historical item, and each node being connected to at least one other node;

[0038] an extraction unit configured to perform graph feature extraction on the association logic graph to obtain an association logic graph feature;

[0039] a prediction unit configured to predict an interaction probability of the target object to the to-be-recommended item based on a preset item feature of the to-be-recommended item and the association logic graph feature through a pre-stored interaction probability prediction model.

[0040] Optionally, the node information of each node comprises a target item feature of a corresponding target historical item.

[0041] The extraction unit is specifically configured to:

[0042] The node information of each node in the association logic graph is vectorized to obtain the corresponding node representation vector;

[0043] Feature extraction is performed on the node representation vectors of each of the multiple nodes in the associated logic graph to obtain the node feature vectors of each of the multiple nodes.

[0044] The node feature vectors of the multiple nodes are fused to obtain the features of the association logic graph.

[0045] Optionally, the node information for each node may also include: the association reason included in the association relationship to which the corresponding target historical item belongs; the target item characteristics include the item attributes and historical interaction types of the corresponding target historical item;

[0046] When converting the node information of each node in the association logic graph into a corresponding node representation vector, the extraction unit is specifically used for:

[0047] The associated reason, item attribute and historical interaction type of each node are transformed into vectors to obtain the first sub-vector, the second sub-vector and the third sub-vector;

[0048] The first sub-vector, the second sub-vector, and the third sub-vector are fused together to obtain the corresponding node representation vector.

[0049] On one hand, an electronic device provided in this application includes a processor and a memory, wherein the memory stores a computer program that, when executed by the processor, causes the processor to perform the steps of any of the above-described item recommendation methods.

[0050] On one hand, embodiments of this application provide a computer-readable storage medium including a computer program, which, when run on an electronic device, causes the electronic device to perform the steps of any of the above-described item recommendation methods.

[0051] On one hand, embodiments of this application provide a computer program product, the computer program product including a computer program stored in a computer-readable storage medium; when a processor of an electronic device reads the computer program from the computer-readable storage medium, the processor executes the computer program, causing the electronic device to perform the steps of any of the above-described item recommendation methods.

[0052] The above-described solutions in this application have at least the following beneficial effects:

[0053] In this embodiment, based on the item features of each historical item in the target object's historical interactions, a set of association relationships between each historical item is obtained through reasoning. Each association relationship involves at least two related historical items and the reason for their association. To verify the accuracy of the association relationship set, multiple candidate historical items are selected from the association relationship set and replaced with corresponding masks. For each mask, contextual semantic analysis is performed based on the item features of other historical items associated with the mask. The estimated historical item corresponding to the mask is selected from the multiple candidate historical items, and then the estimated historical item is compared with the candidate historical item replaced by the mask to see if they match. If they do not match, it is considered that the association relationship between the candidate historical item replaced by the mask and other historical items cannot be accurately inferred. That is, the candidate historical item replaced by the mask is not trusted, and the candidate historical item replaced by the mask and its item features are deleted.

[0054] By validating the aforementioned set of associations, unconvincing candidate historical items can be removed, resulting in convincing target historical items. Then, after obtaining the items to be recommended, based on the recommended item characteristics of the items to be recommended and the target item characteristics of each target historical item, the target association between the items to be recommended and at least one target historical item can be accurately inferred. The reasons for the target association contained in these relationships are then used as recommendation reasons and pushed to the target audience along with the items to be recommended. In this way, the obtained recommendation reasons are relevant to the target audience's historical interactions with these items, making them persuasive. Through these reasons, the target audience can understand why the recommendation system pushed this item, thereby increasing their trust and satisfaction with the recommended item, increasing the probability that the recommended item will be interacted with, and ultimately improving the recommendation effect.

[0055] Other features and advantages of this application will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the application. The objectives and other advantages of this application may be realized and obtained by means of the structures particularly pointed out in the written description, claims, and drawings. Attached Figure Description

[0056] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0057] Figure 1 This is a schematic diagram illustrating an application scenario of an item recommendation method according to an embodiment of this application;

[0058] Figure 2 This is a flowchart of an item recommendation method according to an embodiment of this application;

[0059] Figure 3 This is a schematic diagram of an association logic diagram in an embodiment of this application;

[0060] Figure 4 This is a schematic diagram of an association logic diagram including a mask, as described in an embodiment of this application.

[0061] Figure 5 This is a schematic diagram illustrating a process of reasoning about a set of association relationships based on an initial large language model in an embodiment of this application;

[0062] Figure 6 This is a schematic diagram illustrating the transformation process of an association set in an embodiment of this application;

[0063] Figure 7 This is a schematic diagram of a mask prediction process based on an initial large language model in an embodiment of this application;

[0064] Figure 8 This is a schematic diagram illustrating the verification process of estimated historical items and candidate historical items in an embodiment of this application;

[0065] Figure 9 This is a schematic diagram illustrating the fine-tuning process of an initial large language model in an embodiment of this application;

[0066] Figure 10 This is a schematic diagram illustrating the reasoning process for recommending an item in an embodiment of this application;

[0067] Figure 11 This is a schematic diagram of an article push interface in an embodiment of this application;

[0068] Figure 12 This is a schematic diagram of the prediction process of an interactive probability prediction model in an embodiment of this application;

[0069] Figure 13 This is a schematic diagram of the composition of an item recommendation device according to an embodiment of this application;

[0070] Figure 14 This is a schematic diagram of the composition structure of an electronic device according to an embodiment of this application;

[0071] Figure 15 This is a schematic diagram of the composition structure of another electronic device using an embodiment of this application. Detailed Implementation

[0072] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings of the embodiments of this application. Obviously, the described embodiments are only some embodiments of the technical solutions of this application, and not all embodiments. Based on the embodiments recorded in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the technical solutions of this application.

[0073] The following describes some of the concepts involved in the embodiments of this application.

[0074] 1. Large Language Model (LLM): LLM is a type of natural language processing technology based on deep learning. Its main purpose is to enable machines to better understand and generate human natural language text, such as articles, dialogues, and searches.

[0075] 2. Embedding: A numerical vector composed of multiple floating-point numbers, describing the various attributes and properties of an item in a high-dimensional space. It is generally obtained by training a neural network model.

[0076] 3. Association Logic Graph: Used to describe the association relationships between multiple nodes, with each node connected to one or more other nodes. In this embodiment, each node represents a historical item, and the edge between two nodes can represent the association reason between the two historical items.

[0077] The word “exemplary” as used below means “serving as an example, embodiment, or illustration.” Any embodiment illustrated as an “exemplary” need not be construed as superior to or better than other embodiments.

[0078] The terms "first" and "second" used in this document are for descriptive purposes only and should not be construed as indicating relative importance or implying the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of the embodiments of this application, unless otherwise stated, "multiple" means two or more.

[0079] The design concept of the embodiments of this application will be briefly introduced below.

[0080] Currently, when recommending items to an object, recommendation systems typically determine items that the object might be interested in based on the object's historical behavioral data, and then select recommended items from those items to obtain the recommendation result. However, the recommendation results output by recommendation systems often lack interpretability. Objects cannot understand why the recommendation system is pushing certain recommended items, thus lacking trust and satisfaction with the recommended items. This results in a lower probability that the object will interact with the recommended items, leading to poor recommendation performance.

[0081] In related technologies, while some recommendation systems can generate reasons for recommending items, these reasons are usually based on the item's own characteristics (such as tags, attributes, keywords, etc.). For example, in movie recommendations, by analyzing the types of movies a user likes, directors, actors, etc., movies with similar characteristics can be recommended, and these characteristics can be used as reasons for the recommendations. However, for the user, reasons related to the item's own characteristics may not be persuasive enough.

[0082] In view of this, in order to generate persuasive recommendation reasons for the items to be recommended, thereby increasing the probability that the recommended items pushed by the recommendation system will be interacted with by the object, and thus improving the recommendation effect, this application provides an item recommendation method, apparatus, electronic device, and storage medium. Based on the item characteristics of each historical item interacted with by the target object, a set of association relationships between historical items is obtained through reasoning. Each association relationship involves at least two related historical items and the reasons for their association. Then, by verifying the above association relationship set, unconvincing candidate historical items are deleted from the association relationship set to obtain convincing target historical items. Afterwards, the items to be recommended are... After identifying the item, based on the recommended item characteristics of the item to be recommended and the target item characteristics of each target historical item, the target association relationship between the item to be recommended and at least one target historical item can be accurately inferred. The target association reason contained in the target association relationship is used as the recommendation reason and pushed to the target object along with the item to be recommended. In this way, the obtained recommendation reason is related to the target historical items that the target object has interacted with in the past, and is persuasive. Through this recommendation reason, the target object can understand why the recommendation system pushed the recommended item, thereby increasing the target object's trust and satisfaction with the recommended item, increasing the probability that the recommended item will be interacted with by the object, and thus improving the recommendation effect.

[0083] The preferred embodiments of this application are described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit this application. Furthermore, the embodiments and features in the embodiments of this application can be combined with each other without conflict.

[0084] like Figure 1The diagram shown is an application scenario illustration of an embodiment of this application. The application scenario diagram includes a terminal device 110 and a server 120.

[0085] In one alternative implementation, the terminal device 110 and the server 120 can communicate via a communication network; wherein the communication network can be a wired network or a wireless network.

[0086] In this embodiment, the terminal device 110 includes, but is not limited to, mobile phones, tablets, laptops, desktop computers, e-book readers, smart voice interaction devices, smart home appliances, and in-vehicle terminals. The terminal device may have a client application related to item recommendations installed. This client application can be software (e.g., browsers, video software, shopping software, music software), or it can be a webpage, mini-program, etc. The server 120 can be a backend server corresponding to the software, webpage, mini-program, etc., or a server specifically used for item recommendations; this application does not impose specific limitations. The server 120 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, content delivery networks (CDN), and big data and artificial intelligence platforms.

[0087] It should be noted that the item recommendation method in each embodiment of this application can be executed by the terminal device 110 or the server 120, or it can be executed by both the terminal device 110 and the server 120.

[0088] In some embodiments, taking server 120 as an example, a client related to item recommendation is installed in terminal device 110. The recommendation system in server 120 can recommend items to objects using the client. The target object can perform interactive operations on the recommended items in the client. For example, if the recommended item is an article, the interactive operation can be clicking, reading, sharing, etc. Server 120 can obtain all historical items involved in the target object's various historical interactions, and based on the item characteristics of each historical item, infer a set of association relationships between them. Each association relationship involves at least two related historical items and the reason for their association. Then, by verifying the above association relationship set, unconvincing candidate historical items are deleted from the association relationship set to obtain convincing target historical items. After obtaining the item to be recommended, based on the recommended item characteristics of the item to be recommended and the target item characteristics of each target historical item, the target association relationship between the item to be recommended and at least one target historical item can be accurately inferred. The target association reason contained in the target association relationship is used as the recommendation reason. The item to be recommended and the recommendation reason are sent to the client as recommendation information. The client displays the item to be recommended and the recommendation reason at the same time, so that the target object can intuitively understand why the recommendation system pushes the item to be recommended.

[0089] Product recommendation has wide applications in many fields. Here are some specific application scenarios for product recommendation and their detailed explanations:

[0090] 1. E-commerce. On e-commerce websites, recommendation systems can recommend products that users might be interested in, thereby increasing sales and user satisfaction. For example, personalized product recommendations suggest related products based on a user's browsing history, purchase history, and click behavior; cross-recommendations suggest other products related to those already purchased; new product recommendations suggest new products, especially those similar to those previously purchased; and products from subscribed stores can also be recommended.

[0091] 2. Video Streaming Services. Video streaming services use recommendation systems to recommend video content to users, aiming to increase user engagement and viewing time. For example, they may recommend similar videos based on a user's viewing history; offer trending recommendations, suggesting currently popular video content; generate personalized playlists that users might be interested in; and recommend subscribed video content.

[0092] 3. News Recommendations. News recommendation systems can recommend personalized news content to users. For example, based on a user's reading history and preferences, it can recommend relevant news articles; it can also recommend current trending topics and related news; and it can recommend subscribed news content.

[0093] 4. Music Recommendations. Music recommendation systems can enhance a user's music experience by showing them songs they are interested in. For example, they can generate personalized playlists based on a user's listening history and preferences; recommend newly released songs, especially those that match their preferences; recommend relevant concert and live performance information based on their music preferences; and even recommend subscribed music.

[0094] 5. Social Media. Social media platforms recommend content that users might be interested in through recommendation systems. For example, based on users' interests and historical interaction behavior, they can recommend posts, articles, videos, and content from public accounts that they might be interested in; they can also recommend content that users have subscribed to on social media platforms.

[0095] It should be noted that the application scenarios listed above are just simple examples. Other application scenarios are also applicable to the embodiments of this application, and will not be described in detail here.

[0096] Furthermore, it should be emphasized that in the specific embodiments of this application, the historical interaction operations, historical items, item characteristics, and other data involved in the objects involved need to obtain the permission or consent of the objects when the embodiments of this application are applied to specific products or technologies, and the collection, use, and processing of related data need to comply with the relevant laws, regulations, and standards of the relevant countries and regions.

[0097] The following describes the item recommendation method provided by exemplary embodiments of this application in conjunction with the application scenarios described above and with reference to the accompanying drawings. It should be noted that the above application scenarios are only shown to facilitate understanding of the spirit and principles of this application, and the embodiments of this application are not limited in any way in this respect.

[0098] See Figure 2 The diagram shown is an implementation flowchart of an item recommendation method provided in this application embodiment. Taking the server as the execution entity as an example, the specific implementation process of this method includes the following S21-S23:

[0099] S21. Based on the characteristics of each historical item involved in each historical interaction operation of the target object, reason to obtain the set of association relationships between each historical item; each association relationship includes: at least two related historical items and the reason for the association.

[0100] In this application embodiment, historical items include, but are not limited to: videos, information streams (e.g., articles, news), Q&A, topics, encyclopedias, posts, products, music, e-books, etc.

[0101] The server's item recommendation platform can store a large number of historical item sequences of object interactions. For each object, the historical item sequence can include all historical items involved in the object's historical interaction operations within a set historical time period, such as 100 historical items that the object has interacted with in the last 90 days. Each historical item is identified by a unique item ID (Identifier), and the historical items can be arranged in the order of interaction time.

[0102] The target object can be any object from a large number of objects, specifically a user. The historical interaction operations performed by the target object on historical items include, but are not limited to: browsing, clicking, reading (including long-term reading), playing, liking, commenting, sharing, purchasing, etc.

[0103] Each historical item's characteristics can include: item attributes and historical interaction types. Item attributes can represent the historical item, including the item ID mentioned above, and can also include other representative information. For example, if the historical item is an article, other representative information includes, but is not limited to, the article title, author (or account), and article category. Similarly, if the historical item is a video, other representative information includes, but is not limited to, the video title, video source, and video category. For example, if the historical item is a product, other representative information includes, but is not limited to, the video title, video source, and video category. Historical interaction types can be understood as the operation types of the aforementioned historical interactions. When a target object performs multiple historical interaction operations on the same historical item, it can arbitrarily select one historical interaction operation from among the multiple historical interaction operations, or it can select a specific historical interaction operation, and the operation type of the selected historical interaction operation will be used as the historical interaction type.

[0104] After obtaining the individual item features of each historical item in the target object's historical interactions, these features can be arranged according to the order of their interactions to form an item feature sequence. Then, association analysis can be performed on this sequence to infer a set of relationships between the historical items. Specifically, based on the semantic similarity between item attributes within each item feature, combined with the order of interactions and the types of historical interactions within each item feature, association analysis can be performed to infer multiple relationships between the historical items, thus obtaining a set of relationships. The historical items in this set can be some or all of the historical items from the initially obtained historical item sequence.

[0105] On the one hand, each association in the association set can include two related historical items and an association reason. For example, for related historical items 1 and 2, the target object's interaction with historical item 1 triggers the target object's interaction with historical item 2. At this time, historical item 2 corresponds to an association reason, that is, an interaction reason.

[0106] On the other hand, each association in the association set can include two or more related historical items. When an association includes two or more historical items, there are multiple association reasons. Two or more historical items can be associated sequentially to form an association chain. For example, for related historical items 1, 2, and 3, the target object's interaction with historical item 1 triggers the target object's interaction with historical item 2. At this time, historical item 2 corresponds to an association reason. The target object's interaction with historical item 2 then triggers the target object's interaction with historical item 3. At this time, historical item 3 also corresponds to an association reason.

[0107] Specifically, the set of relationships can be converted into a relationship logic graph, which includes multiple nodes. Each node represents a historical item, and the node information of each node contains the item characteristics of the corresponding historical item. Each node is connected to one or more other nodes, and the edge between each two nodes has an arrow to point from one node to another. Each edge can correspond to a relationship reason, that is, the node pointed to by the arrow of each edge also contains the relationship reason in its node information.

[0108] For example, such as Figure 3 As shown, assuming the association logic diagram includes nodes V1, V2, V3, V4, V5, V6, and V7, when each association link is considered as an association relationship, V1->V2->V6 is one association relationship, V1->V3->V4 is another association relationship, and V1->V5->V7 is yet another association relationship. In addition, each association relationship can also include two associated nodes (i.e., historical items), such as V1->V2, V2->V6, V1->V3, V3->V4, V1->V5, and V5->V7.

[0109] One implementation method is to use a pre-stored initial large language model to infer the set of relationships between historical items based on the item feature sequences of each historical item. The initial large language model possesses powerful text semantic understanding capabilities and a large amount of universally applicable knowledge, which can be used to infer the interaction logic between the target object and each historical item. The initial large language model can be an existing large language model, such as a Generative Pre-trained Transformer (GPT) model.

[0110] In practical applications, appropriate prompts can be constructed based on the item feature sequences of the aforementioned historical items to guide the initial large language model inference to obtain the set of relationships between the historical items. The construction process of the prompts will be described in detail in subsequent embodiments.

[0111] S22. In the set of association relationships, the selected candidate historical items are replaced with the corresponding masks, and the following operations are performed for each mask: perform contextual semantic analysis on the item features of other historical items associated with a mask, and select the estimated historical item corresponding to a mask from the multiple candidate historical items based on the analysis results. When the estimated historical item does not match the corresponding replaced candidate historical item, delete the replaced candidate historical item and its corresponding item features.

[0112] To test the set of relationships obtained through reasoning, some historical items in the set of relationships were obscured, and the historical items corresponding to each obscured position were estimated.

[0113] Specifically, from the set of historical items included in the association relationship, multiple candidate historical items are randomly selected, and each candidate historical item is replaced with a mask. When a candidate historical item corresponds to an association reason, the candidate historical item and the corresponding association reason can be replaced with a mask.

[0114] As an optional implementation, when replacing multiple candidate historical items with their corresponding masks in the association set, the following operations can be performed on each of the historical items in the association set: Randomly determine whether to select a historical item according to a set probability; if a historical item is selected, designate it as a candidate historical item and replace it with its corresponding mask. This ensures that each historical item in the association set has the same probability of being selected.

[0115] The probability setting can be configured as needed, such as 0.2, 0.3, etc., without limitation. In practice, for a historical item, a random number generator can be used to generate a random number between 0 and 1. If this random number is less than or equal to the set probability, the historical item is considered selected; otherwise, if the random number is greater than the set probability, the historical item is not selected.

[0116] For example, the association logic diagram after the association set is transformed is as follows: Figure 3 For example, this association logic diagram is only a schematic diagram and does not include nodes corresponding to all historical items. The candidate historical items randomly selected from the association set are replaced with masks, such as... Figure 4The MASK1 and MASK2 shown are shown. The candidate historical item to be replaced for MASK1 is node V2, and the candidate historical item to be replaced for MASK2 is node V7.

[0117] After obtaining the set of associations for each mask, for each mask, contextual semantic analysis is performed based on the item features of other historical items associated with that mask to select the estimated historical item corresponding to that mask from multiple candidate historical items. Specifically, when the above set of associations is obtained through inference using an initial large language model, the initial large language model can be used to predict the estimated historical item corresponding to each mask.

[0118] In practical applications, appropriate prompts can be constructed based on the set of associations for each mask and multiple selected candidate historical items to guide the initial large language model inference to obtain the estimated historical item corresponding to each mask. The construction process of the prompts will be described in detail in subsequent embodiments.

[0119] For each mask, if the estimated historical item does not match the corresponding candidate historical item to be replaced, it is considered that the association relationship corresponding to the candidate historical item cannot be accurately inferred, that is, the candidate historical item is not trusted, and the candidate historical item and its corresponding item features are deleted.

[0120] After deleting one or more candidate historical items and their features that cannot be accurately inferred from the initially obtained historical items, the remaining historical items are taken as target historical items, thereby obtaining the target historical items and their features that were not deleted, so as to be used for subsequent inference of the relationship between the recommended items and the target historical items.

[0121] S23. Based on the recommended item features of the item to be recommended and the target item features of each target historical item that has not been deleted, infer the target association relationship between the item to be recommended and at least one target historical item, and combine the target association reason contained in the target association relationship with the item to be recommended as recommendation information to push to the target object.

[0122] Among these features, the product recommendation platform can push recommended items to a target audience. Specifically, it can push recommended items based on the target audience's subscription information or proactively push recommended items to the target audience. Before pushing recommended items, the reasons for the recommendation can be explored.

[0123] The characteristics of the recommended item can include recommended item attributes and a set interaction type. The recommended item attributes are similar to those of the historical items mentioned above. The set interaction type can be a specified interaction type, or it can be any interaction type selected from multiple interaction types. For example, if the recommended item is an article, the set interaction type can be a specified interaction type, such as click, or any interaction type such as click, read, share, like, etc.

[0124] Each target historical item's target item features are arranged according to the interaction sequence of each target historical item to obtain a target item feature sequence. Based on this target item feature sequence and the recommended item features, the target association relationship between the recommended item and at least one target historical item is inferred. This inference process is similar to the inference process of the item feature sequence in S21 above.

[0125] Optionally, the initial large language model described above can be used to infer the target association relationship between the item to be recommended and at least one target historical item based on the feature sequence of the target item and the features of the recommended item. Furthermore, the initial large language model can be fine-tuned. The fine-tuned sample set can be obtained from the aforementioned association set. For example, some associations in the association set can be used as sample associations. These partial associations can be the associations to which each candidate historical item belongs, as determined in S22 above, and can be accurately inferred. Then, the fine-tuned target large language model is used to infer the aforementioned target association relationship.

[0126] When reasoning yields a target association relationship between the item to be recommended and a target historical item, the target association reason in that target association relationship is used as the recommendation reason. When reasoning yields a target association relationship between the item to be recommended and multiple target historical items, the item to be recommended and multiple target historical items form an association link. This target association relationship contains multiple target association reasons, that is, there is a target association reason between every two items (the item to be recommended or the target historical item). In this case, the target association reason corresponding to the item to be recommended is used as the recommendation reason, and the recommendation reason is pushed to the target object together with the item to be recommended.

[0127] It should be noted that when multiple items to be recommended are obtained corresponding to the target object, the above S23 can be executed for each item to be recommended, or the above S23 can be executed for multiple items to be recommended at the same time. That is, based on the recommended item features of each of the multiple items to be recommended and the above target item feature sequence, the target association relationship between each item to be recommended and at least one target historical item is inferred.

[0128] In this embodiment, based on the item characteristics of each historical item interacted with by the target object, a set of association relationships between historical items is inferred. Each association relationship involves at least two related historical items and the reason for their association. Then, by verifying the above association relationship set, unconvincing candidate historical items are deleted, and convincing target historical items are obtained. After obtaining the item to be recommended, based on the recommended item characteristics of the item to be recommended and the target item characteristics of each target historical item, the target association relationship between the item to be recommended and at least one target historical item can be accurately inferred. The target association reason contained in the target association relationship is used as the recommendation reason and pushed to the target object along with the item to be recommended. In this way, the obtained recommendation reason is related to the target historical items interacted with by the target object and is persuasive. Through this recommendation reason, the target object can understand why the recommendation system pushes the recommended item, thereby increasing the target object's trust and satisfaction with the recommended item, increasing the probability that the recommended item will be interacted with by the object, and thus improving the recommendation effect.

[0129] In some embodiments, in S21 above, a pre-stored initial large language model can be used to infer the set of relationships between historical items based on the item characteristics of each historical item involved in each historical interaction operation performed by the target object. Specifically, this includes the following steps A1-A2:

[0130] A1. Based on the characteristics of each historical item, construct the first prompt information.

[0131] As can be seen from the above embodiments of this application, the item characteristics of each historical item may include: the item attributes of the corresponding historical item and the historical interaction type of the target object with the corresponding historical item; and, the historical items are arranged in the order of interaction, and correspondingly, the item characteristics of each historical item are also arranged in the order of interaction, forming an item characteristic sequence.

[0132] To enable the initial large language model to infer the set of relationships between historical items, the above item feature sequences can be textualized according to a preset first prompt word template to construct first prompt information. This first prompt information is used to describe the individual item features of each historical item in the item feature sequence, as well as the analysis method of the item feature sequence and the setting format of the output results.

[0133] The aforementioned first prompt can guide the initial large language model to analyze the behavioral logic of the target object interacting with each historical item, thereby outputting a set of relationships between each historical item.

[0134] A2. Using a pre-stored initial large language model, based on the item attributes and historical interaction types of each historical item contained in the first prompt information, as well as the interaction sequence of each historical item, analyze the relationships between each historical item and the reasons for each relationship to obtain the set of relationships between each historical item.

[0135] After the first prompt information is input into the initial large language model, the initial large language model can perform semantic analysis on the first prompt information. Then, based on the item attributes and historical interaction types of each historical item contained in the first prompt information, as well as the interaction sequence of each historical item, and according to the analysis method indicated by the first prompt information, it analyzes the relationships between each historical item and the reasons for each relationship. Finally, according to the output result setting format indicated by the first prompt information, it outputs the set of relationships between each historical item.

[0136] For example, such as Figure 5 As shown, the item feature sequence of each historical item interacted with by the target object is S[V1, V2, V3, ..., V...]. n Assuming the historical items are articles, each item's characteristics include: item ID, title, account holder, article category, and historical interaction type. For example, the first prompt information constructed based on the item characteristic sequence is as follows:

[0137] The following input information describes a sequence of item characteristics for multiple historical items interacting with the target object. The items in the sequence are arranged in chronological order. Each historical item's characteristics include five categories of information: unique ID, title, account author, article category, and interaction type, recorded within parentheses "()".

[0138] (1, "Under the algorithm, real problems are ignored", "xxAPP", "Internet", "click"), ..., (998, ...), ... (10000, ...).

[0139] Please analyze the interaction reasons of the target object with each historical item, striving to uncover these reasons from a global perspective of the item feature sequence. Unconvincing historical items can be ignored. Output the analysis results in the following format:

[0140] {{Interaction Type 1}} Item {{ID1}} triggered {{Interaction Type 2}} Item {{ID2}} because it was the user's {{Behavior Reason 1}}.

[0141] The first prompt message can use special identifiers, such as "{{}}", to represent input information or information that needs to be output according to a set format.

[0142] After inputting the initial prompt information into the initial large language model, the initial large language model analyzes the set of relationships between historical items based on the prompt information. Each relationship can be understood as an interaction causal relationship between at least two historical items. The order of historical items in the historical item sequence is reorganized, and the interaction reasons of the target object with each historical item are analyzed. The output result is as follows:

[0143] "Clicking item {10000} triggered reading sessions for item {998} due to the user's shared interests; sharing item {81} triggered clicks for item {82} due to the user's viewed accounts..."

[0144] The interaction reason for each historical item serves as the associated reason for that historical item.

[0145] The output of the initial large language model is a set of associations between historical items, such as... Figure 6 As shown, after the set of relationships is converted into a relationship logic graph, the nodes and adjacency relationships in the relationship logic graph can be formatted and stored to obtain an adjacency table. For example: (1, Click, "") -> (2, High-duration reading, "Same interest category"), (2, Share, "") -> (6, Click, "Accounts viewed"), that is, each node in the adjacency table can correspond to the item ID, historical interaction type and association reason (if any).

[0146] In this embodiment of the application, since the initial large language model has a powerful text semantic understanding capability and has accumulated a large amount of universal knowledge, it is able to analyze the relationship between historical items based on the first prompt information and the item feature sequence of each historical item, and generate or summarize convincing reasons for the relationship.

[0147] In one possible implementation, considering the vast textual expression space, the association reasons in each association relationship in the initial large language model inference may be highly divergent and lack refinement. To address this issue, this embodiment of the application defines a preset set of association reasons in the aforementioned first prompt information, instructing the initial large language model to select association reasons from this set for each association relationship. The set of association reasons can be set empirically.

[0148] At this point, in step A1 above, the first prompt information can be constructed based on the individual characteristics of each historical item and the preset set of related reasons.

[0149] Furthermore, using the initial large language model, based on the item attributes and historical interaction types of each historical item contained in the first prompt information, as well as the interaction sequence of each historical item, we analyze the relationships between each historical item. For each relationship obtained from the analysis, we select the corresponding relationship reason from the set of relationship reasons to obtain the set of relationship relationships between each historical item.

[0150] In this embodiment of the application, by introducing a preset set of association reasons in the first prompt information, the initial large language model selects appropriate association reasons from the set of association reasons when reasoning about each association relationship, thus ensuring that the association reasons reasoned by the initial large language model are relatively concise.

[0151] In some embodiments, in S22 above, an initial large language model can be used to predict the estimated historical item corresponding to each mask. Specifically, a second prompt information can be constructed based on the association set containing each mask and multiple candidate historical items. Using the initial large language model, the following operations are performed for each mask: based on the item features of other historical items associated with a mask contained in the second prompt information, the semantic similarity between these features and the item features of each of the multiple candidate historical items is calculated. From the multiple candidate historical items, candidate historical items whose semantic similarity meets the set conditions are selected as the estimated historical item corresponding to a mask.

[0152] The initial large language model possesses powerful semantic understanding capabilities. It can analyze the semantic information of the features of other historical items and the semantic information of the features of multiple candidate historical items, thereby determining the semantic similarity between the features of other historical items and the features of each candidate historical item. Specifically, the set condition can be the maximum semantic similarity.

[0153] Specifically, the set of relationships containing each mask can be represented as a logical association diagram containing each mask. Based on this logical association diagram and multiple candidate historical items, combined with a preset second prompt word template, a second prompt message is constructed.

[0154] For example, such as Figure 7 As shown, the set of relationships containing each mask is represented as a relationship logic graph. This relationship graph is only exemplary and does not show all historical items. Based on the preset first prompt word template, the relationship logic graph, and multiple candidate historical items, the following second prompt information is constructed:

[0155] "Given input information is the existing historical interaction items and interaction reasons of the target object. Select the most logical candidate historical item from the candidate historical item set and fill it into [MASK] to complete the set of association relationships."

[0156] The following is the input information:

[0157] (1, "Under the algorithm, real problems are ignored", click, "")->[MASK1]->(6, "Information accessibility", share, ""),(1, "Under the algorithm, real problems are ignored", click, "")->(5, "The lack of human factors in data-driven decision-making", share, "")->[MASK2],…

[0158] The following is a collection of candidate historical items:

[0159] (2, "Limitations of the Algorithm", "Same Interest Categories")

[0160] (7, "The real summer's top influencers", "Accounts I've watched")

[0161] …

[0162] Output the analysis results in the following format:

[0163] {{MASK1}}:{{Candidate Historical Item 1}},{{MASK2}}:{{Candidate Historical Item 2}},…”

[0164] After inputting the second prompt information into the initial large language model, the output of the initial large language model is the estimated historical item corresponding to each mask, that is, a candidate historical item selected from the set of candidate historical items. For example, the output of the initial large language model is as follows:

[0165] MASK1:(7, "The real summer's top influencer", "Accounts I've watched")

[0166] MASK2:(2, "Limitations of the algorithm", "Same interest category")

[0167] ."

[0168] Compared to the adjacency list storage mentioned above (including item ID + historical interaction type + interaction reason), the node information of each node in the masked association logic graph also includes item attributes (such as title text), which makes it easier for the initial large language model to understand the unmasked historical items and candidate historical items at the semantic level.

[0169] like Figure 8As shown, after obtaining the output of the initial large language model, the candidate historical items replaced by each mask can be obtained according to the aforementioned stored adjacency list. For each mask, the corresponding estimated historical item is checked to see if it matches the replaced candidate historical item. For example, if the item ID of the candidate historical item replaced by MASK1 is 2, while the item ID of the estimated historical item corresponding to MASK1 is 7, they do not match. Similarly, if the item ID of the candidate historical item replaced by MASK2 is 7, while the item ID of the estimated historical item corresponding to MASK2 is 2, they also do not match, and so on, performing the matching process for all masks.

[0170] In this embodiment of the application, for the set of association relationships between historical items obtained by the initial large language model inference, multiple candidate historical items are masked, and the initial large language model infers the estimated historical items corresponding to each mask. For each mask, the corresponding estimated historical item is matched with the corresponding candidate historical item to be replaced, and the candidate historical items that do not match are deleted, so that the initial large language model has the ability to self-correct the inferred set of association relationships.

[0171] In some embodiments, considering that the initial large language model is a general model rather than a task-specific model, even if some of the associations inferred in the above embodiments are accurate, it cannot be guaranteed that the output associations will be accurate and consistent for the same multiple historical items. Therefore, the initial large language model is fine-tuned to better suit the task of inferring associations. To enable the initial large language model to more accurately infer the associations between the item to be recommended and each target historical item, supervised fine-tuning can be performed on the initial large language model.

[0172] First, obtain the sample set for supervised fine-tuning.

[0173] When using the initial large language model to predict each mask in the association set, for each mask that is accurately predicted, that is, the predicted historical item corresponding to the mask matches the corresponding candidate historical item to be replaced, it is considered that the initial large language model is relatively accurate in its inference of the association to which the candidate historical item belongs. The accuracy of the association to which the candidate historical item belongs can be further verified, and the association that passes the verification is taken as a sample association.

[0174] For example, suppose we assume a set of associations containing each mask, which is converted to the above... Figure 4 The association logic diagram shown can be used to verify the accuracy of the association relationship between V2 and V6 corresponding to V1->MASK1 when the initial large language model accurately predicts the mask MASK1. After the verification is passed, the association relationship is used as a sample association relationship.

[0175] Furthermore, for each accurately predicted mask corresponding to a relationship, based on the item characteristics of each historical item contained in the relationship, the relationship between these historical items can be inferred again through the initial large language model. The inference process is similar to the inference process in S21 of the above embodiment. The accuracy of the re-output relationship is verified, and after the verification is passed, the re-output relationship is used as a sample relationship.

[0176] Based on the above process, multiple sample associations can be obtained, which can serve as a sample set for supervised fine-tuning of the initial large language model.

[0177] Then, in order to reduce the complexity of fine-tuning the initial large language model, the trained parameters in the set network layer of the initial large language model can be replaced with the set parameters to be trained to obtain the large language model to be trained.

[0178] Among them, the network layer can be a key network layer in the initial large language model, that is, a network layer that has an important impact on the model performance, such as an attention network layer, a feedforward neural network layer, etc. The specific parameters to be trained can be low-rank matrices, which can reduce the number of model parameters and computational complexity, while keeping the model performance unchanged as much as possible. If the network layer is set as an attention network layer, when constructing the low-rank matrix, the weight matrix of the linear transformation in the attention network layer can be decomposed into a matrix. Singular value (i.e., eigenvalue) decomposition or other appropriate decomposition methods can be used, for example, expressed as the following equation (1):

[0179] W=UΣV T (1)

[0180] Where W represents the weight matrix, U and V represent orthogonal matrices, and ∑ represents the singular matrix.

[0181] Then, retaining the top k largest singular values ​​and their corresponding singular vectors in ∑, a low-rank matrix is ​​constructed, which can be expressed as the following equation (2):

[0182] W k =U k ∑ k V k T (2)

[0183] Among them, U k and V k It contains the first k columns of U and V respectively, Σ k Let represent the k*k singular values ​​in Σ.

[0184] Using a low-rank matrix W k Replace the original weight matrix W to obtain the large language model to be trained.

[0185] Finally, based on the obtained association relationships among multiple samples, the large language model to be trained undergoes multiple rounds of iterative training to obtain the target large language model; each round of training includes:

[0186] The sample item features of at least two historical items included in a sample association are input into a large language model to obtain the predicted association. Based on the difference between the predicted association and the sample association, the loss value is determined, and the training parameters in the set network layer are adjusted based on the loss value.

[0187] The sample item characteristics of each historical item are the same as those of the historical items in the above embodiments of this application.

[0188] Specifically, the loss value between the predicted association and the sample association is calculated based on a set loss function, such as the cross-entropy loss function. Then, the gradient is calculated using the backpropagation algorithm based on the loss value, and the trainable parameters in the set network layers are adjusted to minimize the loss value. The initial language model iteration stopping condition can be that the number of iterations reaches a set threshold, or the model loss value is less than a set loss threshold.

[0189] For example, such as Figure 9 The diagram illustrates the fine-tuning process of the large language model to be trained. Assuming that the features of each sample item in a sample association relationship are represented as V1, V2, and V6, V1, V2, and V6 are input into the large language model to be trained (with a low-rank matrix introduced in the network layers). The output is the predicted association relationship, i.e., the predicted association relationship between the sample items corresponding to V1, V2, and V6. Based on the difference between the predicted association relationship and the actual sample association relationship, a loss value is calculated, and the parameters of the low-rank matrix are adjusted based on this loss value. Finally, after the large language model to be trained is fine-tuned, the target large language model is obtained.

[0190] In this embodiment of the application, by supervising the fine-tuning of the initial large language model, the fine-tuned target large language model is made more suitable for the reasoning task of the above-mentioned set of associations. Subsequently, after obtaining the item to be recommended, the association between the item to be recommended and each target historical item can be inferred more accurately, thereby accurately mining the reasons for recommending the item to be recommended.

[0191] In some embodiments, when reasoning about the target association between the item to be recommended and at least one target historical item in S23 above, the above-described fine-tuned target large language model can be used.

[0192] The features of the recommended item include: the recommended item attributes and the set interaction type of the recommended item. The features of the target item include: the item attributes and historical interaction types of the corresponding target historical items. Specifically, based on the features of the recommended item and each target item, a third prompt message is constructed. Using a target large language model, based on the item attributes and historical interaction types of each target historical item contained in the third prompt message, the recommended item attributes and the set interaction type, as well as the interaction sequence of each target historical item, a correlation analysis is performed on the recommended item and each target historical item to obtain the target association relationship between the recommended item and at least one target historical item.

[0193] The process of constructing the third prompt information based on the features of the recommended items and the features of each target item is similar to the process of constructing the first prompt information based on the item feature sequence in the above embodiment. That is, the first prompt word template mentioned above can be used to construct the third prompt information.

[0194] In this embodiment of the application, the semantic-level recommendation reasons of the items to be recommended can be mined based on the target large language model. These reasons mainly revolve around the target object's interest categories and historical items of historical interaction. For example, the recommendation reasons include "interest category: cars", "accounts I often watch", and "related events of the content I have watched".

[0195] The following is combined with Figure 10 The reasoning process for recommending a particular item is illustrated with an example.

[0196] For example, such as Figure 10 As shown, when a target user logs into a recommendation-related client on their terminal device, the item recommendation platform on the server can push recommended items to the target user. At this time, the reasons for recommending the item can be analyzed. Specifically, the server can construct a third prompt based on the item feature sequence S'[V1, V2, V3, ..., Vm] of each target historical item and the recommended item feature [W] of the item to be recommended. This third prompt is used to instruct the target big language model to infer the target association relationship between the item to be recommended and at least one target historical item. After inputting the third prompt into the target big language model, the target big language model analyzes and obtains the target association relationship based on the third prompt: the target association relationship between the target historical item corresponding to V3 and the item to be recommended, as well as the reason for the target association. This reason for the target association is then used as the recommendation reason and pushed to the target user along with the item to be recommended.

[0197] For example, such as Figure 11The image shown is a screenshot of the message interface after subscription. The recommended item pushed to the target audience is the article "Who invented this recipe? The taste is absolutely amazing." At the same time, the reason for the recommendation is "A public account I often read," so that the target audience can intuitively understand why the article is recommended.

[0198] In some embodiments, a corresponding association logic graph is generated based on the latest set of association relationships formed by each target historical item obtained in the above embodiments; then, graph feature extraction is performed on the association logic graph to obtain association logic graph features; based on the preset item features and association logic graph features of the item to be recommended, the interaction probability of the target object with the item to be recommended is predicted through a pre-stored interaction probability prediction model.

[0199] The association logic graph represents the interaction logic between various historical items of the target object, reflecting rich historical interaction information of the target object. By extracting features from the association logic graph, it is easy to integrate it into the interaction probability prediction model as an extended feature input, which can improve the prediction accuracy of interaction probabilities. Interaction probabilities include, for example, click-through rate, negative feedback rate, positive feedback rate, etc. The preset item features of the item to be recommended can represent the item attributes, etc.

[0200] It should be noted that in practical applications, as the number of historical items interacting with the target object increases, the set of relationships for reasoning can be updated periodically to obtain the latest relationship logic diagram.

[0201] Specifically, after obtaining multiple items to be recommended for a target audience, the item recommendation platform on the server can predict the interaction probability of each item and sort the items according to the interaction probability. For example, when the interaction probability is the click rate, the items are arranged from high to low click rate; when the interaction probability is the negative feedback rate, the items are arranged from low to high negative feedback rate; and when the interaction probability is the positive feedback rate, the items are arranged from high to low positive feedback rate.

[0202] For example, such as Figure 12 As shown, the features of the association logic graph (Node embedding, Graph embedding, etc.) and the preset item features are input into the interaction probability prediction model to predict the interaction probability of the target object with the recommended item, such as click rate, negative feedback rate, etc.

[0203] As an optional implementation, the node information of each node includes: the target item characteristics of the corresponding target historical item; the above-mentioned feature extraction of the association logic graph to obtain the association logic graph features may specifically include the following steps B1-B3:

[0204] B1. Perform vector transformation on the node information of each node in the association logic graph to obtain the corresponding node representation vector.

[0205] The vector transformation method for node information can be selected based on the specific content of the node information.

[0206] As an optional implementation, the node information of each node also includes: the association reason included in the association relationship to which the corresponding target historical item belongs; the target item characteristics include the item attributes and historical interaction types of the corresponding target historical item.

[0207] When converting the node information of each node in the association logic graph into a corresponding node representation vector, the association reason, item attribute and historical interaction type of each node can be converted into vectors respectively to obtain the first sub-vector, the second sub-vector and the third sub-vector. Then, the first sub-vector, the second sub-vector and the third sub-vector are merged to obtain the corresponding node representation vector.

[0208] Specifically, the following vector transformation process is performed on the node information of each node (including item ID, historical interaction type, and association reason):

[0209] 1) For the reasons for association (which belong to the text), a pre-trained word vector model, such as the Word-to-Vector (Word2Vec) model or the Global Vectors for Word Representation (GloVe) model, is used to convert the text corresponding to the reasons for association into a vector representation. Specifically, each word in the text corresponding to the reasons for association is mapped to its corresponding word vector, and then these word vectors are averaged or summed to obtain the vector representation of the text corresponding to the reasons for association, i.e., the first sub-vector mentioned above.

[0210] It should be noted that when a node does not contain a reason for association, the first word vector corresponding to that node can be defaulted to 0.

[0211] 2) For the item ID, a trainable feature extraction layer (i.e., the Embedding layer) is used to convert it into a vector representation, namely the second sub-vector mentioned above.

[0212] 3) For historical interaction types (belonging to category data), use one-hot encoding to convert them into vector representations, namely the third sub-vector mentioned above.

[0213] 4) Vector aggregation: The vector representations of the interaction reason, the vector representations of the historical interaction type, and the vector representation of the item ID are concatenated or weighted and summed to obtain a comprehensive node representation vector.

[0214] B2. Extract features from the node representation vectors of each node in the associated logic graph to obtain the node feature vectors of each node.

[0215] One approach is to use Node Embedding to extract features from the node representation vector of each node. Node Embedding is used to learn the node feature vectors of each node in the interconnected logical graph, ensuring that the node feature vectors of different nodes with similar structures and attributes are also similar.

[0216] Specifically, the main idea of ​​the Node Embedding method is to learn a class of mappings, utilizing the topological structure of the interconnected logical graph itself, to map the node information of each node in the graph into a low-dimensional dense vector space, and optimize these mappings so that the learned node feature vector of each node can truly reflect the topological structure of the interconnected logical graph itself. Specifically, for each node in the interconnected logical graph, a node feature vector of a specified dimension is learned, and each node's respective feature vector is used for downstream tasks. For example, the Node Embedding method can use algorithms such as DeepWalk, Node to Vector (Node2Vec), or graph neural networks.

[0217] B3. Merge the node feature vectors of multiple nodes to obtain the features of the association logic graph.

[0218] Specifically, the node feature vectors of multiple nodes are concatenated or weighted and summed to obtain the features of the association logic graph.

[0219] In this embodiment, the node information of each node in the association logic graph is converted into a node vector representation using a suitable vector transformation method, which facilitates feature extraction of each node and obtains the node feature vector of each node. Then, the node feature vectors of multiple nodes are fused into the weight features of the association logic graph, so as to be used for downstream interaction probability prediction tasks and improve the prediction accuracy of interaction probability.

[0220] Based on the same inventive concept, this application also provides an item recommendation device. The principle of this device in solving the problem is similar to the method in the above embodiments. Therefore, the implementation of this device can refer to the implementation of the above method, and the repeated parts will not be described again.

[0221] like Figure 13 As shown, this is a structural schematic diagram of the item recommendation device 1300, which may include:

[0222] The reasoning unit 1301 is used to reason about the set of relationships between historical items based on the item characteristics of each historical item involved in each historical interaction operation performed by the target object; each relationship includes: at least two related historical items and the reason for the relationship;

[0223] The verification unit 1302 is used to replace the selected candidate historical items with corresponding masks in the association set, and perform the following operations for each mask: perform contextual semantic analysis based on the item features of other historical items associated with a mask, and select the estimated historical item corresponding to a mask from the multiple candidate historical items based on the analysis results. When the estimated historical item does not match the corresponding replaced candidate historical item, delete the replaced candidate historical item and its corresponding item features.

[0224] The recommendation unit 1303 is used to infer the target association relationship between the item to be recommended and at least one target historical item based on the recommended item features of the item to be recommended and the target item features of each target historical item that has not been deleted, and to combine the target association reason contained in the target association relationship with the item to be recommended as recommendation information and push it to the target object.

[0225] Optionally, the characteristics of each historical item include: the item attributes of the corresponding historical item, the historical interaction types of the target object with the corresponding historical item; and the historical items are arranged in the order of interaction.

[0226] Inference unit 1301 is specifically used for:

[0227] Based on the unique characteristics of each historical item, construct the first prompt information;

[0228] Using a pre-stored initial large language model, based on the item attributes and historical interaction types of each historical item contained in the first prompt information, as well as the interaction sequence of each historical item, we analyze the relationships between each historical item and the reasons for each relationship, and obtain the set of relationships between each historical item.

[0229] Optionally, the inference unit 1301 is specifically used for:

[0230] Based on the individual characteristics of each historical item and a pre-defined set of related reasons, construct the first prompt information;

[0231] Using an initial large language model, based on the item attributes and historical interaction types of each historical item contained in the first prompt information, as well as the interaction sequence of each historical item, we analyze the relationships between each historical item. For each relationship obtained from the analysis, we select the corresponding relationship reason from the set of relationship reasons to obtain the set of relationship relationships between each historical item.

[0232] Optionally, the verification unit 1302 is specifically used for:

[0233] Based on the set of relationships containing each mask and multiple candidate historical items, construct the second prompt information;

[0234] Using the initial large language model, perform the following operations for each mask: based on the item features of other historical items associated with a mask contained in the second prompt information, calculate the semantic similarity between the mask and the item features of multiple candidate historical items, and select the candidate historical items whose semantic similarity meets the set conditions from the multiple candidate historical items as the estimated historical item corresponding to a mask.

[0235] Optionally, the device also includes a training unit for:

[0236] Whenever the estimated historical item corresponding to a mask matches the corresponding candidate historical item to be replaced, the accuracy of the association to which the candidate historical item to be replaced belongs is checked in the association set, and the association that passes the check is taken as a sample association.

[0237] In the initial large language model's defined network layers, the trained parameters are replaced with the pre-defined parameters to be trained, resulting in a large language model to be trained. Based on the obtained association relationships among multiple samples, the large language model to be trained undergoes multiple rounds of iterative training to obtain the target large language model. Each round of training includes:

[0238] The sample item features of at least two historical items included in a sample association are input into a large language model to obtain the predicted association. Based on the difference between the predicted association and the sample association, the loss value is determined, and the training parameters in the set network layer are adjusted based on the loss value.

[0239] Optionally, the features of the recommended item include: the recommended item attributes and the set interaction type of the item to be recommended; the features of the target item include: the item attributes and historical interaction types of the corresponding target historical item.

[0240] Recommended unit 1303 is specifically used for:

[0241] Based on the characteristics of recommended items and the characteristics of each target item, a third prompt message is constructed;

[0242] Using a target large language model, based on the item attributes and historical interaction types of each target historical item contained in the third prompt information, the recommended item attributes and set interaction types, as well as the interaction sequence of each target historical item, a correlation analysis is performed on the item to be recommended and each target historical item to obtain the target association relationship between the item to be recommended and at least one target historical item.

[0243] Optionally, when replacing multiple selected candidate historical items with their corresponding masks in the association set, the verification unit 1302 is specifically used for:

[0244] For each of the multiple historical items contained in the association set, perform the following operations:

[0245] Based on a set probability, randomly determine whether to select a historical item;

[0246] When selecting a historical item, a historical item is considered as a candidate historical item, and the candidate historical item is replaced with the corresponding mask.

[0247] Optionally, the device also includes:

[0248] The generation unit is used to generate a corresponding association logic graph based on the latest set of association relationships formed by each target historical item that has not been deleted; the association logic graph includes multiple nodes, each node corresponds to a target historical item, and each node is connected to at least one other node;

[0249] The extraction unit is used to extract graph features from the association logic graph to obtain the features of the association logic graph.

[0250] The prediction unit is used to predict the interaction probability of the target object with the recommended item based on the preset item features and association logic graph features of the item to be recommended, through a pre-stored interaction probability prediction model.

[0251] Optionally, the node information for each node includes: the target item characteristics of the corresponding target historical item;

[0252] The extraction unit is specifically used for:

[0253] The node information of each node in the association logic graph is transformed into a vector to obtain the corresponding node representation vector;

[0254] Feature extraction is performed on the node representation vectors of multiple nodes in the association logic graph to obtain the node feature vectors of each node.

[0255] The node feature vectors of multiple nodes are fused together to obtain the features of the association logic graph.

[0256] Optionally, the node information for each node may also include: the association reason included in the association relationship to which the corresponding target historical item belongs; and the target item characteristics including the item attributes and historical interaction types of the corresponding target historical item.

[0257] When converting the node information of each node in the association logic graph into a corresponding node representation vector, the extraction unit is specifically used for:

[0258] The associated reason, item attribute and historical interaction type of each node are transformed into vectors to obtain the first sub-vector, the second sub-vector and the third sub-vector;

[0259] The first sub-vector, the second sub-vector, and the third sub-vector are merged to obtain the corresponding node representation vector.

[0260] For ease of description, the above sections are divided into modules (or units) according to their functions and described separately. Of course, in implementing this application, the functions of each module (or unit) can be implemented in one or more software or hardware components.

[0261] In this application embodiment, the terms "module" or "unit" refer to a computer program or part of a computer program that has a predetermined function and works with other related parts to achieve a predetermined goal, and can be implemented wholly or partially using software, hardware (such as processing circuitry or memory), or a combination thereof. Similarly, a processor (or multiple processors or memory) can be used to implement one or more modules or units. Furthermore, each module or unit can be part of an overall module or unit that includes the functionality of that module or unit.

[0262] Having introduced the article recommendation method and apparatus according to exemplary embodiments of this application, we will now introduce an electronic device according to another exemplary embodiment of this application.

[0263] Based on the same inventive concept as the above-described method embodiments, this application also provides an electronic device. In one embodiment, the electronic device may be a server, such as... Figure 1 The server 120 is shown. In this embodiment, the structure of the electronic device can be as follows: Figure 14 As shown, it includes a memory 1401, a communication module 1403, and one or more processors 1402.

[0264] The memory 1401 is used to store computer programs executed by the processor 1402. The memory 1401 may mainly include a program storage area and a data storage area. The program storage area may store the operating system and programs required to run instant messaging functions, etc.; the data storage area may store various instant messaging information and operation instruction sets, etc.

[0265] Memory 1401 may be volatile memory, such as random-access memory (RAM); memory 1401 may also be non-volatile memory, such as read-only memory, flash memory, hard disk drive (HDD), or solid-state drive (SSD); or memory 1401 may be any other medium capable of carrying or storing a desired computer program having the form of instructions or data structures and accessible by a computer, but is not limited thereto. Memory 1401 may be a combination of the above-described memories.

[0266] Processor 1402 may include one or more central processing units (CPUs) or digital processing units, etc. Processor 1402 is used to implement the recommended method for the above-mentioned items when calling computer programs stored in memory 1401.

[0267] The communication module 1403 is used to communicate with terminal devices and other servers.

[0268] This application embodiment does not limit the specific connection medium between the memory 1401, communication module 1403, and processor 1402. This application embodiment... Figure 14 The memory 1401 and the processor 1402 are connected via a bus 1404, and the bus 1404 is in Figure 14 The diagram uses thick lines to describe the connections between other components; these are for illustrative purposes only and should not be considered limiting. The 1404 bus can be divided into address bus, data bus, control bus, etc. For ease of description, Figure 12 It is described using only a thick line, but does not indicate that there is only one bus or one type of bus.

[0269] The memory 1401 stores a computer storage medium, which stores computer-executable instructions for implementing the item recommendation method of this application embodiment. The processor 1402 is used to execute the above-described item recommendation method, such as... Figure 2 As shown.

[0270] In another embodiment, the electronic device may also be other electronic devices, such as... Figure 1 The terminal device 110 is shown. In this embodiment, the electronic device can be structured as follows: Figure 15As shown, it includes components such as: communication component 1510, memory 1520, display unit 1530, camera 1540, sensor 1550, audio circuit 1560, Bluetooth module 1570, processor 1580, etc.

[0271] The communication component 1510 is used to communicate with the server. In some embodiments, it may include a Circuit-Based Wireless Fidelity (WiFi) module, which is a short-range wireless transmission technology. Electronic devices can use the WiFi module to help users send and receive information.

[0272] The memory 1520 can be used to store software programs and data. The processor 1580 executes various functions of the terminal device 110 and performs data processing by running the software programs or data stored in the memory 1520. The memory 1520 may include high-speed random access memory, and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other volatile solid-state storage device. The memory 1520 stores an operating system that enables the terminal device 110 to run. In this application, the memory 1520 may store the operating system and various application programs, and may also store computer programs that execute the methods recommended in the embodiments of this application.

[0273] The display unit 1530 can also be used to display information input by the user or information provided to the user, as well as various menus of the terminal device 110, in a graphical user interface (GUI). Specifically, the display unit 1530 may include a display screen 1532 disposed on the front of the terminal device 110. The display screen 1532 may be configured as a liquid crystal display, a light-emitting diode, or the like. The display unit 1530 can be used to display a display interface for recommended items in the embodiments of this application.

[0274] The display unit 1530 can also be used to receive input digital or character information and generate signal inputs related to user settings and function control of the terminal device 110. Specifically, the display unit 1530 may include a touch screen 1531 disposed on the front of the terminal device 110, which can collect touch operations of the user on or near it, such as clicking buttons, dragging scroll boxes, etc.

[0275] The touchscreen 1531 can be placed on top of the display screen 1532, or the touchscreen 1531 and the display screen 1532 can be integrated to realize the input and output functions of the terminal device 110. After integration, it can be referred to as a touch display screen. In this application, the display unit 1530 can display the application and the corresponding operation steps.

[0276] Camera 1540 can be used to capture still images, which users can then share via an application. There can be one or multiple cameras 1540. An object is projected onto a photosensitive element through a lens, generating an optical image. This photosensitive element can be a charge-coupled device (CCD) or a complementary metal-oxide-semiconductor (CMOS) phototransistor. The photosensitive element converts the light signal into an electrical signal, which is then transmitted to the processor 1580 for conversion into a digital image signal.

[0277] The terminal device may also include at least one sensor 1550, such as an accelerometer 1551, a proximity sensor 1552, a fingerprint sensor 1553, and a temperature sensor 1554. The terminal device may also be equipped with other sensors such as a gyroscope, barometer, hygrometer, thermometer, infrared sensor, light sensor, and motion sensor.

[0278] Audio circuitry 1560, speaker 1561, and microphone 1562 provide an audio interface between the user and terminal device 110. Audio circuitry 1560 converts received audio data into electrical signals, which are then transmitted to speaker 1561, where they are converted into sound signals for output. Terminal device 110 may also be equipped with volume buttons for adjusting the volume of the sound signal. On the other hand, microphone 1562 converts collected sound signals into electrical signals, which are received by audio circuitry 1560, converted into audio data, and then output to communication component 1510 for transmission to, for example, another terminal device 110, or to memory 1520 for further processing.

[0279] The Bluetooth module 1570 is used to interact with other Bluetooth devices that also have a Bluetooth module via the Bluetooth protocol. For example, a terminal device can establish a Bluetooth connection with a wearable electronic device (such as a smartwatch) that also has a Bluetooth module through the Bluetooth module 1570, thereby exchanging data.

[0280] The processor 1580 is the control center of the terminal device, connecting various parts of the terminal through various interfaces and lines. It executes various functions and processes data by running or executing software programs stored in the memory 1520 and calling data stored in the memory 1520. In some embodiments, the processor 1580 may include one or more processing units; the processor 1580 may also integrate an application processor and a baseband processor, wherein the application processor mainly handles the operating system, user interface, and applications, and the baseband processor mainly handles wireless communication. It is understood that the baseband processor may not be integrated into the processor 1580. In this application, the processor 1580 can run the operating system, applications, user interface display and touch response, and the article recommendation method of this application embodiment. Furthermore, the processor 1580 is coupled to the display unit 1530.

[0281] In some possible implementations, various aspects of the item recommendation method provided in this application can also be implemented in the form of a program product, which includes a computer program. When the program product is run on an electronic device, the computer program causes the electronic device to perform the steps in the item recommendation method according to the various exemplary embodiments of this application described above. For example, the electronic device can perform actions such as... Figure 2 The steps are shown in the figure.

[0282] The program product may employ any combination of one or more readable media. A readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples (a non-exhaustive list) of readable storage media include: electrical connections having one or more wires, portable disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0283] The program product of the embodiments of this application may employ a portable compact disc read-only memory (CD-ROM) and include a computer program, and may run on an electronic device. However, the program product of this application is not limited thereto. In this document, the readable storage medium may be any tangible medium that contains or stores a program that may be used by or in conjunction with a command execution system, apparatus, or device.

[0284] A readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying a readable computer program. This propagated data signal may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A readable signal medium may also be any readable medium other than a readable storage medium, capable of sending, propagating, or transmitting a program for use by or in conjunction with a command execution system, apparatus, or device.

[0285] Computer programs contained on readable media may be transmitted using any suitable medium, including but not limited to wireless, wired, optical fiber, RF, etc., or any suitable combination thereof.

[0286] Computer programs for performing the operations of this application can be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java and C++, and conventional procedural programming languages ​​such as C or similar languages. The computer program can execute entirely on the user's electronic device, partially on the user's electronic device, as a standalone software package, partially on the user's electronic device and partially on a remote electronic device, or entirely on a remote electronic device or server. In cases involving remote electronic devices, the remote electronic device can be connected to the user's electronic device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external electronic device (e.g., via the Internet using an Internet service provider).

[0287] It should be noted that although several units or sub-units of the device have been mentioned in the detailed description above, this division is merely exemplary and not mandatory. In fact, according to embodiments of this application, the features and functions of two or more units described above can be embodied in one unit. Conversely, the features and functions of one unit described above can be further divided and embodied by multiple units.

[0288] Furthermore, although the operations of the method of this application are described in a specific order in the accompanying drawings, this does not require or imply that these operations must be performed in that specific order, or that all the operations shown must be performed to achieve the desired result. Additionally or alternatively, certain steps may be omitted, multiple steps may be combined into one step, and / or one step may be broken down into multiple steps.

[0289] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing a computer-usable computer program.

[0290] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, produce a machine for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0291] These computer program commands may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the commands stored in the computer-readable storage medium produce an article of manufacture including command means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0292] These computer program commands can also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing the commands executed on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0293] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0294] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. An article recommendation method characterized by comprising: The method comprises: Based on the characteristics of each historical object involved in each historical interaction operation performed by the target object, the correlation set between the historical objects is inferred; each correlation contains at least two correlated historical objects and a correlation reason; In the correlation set, replace a plurality of selected candidate historical objects with corresponding masks respectively, and for each mask, perform the following operations: based on the characteristics of other historical objects associated with the mask, perform context semantic analysis, and based on the analysis result, select the estimated historical object corresponding to the mask from the plurality of candidate historical objects; when the estimated historical object does not match the corresponding replaced candidate historical object, delete the replaced candidate historical object and the corresponding characteristics; Based on the characteristics of the recommended object and the characteristics of each target historical object, the target correlation between the recommended object and at least one target historical object is inferred, and the target correlation reason included in the target correlation is combined with the recommended object as recommendation information and pushed to the target object.

2. The method of claim 1, wherein, The characteristics of each historical object include the characteristics of the corresponding historical object and the historical interaction type of the target object with the corresponding historical object; the historical objects are arranged in order of interaction; Based on the characteristics of each historical object involved in each historical interaction operation performed by the target object, the correlation set between the historical objects is inferred, comprising: Based on the characteristics of each historical object, a first prompt information is constructed; Based on the characteristics of each historical object and the initial large language model, the characteristics of each historical object and the order of interaction are analyzed to obtain the correlation set between the historical objects.

3. The method of claim 2, wherein, Based on the characteristics of each historical object and the initial large language model, the characteristics of each historical object and the order of interaction are analyzed to obtain the correlation set between the historical objects. Based on the characteristics of each historical object and the initial large language model, the characteristics of each historical object and the order of interaction are analyzed to obtain the correlation set between the historical objects. Based on the characteristics of each historical object and the initial large language model, the characteristics of each historical object and the order of interaction are analyzed to obtain the correlation set between the historical objects. Based on the characteristics of each historical object and the initial large language model, the characteristics of each historical object and the order of interaction are analyzed to obtain the correlation set between the historical objects.

4. The method of claim 2, wherein, The method comprises the following steps of: for each mask, performing the following operations based on the item features of other historical items associated with the mask: performing context semantic analysis, and selecting an estimated historical item corresponding to the mask from the plurality of candidate historical items based on the analysis result. Based on the association set containing each mask and the plurality of candidate historical items, a second prompt information is constructed. For each mask, the following operations are performed based on the item features of other historical items associated with the mask contained in the second prompt information: the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and the semantic similarity between the item features of the plurality of candidate historical items respectively, and 5. The method according to any one of claims 2 to 4, characterized in that, ​ ​ ​ ​ 6. The method of claim 5, wherein, ​ ​ ​ ​ 7. The method according to any one of claims 1 to 4, characterized in that, ​ ​ ​ In determining the one historical item, the one historical item is taken as one candidate historical item, and the one candidate historical item is replaced by a corresponding mask.

8. The method according to any one of claims 1 to 4, characterized in that, The method further comprises: generating a corresponding association logic graph based on a latest association relation set constituted by the target historical items not deleted; the association logic graph comprises a plurality of nodes, each node corresponding to one target historical item, and each node being connected with at least one other node; performing graph feature extraction on the association logic graph to obtain association logic graph features; predicting an interaction probability of the target object to the to-be-recommended item based on a preset item feature of the to-be-recommended item and the association logic graph features through a pre-stored interaction probability prediction model.

9. The method of claim 8, wherein, The node information of each node comprises a target item feature of the corresponding target historical item; The feature extraction on the association logic graph to obtain the association logic graph features comprises: performing vector conversion on the node information of each node in the association logic graph to obtain a corresponding node representation vector; performing feature extraction on the node representation vectors of the plurality of nodes in the association logic graph respectively to obtain node feature vectors of the plurality of nodes respectively; fusing the node feature vectors of the plurality of nodes respectively to obtain the association logic graph features.

10. The method of claim 9, wherein, The node information of each node further comprises an association reason contained in an association relation to which the corresponding target historical item belongs; and the target item feature comprises an item attribute and a historical interaction type of the corresponding target historical item. The conversion of the node information of each node in the association logic graph into a corresponding node representation vector comprises: performing vector conversion on the association reason, the item attribute and the historical interaction type corresponding to each node respectively to obtain a first sub-vector, a second sub-vector and a third sub-vector; performing fusion processing on the first sub-vector, the second sub-vector and the third sub-vector to obtain a corresponding node representation vector. 11.An article recommendation device characterized by comprising: comprise: an inference unit configured to infer an association relation set among historical items based on target object execution of each historical item respectively involved in each historical interaction operation; each association relation comprising: at least two associated historical items and an association reason; a verification unit configured to replace a plurality of candidate historical items selected in the association relation set with corresponding masks respectively, and perform the following operations on each mask: performing context semantic analysis based on item features of other historical items associated with one mask, and selecting a pre-estimated historical item corresponding to the one mask from the plurality of candidate historical items based on an analysis result; and when the pre-estimated historical item does not match a corresponding replaced candidate historical item, deleting the replaced candidate historical item and a corresponding item feature. The recommendation unit is configured to infer a target association relationship between the to-be-recommended item and at least one target historical item based on a recommendation item feature of the to-be-recommended item and a target item feature of each target historical item that has not been deleted, and push a target association reason included in the target association relationship to the target object as recommendation information in combination with the to-be-recommended item.

12. An electronic device, comprising: It comprises a processor and a memory, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the method in any one of claims 1-10.

13. A computer-readable storage medium, characterized in that, It comprises a computer program, and when the computer program runs on an electronic device, the computer program is used to make the electronic device execute the steps of the method in any one of claims 1-10.

14. A computer program product, characterised in that, It comprises a computer program stored in a computer readable storage medium, and when a processor of an electronic device reads the computer program from the computer readable storage medium, the processor executes the computer program, so that the electronic device executes the steps of the method in any one of claims 1-10.

Images