Model processing method for predicting user interest information, electronic device, and storage medium

Abstract

Embodiments of the present application provide a model processing method and device for predicting user interest information, a user interest information processing method and device, a data object processing method and device, an electronic device and a storage medium. According to the embodiments of the present application, when a prediction model for predicting user interest information is established, first, the historical behavior data of the user is data mined, the historical behavior data of multiple users is obtained, multiple data objects of corresponding multiple categories having similar characteristics with the historical behavior data are determined, further, the user interest information represented by the historical behavior data is constructed according to the multiple data objects, and the prediction model for predicting the user interest information is trained in combination with the above-mentioned user historical behavior data and user interest information. The prediction model can increase the probability of matching the search result with the user's intention by searching the data objects under multiple object categories in the case that the user's search intention is not clear, and improve the access traffic of multiple object categories.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a model processing method and apparatus for predicting user interest information, a user interest information processing method and apparatus, a data object processing method and apparatus, an electronic device, and a storage medium. Background Technology

[0002] With the rapid development of the internet, information is exploding. When users need information, they can use query terms to express their needs and locate the required information in the data system by searching for these terms. However, in some cases, users' query terms are vague, which leads to the server retrieving information that does not meet the user's needs. This prevents users from obtaining the information they need in a timely manner, reduces the user experience, and consequently affects the traffic generated by the client.

[0003] How to quickly provide users with the information they need when the search terms are vague or even missing has become one of the problems to be solved. Summary of the Invention

[0004] This application provides a model processing method and apparatus for predicting user interest information, a user interest information processing method and apparatus, a data object processing method and apparatus, an electronic device, and a storage medium, so as to realize the prediction of user interest information and further realize the prediction of user click rate.

[0005] In a first aspect, embodiments of this application provide a model processing method for predicting user interest information, characterized in that it includes:

[0006] Obtain historical behavior data from multiple users;

[0007] Identify multiple target data objects that have similar attributes to the historical data objects accessed corresponding to the historical behavior data, wherein the multiple target data objects correspond to multiple object categories;

[0008] The user interest information represented by the historical behavior data is constructed based on the multiple target data objects;

[0009] Based on the differences between the predicted data object and the historical data object, a prediction model for predicting user interest information based on the user's historical behavior is determined, wherein the predicted data object is determined according to the user interest information.

[0010] Secondly, embodiments of this application provide a method for processing user interest information, characterized in that it includes:

[0011] Retrieve the historical behavior data of the current user who triggered the interaction;

[0012] The historical behavior data is input into the prediction model to obtain user interest information. The prediction model is used to construct the user interest information represented by the historical behavior data based on target data objects with similar attributes to the historical data objects accessed corresponding to the historical behavior data. The multiple target data objects correspond to multiple object categories.

[0013] Provide the aforementioned user interest information.

[0014] Thirdly, embodiments of this application provide a method for processing data objects, characterized in that it includes:

[0015] Retrieve the historical behavior data of the current user who triggered the interaction;

[0016] The historical behavior data is input into the prediction model to obtain recommended data objects for the current user; the prediction model is used to predict the user interest information represented by the historical behavior data based on target data objects with similar attributes to the historical data objects accessed corresponding to the historical behavior data, and to predict recommended data objects under at least one object category that the current user tends to access based on the user interest information, wherein the multiple target data objects correspond to multiple object categories;

[0017] The recommended data object is provided to the interactive client.

[0018] Fourthly, embodiments of this application provide another method for processing data objects, characterized by comprising:

[0019] Retrieve the historical behavior data of the current user who triggered the interaction;

[0020] Obtain recommended data objects under at least one object category that the current user tends to visit based on the historical behavior data; the target data objects are predicted based on user interest information represented by the historical behavior data, and the user interest information is constructed based on target data objects with similar attributes to the historical data objects visited corresponding to the historical behavior data, and the multiple target data objects correspond to multiple object categories;

[0021] The recommended data object is provided on the interactive interface.

[0022] Fifthly, embodiments of this application provide an electronic device, including a memory, a processor, and a computer program stored in the memory, wherein the processor, when executing the computer program, implements the method described in any of the above-mentioned embodiments.

[0023] Sixthly, embodiments of this application provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method described in any of the preceding claims.

[0024] Compared with the prior art, this application has the following advantages:

[0025] According to the embodiments of this application, when establishing a prediction model for predicting user interest information, data mining is first performed on the user's historical behavior data to obtain historical behavior data from multiple users. Then, multiple data objects corresponding to multiple categories that have similar characteristics to the historical behavior data are identified. Furthermore, user interest information represented by the historical behavior data is constructed based on these multiple data objects. The prediction model for predicting user interest information is then trained by combining the aforementioned user historical behavior data and user interest information. Thus, on the one hand, since historical behavior data can represent the information that users tend to access, determining similar data objects based on historical behavior data can identify user interest information represented by historical behavior data even in situations where query terms are lacking or vague. On the other hand, since the search involves multiple object categories similar to the historical behavior data, the user interest information constructed based on the identified data objects can represent the user's multi-peak interests across multiple object categories. This provides users with data objects under multiple object categories, especially when the user's search intent is unclear. By retrieving data objects under multiple object categories, the probability of matching search results with the user's intent is increased, thereby improving access traffic to multiple object categories.

[0026] When training the above prediction model, the data objects that users tend to visit can be predicted based on user interest information. Then, the loss function of the model can be determined based on the difference between the predicted data objects and the data objects actually visited in historical behavior data. The prediction model is trained based on this loss function, thereby making the prediction of user interest information more accurate.

[0027] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application, it can be implemented according to the contents of the specification. In order to make the above and other objects, features and advantages of this application more obvious and understandable, specific embodiments of this application are given below. Attached Figure Description

[0028] In the accompanying drawings, unless otherwise specified, the same reference numerals throughout the various drawings denote the same or similar parts or elements. These drawings are not necessarily drawn to scale. It should be understood that these drawings depict only some embodiments according to this application and should not be construed as limiting the scope of this application.

[0029] Figure 1This is a schematic diagram of a model processing scheme for predicting user interest information provided in this application.

[0030] Figure 2 This is a flowchart of a model processing method for predicting user interest information according to an embodiment of this application;

[0031] Figure 3 This is a flowchart of a method for processing user interest information according to an embodiment of this application;

[0032] Figure 4 This is a flowchart of a data object processing method according to an embodiment of this application;

[0033] Figure 5 This is a flowchart of a data object processing method according to another embodiment of this application;

[0034] Figure 6 This is a schematic diagram illustrating a specific example of an embodiment of this application;

[0035] Figure 7 This is a structural block diagram of a model processing apparatus for predicting user interest information according to an embodiment of this application;

[0036] Figure 8 This is a structural block diagram of a user interest information processing apparatus according to an embodiment of this application;

[0037] Figure 9 This is a structural block diagram of a data object processing apparatus according to an embodiment of this application;

[0038] Figure 10 This is a structural block diagram of a pre-data object processing apparatus according to another embodiment of this application; and

[0039] Figure 11 This is a block diagram of an electronic device used to implement the embodiments of this application. Detailed Implementation

[0040] In the following description, only certain exemplary embodiments are briefly described. As those skilled in the art will recognize, the described embodiments can be modified in various ways without departing from the concept or scope of this application. Therefore, the drawings and description are considered to be exemplary in nature and not restrictive.

[0041] To facilitate understanding of the technical solutions of the embodiments of this application, the relevant technologies of the embodiments of this application are described below. The following relevant technologies are optional solutions and can be combined with the technical solutions of the embodiments of this application in any way, and all of them fall within the protection scope of the embodiments of this application.

[0042] In view of the fact that the existing technology is unable to provide users with the information they need when the query terms used by the user are vague or lack query terms, the embodiments of this application provide a new model processing scheme for predicting user interest information, so as to solve all or part of the above-mentioned technical problems.

[0043] To more clearly demonstrate the model processing scheme for predicting user interest information provided in the embodiments of this application, examples of the schemes in the embodiments of this application are given below. Figure 1 This is a schematic diagram illustrating an exemplary solution for implementing an embodiment of this application. For example... Figure 1 As shown, this application embodiment can construct a prediction model based on user historical behavior data. In cases where user query terms are missing or vague, the prediction model can predict user interest information. Combined with the user-side interactive client, the obtained user interest information can be used to predict the click-through rate of data objects. In the context of travel e-commerce, data objects can include destination information, hotel information, travel guides, attraction ticket information, local food information, recommended blog content, etc. Furthermore, the predicted data objects that users are likely to access (i.e., recommended data objects) can be provided in a prominent position when displaying data objects on the user-side client, facilitating efficient access to data objects that match user interests, increasing the probability that the data objects provided by the client match the user's intent, and improving client access traffic.

[0044] This application provides a model processing method for predicting user interest information. For example... Figure 2 The diagram shown is a flowchart of a model processing method 200 for predicting user interest information according to an embodiment of this application, which may include:

[0045] Step S201: Obtain historical behavior data from multiple users.

[0046] This application's embodiments involve data mining of users' historical behavior data. Behavioral data records one or more user behaviors and can be obtained by capturing user interactions on interactive media such as clients or web pages. Historical behavior data can be collected from multiple clients and recorded in user-related data systems. When it is necessary to obtain historical behavior data from multiple users, historical behavior data can be directly collected from the clients or obtained from user-related data systems.

[0047] For a single user action, the data that can be captured includes the action itself, the data object, and the time of occurrence. User actions can include clicking, saving, downloading, adding to cart, and purchasing. Data objects can include information accessible to the user online, such as documents, images, videos, music, products, and advertisements. Historical behavior data refers to behavioral data that occurred over a past period, and can include short-term and long-term behavior data. Short-term behavior data is generated from user behavior within a relatively short time period (e.g., one hour), while long-term behavior data is generated from user behavior within a relatively long time period (e.g., one year).

[0048] Before building the model, historical behavior data from multiple users is acquired for subsequent mining or processing. It is understood that after acquiring the historical behavior data, it can be preprocessed. Preprocessing may include removing outlier data through data cleaning and converting the historical behavior data into vector form for later use. Specifically, the actions, data objects, and occurrence times of the corresponding behaviors in the historical behavior data can also be represented separately as vectors. The method of converting data into vector form can be to first encode the data into one-hot high-dimensional vectors, and then convert these vectors into low-dimensional dense vectors by multiplying them with embedding vectors. The data or information involved in the embodiments of this application may also include other forms, and this application does not limit them.

[0049] Since users often choose the data objects they access based on their preferences and interests, data mining of users' historical behavior data can reveal their preferences and interests. This is the process of predicting user interest information through historical behavior data, which will be explained in detail below.

[0050] Step S202: Determine multiple target data objects that have similar attributes to the historical data objects accessed corresponding to the historical behavior data, wherein the multiple target data objects correspond to multiple object categories.

[0051] The attribute information of a data object is used to describe the data object. Taking a product as an example, the product's attribute information may include its price, brand, and product category. After acquiring historical behavior data, it is possible to compare the historical data objects accessed during the historical behavior data with candidate data objects to determine whether they have similar attributes. This allows for the selection of multiple target data objects with similar attributes to the historical data objects, which are then used to further determine user interest information. When comparing historical data objects with candidate data objects, one or more attribute information of the data objects can be combined for comparison, and the similarity between the attribute information can be calculated. Candidate data objects that meet the similarity criteria are then selected as target data objects.

[0052] The aforementioned data objects can correspond to the same object category or multiple object categories. Taking the scenario of travel e-commerce as an example, object categories can include hotels, food, popular scenic spots, tickets, tour groups, travel guides, air tickets, car rentals, etc. Data objects corresponding to the above object categories can include Disney tickets (corresponding to popular scenic spots and tickets), Happy Valley tickets (corresponding to popular scenic spots and tickets), customized Hangzhou trips (corresponding to tour groups and travel guides), international youth hostels (corresponding to hotels), time-honored restaurants (corresponding to food), popular coffee shop check-in lists (corresponding to food and travel guides), discounted air tickets (corresponding to air tickets), etc.

[0053] When multiple data objects correspond to multiple different object categories, the user interest information determined by the data objects corresponding to multiple object categories can represent the user's multi-peak interests in multiple object categories. In this way, data objects under multiple object categories can be provided to the user. Especially when the user's search intent is unclear, by retrieving data objects under multiple object categories, the probability of the search results matching the user's intent is increased, and the access traffic of multiple object categories is improved.

[0054] For example, in a travel search scenario, a user's historical behavior data records that the user has visited restaurants with an average cost of 50-100 yuan per person. Data objects under multiple categories with similar attributes to this visit record include: restaurants with an average cost of 50-100 yuan per person, tickets, and discounted accommodations. When the user searches using the query term "Shanghai", the system further recommends time-honored Shanghai restaurants with an average price of 68 yuan per person, tourist attractions with an average ticket price of 30 yuan per person, and Shanghai youth hostels with an average accommodation price of 99 yuan per person.

[0055] To obtain target data objects from multiple object categories that have similar attributes to historical data objects accessed in the past, candidate data objects can be obtained from each of the multiple object categories. Then, the attribute information of the candidate data objects corresponding to different object categories is compared with that of the historical data objects. Candidate data objects whose attribute information similarity meets the conditions are selected from each of the different object categories, thereby obtaining multiple target data objects corresponding to multiple object categories.

[0056] Step S203: Construct user interest information representing the historical behavior data based on the multiple target data objects.

[0057] User interest information represents the characteristics of data objects that users tend to access, and can be determined based on the target data objects identified above using historical behavior data. Since multiple data objects correspond to multiple object categories, the user interest information constructed through this step can represent the user's multi-peaked interests across various object categories. When constructing user interest information, the multiple target data objects obtained above can be converted into vector representations, and user interest information is constructed by combining the vectors corresponding to multiple target data objects. In this embodiment, when combining vectors, vectors can be directly combined through stacking or concatenation, or a neural network model can be used, employing an attention mechanism to assign corresponding weights to each vector, and combining the vectors into user interest information through weighted summation. Further details are omitted. When using a neural network model, the number of hidden layer units can be selected according to actual needs; this application does not impose any restrictions on this.

[0058] In one optional embodiment, when mining user interest information through historical behavior data, user interest information represented by historical query terms can be constructed by combining the user's historical query terms. Historical query terms refer to query terms previously used by the user in search scenarios. When mining user's historical behavior data by combining historical query terms, at least one historical behavior data related to the historical query terms can be determined from multiple historical behavior data of the same user. Specifically, the relevant historical behavior data and historical query terms can be converted into vector form (multiple historical behavior data corresponding to multiple vectors). By calculating the similarity between the vector corresponding to the historical query terms and the vectors corresponding to the multiple historical behavior data, at least one historical behavior data with a high similarity is determined. Then, user interest information represented by the historical query terms is constructed based on the determined historical behavior data. The construction method can be to use the vector corresponding to at least one historical behavior data and the similarity between this vector and the historical query terms, using the similarity as a weight, and combining at least one historical behavior data into user interest information by weighted summation of the vector corresponding to at least one historical behavior data and its corresponding similarity. The resulting user interest information is both related to the query terms used by the user and conforms to the user's multi-peak interests.

[0059] For example, in a travel search scenario, a user's historical behavior data may contain multiple access records for data objects under the amusement park and family categories. When a user's query is "Shanghai," the search results returned to the user might include data objects such as discounted family tickets to Shanghai Disneyland, Disneyland queue guides, and bookings for family suite hotels near Shanghai Disneyland.

[0060] In one alternative computational instance, a neural network can be used to construct user interest information, where an attention mechanism is used to determine the corresponding weights for each historical behavioral data point.

[0061] In an optional embodiment, since the type and sequence of historical behavior data can indicate the user's interest state, when mining user interest information through historical behavior data, the user's interest state can also be obtained by analyzing the type and sequence of historical behavior data within a certain period. The user's interest state characterizes whether the user has released their interest through one or more current behaviors, i.e., obtained information that meets the user's needs. Specifically, it can include whether the interest has been released or not. Furthermore, corresponding behavior weights can be assigned to each historical behavior data point based on the user's interest state. Higher behavior weights can be assigned to historical behavior data where the user's interest state is "interest not released," and lower behavior weights can be assigned to historical behavior data where the user's interest state is "interest released."

[0062] Historical behavioral data can be categorized into types of user actions, such as clicks, favorites, purchases, and downloads. Behavioral sequence refers to the chronological order in which historical behaviors occur. Analyzing the types of historical behavioral data and their sequence allows us to determine the user's interest state. Specifically, when a user continuously clicks on data objects within a certain category, it indicates a high level of interest in those objects, and the user's interest state is considered "interest not released." If, after continuously clicking on data objects within a category, the type of historical behavior for at least one data object in that category changes from clicking to another type, the user's interest state for that category is considered "interest released." If the clicking behavior continues, the user's interest state is considered "interest not released."

[0063] For example, in a travel e-commerce scenario, if a user clicks on multiple public transportation cards and then purchases a 3-day unlimited travel pass, their interest in public transportation cards can be considered temporarily released. In this case, lower behavioral weights can be assigned to the transportation cards mentioned in the historical behavior data and other products within the corresponding category. However, if a user clicks on multiple public transportation cards and then adds one to their favorites or shopping cart, their interest in public transportation cards or other products within that category can be considered unreleased. In this case, higher behavioral weights can be assigned to the transportation cards mentioned in the historical behavior data and other products within the corresponding category.

[0064] After obtaining the corresponding behavioral weights from historical behavioral data, user interest information represented by the historical behavioral data can be constructed by combining these weights. Specifically, the data objects involved can be represented in vector form, and the aforementioned user interest information can be constructed by analyzing a weighted sum of multiple data objects involved in the historical behavioral data for a specific purpose.

[0065] In one optional computational instance within an e-commerce scenario, the user interest information represented by the constructed historical behavioral data with corresponding behavioral weights is denoted as u. c u c The calculation method can be:

[0066]

[0067]

[0068] Where, α i The weights of the user's historical behavior data, where i is the historical data object, and e is the weight of the historical behavior data. iHistorical data objects are represented as vectors, with Sigmoid as the activation function, FFN as a feed-forward network, and W∈R. d ×4d Sum(·) represents the cumulative pooling operator, x i bc x i bp x i ac x i ap These represent click actions prior to a specific historical action of historical data object i, purchase actions prior to a specific historical action of historical data object i, click actions after a specific historical action of historical data object i, and purchase actions after a specific historical action of historical data object i, respectively. It's understandable that the above-mentioned purchase is a type of user action in e-commerce scenarios that can identify a user's interest state as having been released. In other scenarios, x can be... i bp and x i ap The corresponding user action type is replaced with a user action type that corresponds to the application scenario, which can indicate that the user's interest status has been released.

[0069] Understandably, to make the obtained user interest information more accurate, the three types of user interest information (user interest information represented by historical behavior data, user interest information represented by historical query terms, and user interest information represented by historical behavior data with corresponding behavior weights) can be fused together. Specifically, this fusion can be achieved by using the SoftMax mechanism to combine the three types of user interest information, represented as vectors, into a single vector.

[0070] Step S204: Based on the difference between the predicted data object and the historical data object, determine the prediction model for predicting user interest information based on the user's historical behavior, wherein the predicted data object is determined according to the user interest information.

[0071] The above describes the process of constructing user interest information by mining historical behavioral data. After the user interest information is constructed, it can be used to predict the data objects that users are inclined to visit. The acquired historical behavioral data of multiple users can be used as a training set to train and determine a predictive model for user interest information based on historical behavior. The training direction can be to reduce the difference between the predicted data objects and the historical data objects actually visited in the historical behavioral data. After analyzing the difference between the predicted data objects and the actual historical data objects visited, the deviation between the prediction results and the actual data objects is corrected by configuring a loss function for the model. For example, the loss function can be constructed using the cross-entropy loss function. Combining the loss function, the predictive model for user interest information based on historical behavior can be determined. In practical applications, as needed, required contextual feature data can be added during model training. This application does not limit the contextual feature data that may be added or adjusted in practical applications.

[0072] In one optional embodiment, data mining can be performed separately on short-term and long-term behavioral data included in historical behavioral data. Short-term behavioral data consists of user behavior over a short period (e.g., one hour), while long-term behavioral data consists of user behavior over a longer period (e.g., one year). User interests mined from short-term behavioral data are recorded as short-term interest information, and user interests mined from long-term behavioral data are recorded as long-term interest information. The obtained long-term and short-term interest information are then fused to obtain the final user interest information.

[0073] The aforementioned short-term user interests may change due to short-term events such as new product launches, seasonal changes, birthdays, and holidays, exhibiting a relatively frequent change pattern. Long-term user interests, on the other hand, may include color preferences, clothing size preferences, and price preferences. These long-term interests may be influenced by factors such as family background, age, marital status, and educational background. Because these factors have longer change cycles, long-term user interests are generally more stable and change more slowly compared to short-term interests.

[0074] It is understood that the historical behavior data of multiple users obtained in step S201 of the aforementioned scheme may include both short-term and long-term behavior data, or only one of short-term or long-term behavior data. When the historical behavior data includes short-term behavior data, when constructing the user interest information represented by the historical behavior data based on multiple data objects, short-term interest information represented by the short-term behavior data can be constructed based on the multiple data objects. That is to say, after obtaining the historical behavior data of multiple users, the short-term behavior data included in the historical behavior data can be further obtained, and then multiple target data objects corresponding to multiple categories with similar characteristics to the short-term behavior data can be determined. The multiple target data objects obtained above are converted into vector representations, and short-term interest information is constructed by combining the vectors corresponding to the multiple target data objects.

[0075] After obtaining short-term interest information, long-term interest information can be constructed based on short-term interest information and long-term behavioral data, which is represented by long-term behavioral data.

[0076] It is understandable that the user's long-term interest information can be determined first, followed by the user's short-term interest information. This application embodiment does not restrict the order in which long-term or short-term interest information is determined. After the short-term and long-term interest information is constructed, user interest information can be obtained by merging the short-term and long-term interest information.

[0077] In one optional embodiment, when constructing long-term interest information based on short-term interest information and long-term behavioral data, at least one long-term behavioral data related to the short-term interest information can be found from multiple long-term behavioral data of the same user. Then, long-term interest information represented by historical query terms can be constructed based on the found historical behavioral data. Similarly, multiple long-term behavioral data can be converted into multiple vectors, and long-term behavioral data related to the short-term interest information can be calculated separately. The user's long-term interest information can then be constructed using the related long-term behavioral data.

[0078] In one optional embodiment, when merging short-term and long-term interest information to obtain user interest information, the interest weights corresponding to the short-term and long-term interest information can be determined based on the matching relationship between the most recent historical behavior data and the short-term and long-term interest information, respectively. The short-term and long-term interest information are then weighted according to their respective weights. The weighted short-term and long-term interest information are then merged to obtain the user interest information. For example, an attention mechanism can be used to configure interest weights for the short-term and long-term interest information, and then weighted accordingly. The short-term interest information is denoted as u. sh o rtLong-term interest information is denoted as u l o ng Let user interest information be denoted as u. The following is a possible calculation formula for merging user interest information:

[0079] u=α·u short +(1-α)·u long

[0080] Where α is the interest weight configured by the attention mechanism for short-term interest information.

[0081] In an optional embodiment, to distinguish between short-term and long-term interest information and prevent coupling between them, interest tags corresponding to short-term and long-term interest information can be constructed respectively. A loss function is then constructed based on the difference between user interest information and interest tags to distinguish between short-term and long-term interest information. This constructed loss function is used to train the aforementioned prediction model. To differentiate between the different loss functions, the loss function used to correct the difference between the prediction result and the data object actually accessed by the user is designated as the first loss function, and the loss function used to distinguish between short-term and long-term interest information is designated as the second loss function. The first and second loss functions can be combined by summing them for use in the model established in this embodiment for predicting user interest information.

[0082] To construct interest tags, short-term and long-term interest information can first be converted into vector form. The short-term interest information in vector form is denoted as the first vector, and the long-term interest information in vector form as the second vector. When determining short-term interest tags, the average value of the first vector is denoted as the first average value, which can be used as the short-term interest tag. When determining long-term interest tags, the long-term interest information is divided into multiple information segments according to time intervals. The second average value of multiple data points in the second vector corresponding to each information segment is calculated. The data is then weighted according to the second average value and the time weight of the corresponding information segment, and the weighted result is used as the long-term interest tag.

[0083] When constructing the second loss function to distinguish between short-term and long-term interest information, a contrastive learning task can be constructed to ensure that the similarity between short-term interest information and short-term interest labels is greater than the similarity between long-term interest information and long-term interest labels. That is, four contrastive learning tasks can be constructed such that the similarity between the first vector and the first average is greater than the similarity between the second vector and the second average; the similarity between the first average and the first vector is greater than the similarity between the second average and the second vector; the similarity between the second vector and the second average is greater than the similarity between the first vector and the first average; and the similarity between the second average and the second vector is greater than the similarity between the first average and the first vector. The training results of these contrastive learning tasks are then used as the second loss function to distinguish between the aforementioned short-term and long-term interest information. Specifically, the second loss function can be constructed using pairwise loss BPR (pairwise loss Bayesian Personalized Ranking).

[0084] The model for predicting user interest information provided in this application embodiment can also be further used to predict the click-through rate (CTR) of users for target data objects. Specifically, after obtaining user interest information, the similarity between the target data object and the user interest information can be calculated. Products with higher similarity are the data objects that users are likely to access, analyzed from the predicted user interest information. These data objects are denoted as recommended data objects, and can be provided to users on the user-side interactive client. Since the recommended data objects are predicted based on user interest information, users' access tendency (or click tendency on the client's interactive page) will be higher than for other data objects. It is worth noting that in actual tourism e-commerce scenarios, providing recommended data objects predicted using this application embodiment on the client side, compared to the existing technology in the aforementioned tourism e-commerce scenario, results in a 3.12% increase in UCTR (User CTR), a 3.73% increase in PCTR (Page CTR), and a 1.48% increase in GMV (Gross Merchandise Volume). In the search scenario of travel e-commerce, when users' search terms are relatively vague, the prediction model provided in this application embodiment improves UCTR by 4.16%, PCTR by 3.89%, and GMV by 1.87% compared to the prior art.

[0085] Figure 3 This is a flowchart of a user interest information processing method 300 according to an embodiment of this application.

[0086] Step S301: Obtain the historical behavior data of the current user who triggered the interactive operation.

[0087] Step S302: Input the historical behavior data into the prediction model to obtain user interest information. The prediction model is used to construct the user interest information represented by the historical behavior data based on target data objects with similar attributes to the historical data objects accessed by the historical behavior data. The multiple target data objects correspond to multiple object categories.

[0088] Step S303: Provide the user interest information.

[0089] The aforementioned scheme for predicting user interest information can be used to obtain user interest information predicted by the model by inputting the user's historical behavior data into the prediction model provided by the scheme. The predicted user interest information can also be further used to predict the data objects that users tend to visit, the behaviors that users tend to perform (such as clicking, saving, etc.), and to recommend data objects that match the user's interests.

[0090] In search scenarios, the above-mentioned scheme for predicting user interest information can also be combined with the current query term. While acquiring the historical behavior data of the current user that triggered the interaction, the scheme also acquires the query term currently used by the user. The acquired historical behavior data and the current query term are then input into the prediction model to achieve the prediction of user interest information based on the query term. The relevant details can be found in the descriptions of the above embodiments and will not be repeated here.

[0091] This application provides a method for processing data objects, such as... Figure 4 The diagram shown is a flowchart of a data object processing method 400 according to an embodiment of this application, which may include:

[0092] Step S401: Obtain the historical behavior data of the current user who triggered the interactive operation.

[0093] Step S402: Input the historical behavior data into the prediction model to obtain recommended data objects for the current user; the prediction model is used to predict the user interest information represented by the historical behavior data based on target data objects with similar attributes to the historical data objects accessed corresponding to the historical behavior data, and to predict recommended data objects under at least one object category that the current user tends to access based on the user interest information, wherein the multiple target data objects correspond to multiple object categories.

[0094] Step S403: Provide the recommended data object to the interactive client.

[0095] Based on the predicted user interest information provided in the above embodiments, this application embodiment can recommend data objects that match the user's interest information. In other words, it can predict one or more recommended data objects that the current user is inclined to visit based on the user interest information.

[0096] In search scenarios, the aforementioned scheme for predicting user interest information can be further combined with the current query term to predict user interest information based on the user's current query term and historical behavior data, and to further predict and recommend data objects that the user is inclined to access based on the user's interest information. The relevant details can be found in the descriptions of the above embodiments and will not be repeated here.

[0097] This application also provides another method for processing user information. For example... Figure 5 The diagram shown is a flowchart of a user interest information processing method 500 according to another embodiment of this application.

[0098] Step S501: Obtain the historical behavior data of the current user who triggered the interactive operation;

[0099] Step S502: Obtain recommended data objects under at least one object category that the current user tends to visit based on the historical behavior data; the target data objects are predicted based on user interest information represented by the historical behavior data, and the user interest information is constructed based on target data objects with similar attributes to the historical data objects visited corresponding to the historical behavior data, and the multiple target data objects correspond to multiple object categories;

[0100] Step S503: Provide the recommended data object on the interactive interface.

[0101] In this embodiment of the application, based on the user interest information and recommended data objects provided in the above embodiments, data objects that match the user's interests can be recommended to the user in the interactive client used by the user. For example, the recommended data objects can be provided to the user in a relatively prominent position on the interactive page to facilitate the user's access.

[0102] The above-described scheme for predicting user interest information can also be combined with the current query term to further predict user interest information based on the user's current query term and historical behavior data, and to further predict and recommend data objects that the user is inclined to access based on the user's interest information. The relevant details can be found in the descriptions of the above embodiments and will not be repeated here.

[0103] To enable those skilled in the art to better understand the technical solution of this application, the following specific example illustrates the embodiments of this application. Figure 6 The diagram shown is a specific example of this application. Figure 6 Before building the model, the data can be preprocessed, that is, the data can be converted into a low-dimensional dense vector form by using a matrix embedding layer. Figure 6 This involves the construction of user interest information, the construction of a loss function, and the training of the model in the process of building a model for predicting user interest information, as provided in the embodiments of this application. The construction of user interest information includes a short-term interest extractor, a long-term interest extractor, and an interest infusing module. The construction of the loss function includes an interest disentanglement module. The training of the model includes a final prediction module for building the model for predicting user interest information.

[0104] The short-term interest information construction section involves extracting three types of user interest information from short-term user historical behavior data: user interest information represented by historical query terms (u q User interest information represented by historical behavioral data (u t ) and user interest information represented by historical behavioral data with corresponding behavioral weights (u c When extracting user interest information represented by historical query terms, a query-dependent interest encoder can be used to encode it based on short-term user behavior data x. ubs Historical search terms e q Using a corresponding gated recurrent unit (GRU) layer, user interest information represented by historical query terms is obtained. When extracting user interest information represented by historical behavior data, a target-dependent interest encoder can be used, based on short-term user behavior data x. ubs Data object e i Using the corresponding gated recurrent unit (GRU) layer, user interest information represented by historical behavior data is obtained. When extracting user interest information represented by historical behavior data with corresponding behavior weights, a behavioral-dependent interest encoder can be used, based on short-term user behavior data x. ubs User long-term behavior data u blBy combining multilayer neural perceptrons and corresponding activation functions (MLP & sigmoid), weights (α1…α) are assigned to behavioral data. i …α iTs By using a weighted sum, user interest information is obtained, represented by historical behavioral data with corresponding behavioral weights. ubs ={e i1 ,e i2 ,...,e iTs}, x ubl ={e i1 ,e i2 ,...,e iTsl}, where e i1 e represents the first target data object in the historical behavior data. i2 T represents the second target data object in the historical behavior data. s T represents the length of short-term behavioral data. l This indicates the length of long-term behavioral data. Finally, the above three types of user interest information are merged (Short-term Interest Infusing) into short-term user interest information u. short Specifically, when extracting user interest information represented by historical behavior data with corresponding behavioral weights, one can start from short-term user behavior data x. ubs User and long-term behavioral data u bl In this process, by extracting subsequences, we obtain the click behavior x preceding a specific historical behavior for historical data object i. i bc And the purchase behavior prior to this historical behavior x i bp Click behavior following this historical behavior x i ac And subsequent purchasing behavior x i ap and the above x i bc x i bp x i ac x i ap Weights (α1…α2) are assigned to behavioral data by combining multilayer neural perceptrons and corresponding activation functions (MLP & sigmoid). i …α iTs By using a weighted sum, user interest information is obtained from historical behavioral data with corresponding behavioral weights.

[0105] The construction of long-term interest information involves utilizing short-term interest information. short and long-term user historical behavior data x ubl , (x ubl ={e i1 ,e i2 ,...,e iTsl Long-term interest information U is obtained through an attention network. long .

[0106] The comparative learning task construction involves generating short-term proxy labels for short-term interest information and long-term proxy labels for long-term interest information. Specifically, when generating long-term proxy labels, the long-term interest information can be window-segmented, and the average value of the segmented long-term interest information is calculated. This average value is then used as the proxy label for the long-term interest information. long When generating pseudo-labels for short-term interests, the average value of the short-term interest information can be directly calculated, and this average value can be used as the pseudo-label for the short-term interest information (P). short Then, a contrastive learning task is constructed using long-term interest information, short-term interest information, long-term interest pseudo-labels, and short-term interest pseudo-labels, ultimately yielding a contrastive loss function.

[0107] The long-term and short-term interest fusion part includes using an attention network to assign weights to long-term and short-term interest information respectively, and then merging the long-term and short-term interest information according to their corresponding weights (Fusion Predictor) to obtain user interest information u.

[0108] The part of building a model for predicting user interests involves using users' historical behavior data. up Data object e i Query term e q Contextual feature information e c The system concatenates user interest information u, combines the concatenated content with a multilayer perceptron (MLP) and a main loss function for training, and finally constructs a model for predicting user interest information.

[0109] Corresponding to the application scenarios and methods provided in the embodiments of this application, the embodiments of this application also provide a model processing apparatus 700 for predicting user interest information. For example... Figure 7 The diagram shown is a structural block diagram of a model processing apparatus 700 for predicting user interest information according to an embodiment of this application. The apparatus may include:

[0110] Data acquisition module 701 is used to acquire historical behavior data of multiple users;

[0111] The data object determination module 702 is used to determine multiple target data objects that have similar attributes to the historical data objects accessed corresponding to the historical behavior data, and the multiple target data objects correspond to multiple object categories;

[0112] The first information construction module 703 is used to construct user interest information represented by the historical behavior data based on the plurality of target data objects;

[0113] The prediction model determination module 704 is used to determine a prediction model for predicting user interest information based on the difference between the prediction data object and the historical data object, wherein the prediction data object is determined based on the user interest information.

[0114] In an optional embodiment, the device 700 may further include a second information construction module, configured to determine at least one historical behavior data related to a historical query term from multiple historical behavior data of the same user, and construct user interest information represented by the historical query term based on the determined historical behavior data.

[0115] In an optional embodiment, the device 700 may further include a third information construction module, used to assign corresponding behavior weights to the historical behavior data according to the user interest state indicated by the type and behavior sequence of the historical behavior data; and construct user interest information represented by the historical behavior data with corresponding behavior weights.

[0116] In an optional embodiment, the historical behavior data includes short-term behavior data and long-term behavior data, and the first information construction module 703 may include:

[0117] The short-term interest information construction submodule is used to construct short-term interest information representing the short-term behavioral data based on the multiple target data objects.

[0118] In an optional embodiment, the method may further include:

[0119] The long-term interest information construction submodule is used to construct long-term interest information represented by the long-term behavior data based on the short-term interest information and the long-term behavior data.

[0120] The information merging submodule is used to merge the short-term interest information and long-term interest information into user interest information.

[0121] In one optional embodiment, the long-term interest information construction submodule is specifically used to search for at least one long-term behavior data related to the short-term interest information from multiple long-term behavior data of the same user, and construct the long-term interest information represented by the historical query terms based on the searched historical behavior data.

[0122] In one optional embodiment, the information merging submodule is specifically used to determine the interest weights corresponding to the short-term interest information and the long-term interest information respectively based on the matching relationship between the most recent historical behavior data and the short-term interest information and the long-term interest information; and to merge the short-term interest information and the long-term interest information according to the corresponding interest weights to obtain user interest information.

[0123] In an optional embodiment, the device further includes:

[0124] An interest tag construction module is used to construct interest tags corresponding to the short-term interest information and the long-term interest information, respectively.

[0125] The loss function construction module is used to construct a loss function based on the difference between user interest information and interest tags to distinguish between short-term and long-term interest information, and to train the prediction model.

[0126] Corresponding to the application scenarios and methods provided in the embodiments of this application, the embodiments of this application also provide a user interest information processing device 800. For example... Figure 8 The diagram shown is a structural block diagram of a user interest information processing apparatus 800 according to an embodiment of this application, which may include:

[0127] Data acquisition module 801 is used to acquire the historical behavior data of the current user that triggered the interactive operation;

[0128] Data input module 802 is used to input the historical behavior data into a prediction model to obtain user interest information. The prediction model is used to construct user interest information represented by the historical behavior data based on target data objects with similar attributes to the historical data objects accessed corresponding to the historical behavior data. The multiple target data objects correspond to multiple object categories.

[0129] Information providing module 803 is used to provide the user interest information.

[0130] Corresponding to the application scenarios and methods provided in the embodiments of this application, the embodiments of this application also provide a data object processing device 900. Figure 9This is a structural block diagram of a data object processing apparatus according to an embodiment of this application, which may include:

[0131] Data acquisition module 901 is used to acquire the historical behavior data of the current user that triggered the interactive operation;

[0132] Prediction module 902 is used to input the historical behavior data into a prediction model to obtain recommended data objects for the current user; the prediction model is used to predict user interest information represented by the historical behavior data based on target data objects with similar attributes to the historical data objects accessed corresponding to the historical behavior data, and to predict recommended data objects under at least one object category that the current user tends to access based on the user interest information, wherein the multiple target data objects correspond to multiple object categories;

[0133] The object providing module 903 is used to provide the recommendation data object to the interactive client.

[0134] Corresponding to the application scenarios and methods provided in the embodiments of this application, the embodiments of this application also provide another data object processing device 1000. Figure 10 This is a structural block diagram of a pre-data object processing apparatus according to another embodiment of this application, which may include:

[0135] Data acquisition module 1001 is used to acquire the historical behavior data of the current user that triggered the interactive operation;

[0136] Prediction module 1002 is used to obtain recommended data objects under at least one object category that the current user tends to visit based on the historical behavior data; the target data objects are predicted based on user interest information represented by the historical behavior data, the user interest information is constructed based on target data objects with similar attributes to the historical data objects visited corresponding to the historical behavior data, and the multiple target data objects correspond to multiple object categories;

[0137] The object providing module 1003 is used to provide the recommended data object on the interactive interface.

[0138] The functions of each module in each device in the embodiments of this application can be found in the corresponding description in the above method, and they have corresponding beneficial effects, which will not be repeated here.

[0139] Figure 11 This is a block diagram of an electronic device used to implement embodiments of this application. Figure 11As shown, the electronic device includes a memory 1101 and a processor 1102. The memory 1101 stores a computer program that can run on the processor 1102. When the processor 1102 executes the computer program, it implements the method described in the above embodiments. The number of memories 1101 and processors 1102 can be one or more.

[0140] The electronic device also includes:

[0141] Communication interface 1103 is used to communicate with external devices and perform data exchange and transmission.

[0142] If the memory 1101, processor 1102, and communication interface 1103 are implemented independently, they can be interconnected via a bus to communicate with each other. This bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. This bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 11 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0143] Optionally, in a specific implementation, if the memory 1101, processor 1102, and communication interface 1103 are integrated on a single chip, then the memory 1101, processor 1102, and communication interface 1103 can communicate with each other through an internal interface.

[0144] This application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method provided in this application.

[0145] This application also provides a chip including a processor for calling and executing instructions stored in a memory, causing a communication device with the chip installed to perform the method provided in this application.

[0146] This application also provides a chip, including: an input interface, an output interface, a processor, and a memory. The input interface, output interface, processor, and memory are connected through an internal connection path. The processor is used to execute code in the memory. When the code is executed, the processor is used to execute the method provided in the application embodiment.

[0147] It should be understood that the aforementioned processor can be a Central Processing Unit (CPU), or other general-purpose processors, Digital Signal Processors (DSPs), Application Specific Integrated Circuits (ASICs), Field-Programmable Gate Arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. General-purpose processors can be microprocessors or any conventional processor. It is worth noting that the processor can be a processor supporting Advanced Reduced Instruction Set Machines (ARM) architecture.

[0148] Further, optionally, the aforementioned memory may include read-only memory and random access memory. The memory may be volatile memory or non-volatile memory, or may include both. Non-volatile memory may include read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. Volatile memory may include random access memory (RAM), which serves as an external cache. By way of example, but not limitation, many forms of RAM are available. Examples include Static Random Access Memory (SRAM), Dynamic Random Access Memory (DRAM), Synchronous DRAM (SDRAM), Double Data Rate SDRAM (DDR SDRAM), Enhanced Synchronous DRAM (ESDRAM), Synchronous Link DRAM (SLDRAM), and Direct Rambus RAM (DR RAM).

[0149] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. A computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions according to this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transferred from one computer-readable storage medium to another.

[0150] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of those different embodiments or examples.

[0151] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "a plurality of" means two or more, unless otherwise explicitly specified.

[0152] Any process or method described in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing a particular logical function or process. Furthermore, the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functionality involved.

[0153] The logic and / or steps described in the flowchart or otherwise herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus or device (such as a computer-based system, a processor-included system or other system that can fetch and execute instructions from, an instruction execution system, apparatus or device).

[0154] It should be understood that various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. All or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware, the program being stored in a computer-readable storage medium, which, when executed, includes one or a combination of the steps of the method embodiments.

[0155] Furthermore, the functional units in the various embodiments of this application can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. This storage medium can be a read-only memory, a disk, or an optical disk, etc.

[0156] The above description is merely an exemplary embodiment of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various variations or substitutions within the technical scope described in this application, and these should all be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A model processing method for predicting user interest information, characterized in that, include: Obtain historical behavior data from multiple users; Identify multiple target data objects that have similar attributes to the historical data objects accessed corresponding to the historical behavior data, wherein the multiple target data objects correspond to multiple object categories; The user interest information is constructed based on the multiple target data objects to represent the historical behavior data. The construction of user interest information includes: a short-term interest information construction part, a long-term interest information construction part, and a long-short-term interest fusion part. The short-term interest information construction part includes three types of user interest information extracted from short-term user historical behavior data: first user interest information represented by historical query terms, second user interest information represented by historical behavior data, and third user interest information represented by historical behavior data with corresponding behavior weights. The first user interest information includes a user interest representation with a clear intent obtained based on data objects related to historical query terms in the historical behavior data. The second user interest information includes a user interest representation with an unclear search intent obtained based on data objects under various object categories related to the destination in the historical behavior data. The third user interest information includes user interest information obtained by weighted summation of purchase and click behaviors before and after the target historical behavior data in the historical behavior data, configuring corresponding behavior weights for the target historical behavior data. The long-term interest information construction part is used to obtain the long-term interest information through an attention mechanism network by utilizing short-term interest information and long-term user historical behavior data. The long-term and short-term interest fusion part includes using an attention mechanism network to assign weights to long-term interest information and short-term interest information respectively, and then merging the long-term interest information and short-term interest information according to their corresponding weights to obtain user interest information; Based on the differences between the predicted data objects and historical data objects, a prediction model for predicting user interest information based on historical user behavior is determined, wherein the predicted data objects are determined according to the user interest information; wherein, after analyzing the differences between the predicted data objects and historical data objects, a first loss function is determined to correct the difference between the predicted results and the data objects actually accessed by the user; interest tags corresponding to short-term interest information and long-term interest information are determined respectively, and a second loss function is constructed to distinguish between short-term interest information and long-term interest information based on the differences between user interest information and interest tags; the prediction model is trained by combining the first loss function and the second loss function.

2. The method according to claim 1, characterized in that, The method of obtaining an accurate representation of user interests under a clear intent based on data objects related to historical query terms in historical behavior data includes: From multiple historical behavior data of the same user, at least one historical behavior data related to the historical query term is determined, and user interest information represented by the historical query term is obtained based on the determined historical behavior data as a user interest representation under a clear intent.

3. The method according to claim 1, characterized in that, The method of obtaining long-term interest information through an attention mechanism network using short-term interest information and long-term user historical behavior data includes: From multiple long-term user historical behavior data of the same user, at least one long-term user historical behavior data related to the short-term interest information is searched, and the long-term interest information is obtained based on the searched long-term user historical behavior data.

4. The method according to claim 1, characterized in that, The process of weighting and merging long-term and short-term interest information according to their respective weights to obtain user interest information includes: Based on the matching relationship between the most recent historical behavior data and the short-term interest information and the long-term interest information, the interest weights corresponding to the short-term interest information and the long-term interest information are determined respectively. Based on the interest weights corresponding to short-term and long-term interest information, the short-term and long-term interest information are weighted and merged to obtain user interest information.

5. A method for processing user interest information, characterized in that, include: Retrieve the historical behavior data of the current user who triggered the interaction; The historical behavior data is input into a prediction model to obtain user interest information. The prediction model is determined based on the difference between the predicted data object and the historical data object. The predicted data object is determined based on the user interest information. User interest information represented by the historical behavior data is constructed based on target data objects with similar attributes to the historical data objects accessed corresponding to the historical behavior data. Multiple target data objects correspond to multiple object categories. Specifically, after analyzing the difference between the predicted data object and the historical data object, a first loss function is determined to correct the difference between the prediction result and the data object actually accessed by the user. Interest tags corresponding to short-term interest information and long-term interest information are determined respectively. Based on the difference between user interest information and interest tags, a second loss function is constructed to distinguish between short-term interest information and long-term interest information. The prediction model is trained by combining the first loss function and the second loss function. The construction of user interest information includes: a short-term interest information construction part, a long-term interest information construction part, and a long-short-term interest fusion part. The short-term interest information construction part includes three types of user interest information extracted from short-term user historical behavior data: first user interest information represented by historical query terms, second user interest information represented by historical behavior data, and third user interest information represented by historical behavior data with corresponding behavior weights. The first user interest information includes user interest representations with clear intent obtained based on data objects related to historical query terms in historical behavior data; the second user interest information includes user interest representations with unclear search intent obtained based on data objects under various object categories related to the destination in historical behavior data; and the third user interest information includes user interest information obtained by weighted summation of purchase and click behaviors before and after the target historical behavior data in historical behavior data, configuring corresponding behavior weights for the target historical behavior data. The long-term interest information construction part is used to obtain the long-term interest information through an attention mechanism network by utilizing short-term interest information and long-term user historical behavior data. The long-term and short-term interest fusion part includes using an attention mechanism network to assign weights to long-term interest information and short-term interest information respectively, and then merging the long-term interest information and short-term interest information according to their corresponding weights to obtain user interest information; Provide the aforementioned user interest information.

6. A method for processing data objects, characterized in that, include: Retrieve the historical behavior data of the current user who triggered the interaction; The historical behavior data is input into a prediction model to obtain recommended data objects for the current user. The prediction model is determined based on the difference between the predicted data objects and historical data objects. The predicted data objects are determined based on user interest information. Based on target data objects with similar attributes to the historical data objects accessed corresponding to the historical behavior data, user interest information represented by the historical behavior data is constructed. Based on the user interest information, recommended data objects under at least one object category that the current user tends to access are predicted. Multiple target data objects correspond to multiple object categories. Specifically, after analyzing the difference between the predicted data objects and historical data objects, a first loss function is determined to correct the difference between the prediction result and the data objects actually accessed by the user. Interest tags corresponding to short-term interest information and long-term interest information are determined respectively. Based on the difference between user interest information and interest tags, a second loss function is constructed to distinguish between short-term interest information and long-term interest information. The prediction model is trained by combining the first loss function and the second loss function. The construction of user interest information includes: a short-term interest information construction part, a long-term interest information construction part, and a long-short-term interest fusion part. The short-term interest information construction part includes three types of user interest information extracted from short-term user historical behavior data: first user interest information represented by historical query terms, second user interest information represented by historical behavior data, and third user interest information represented by historical behavior data with corresponding behavior weights. The first user interest information includes user interest representations with clear intent obtained based on data objects related to historical query terms in historical behavior data; the second user interest information includes user interest representations with unclear search intent obtained based on data objects under various object categories related to the destination in historical behavior data; and the third user interest information includes user interest information obtained by weighted summation of purchase and click behaviors before and after the target historical behavior data in historical behavior data, configuring corresponding behavior weights for the target historical behavior data. The long-term interest information construction part is used to obtain the long-term interest information through an attention mechanism network by utilizing short-term interest information and long-term user historical behavior data. The long-term and short-term interest fusion part includes using an attention mechanism network to assign weights to long-term interest information and short-term interest information respectively, and then merging the long-term interest information and short-term interest information according to their corresponding weights to obtain user interest information; The recommended data object is provided to the interactive client.

7. A method for processing data objects, characterized in that, include: Retrieve the historical behavior data of the current user who triggered the interaction; The method involves obtaining recommended data objects under at least one object category that the current user is likely to visit, predicted after inputting the historical behavior data into a prediction model. The prediction model is determined based on the difference between the predicted data objects and historical data objects. The predicted data objects are determined based on user interest information. User interest information, represented by the historical behavior data, is constructed based on target data objects with similar attributes to the historical data objects accessed corresponding to the historical behavior data. Recommended data objects under at least one object category that the current user is likely to visit are predicted based on this user interest information. Multiple target data objects correspond to multiple object categories. A first loss function is determined to correct the difference between the predicted data objects and historical data objects after analyzing the difference between the predicted data objects and the data objects actually visited by the user. Interest tags corresponding to short-term and long-term interest information are determined, and a second loss function is constructed to distinguish between short-term and long-term interest information based on the difference between user interest information and interest tags. The prediction model is trained by combining the first and second loss functions. The construction of user interest information includes: a short-term interest information construction part, a long-term interest information construction part, and a long-short-term interest fusion part. The short-term interest information construction part includes three types of user interest information extracted from short-term user historical behavior data: first user interest information represented by historical query terms, second user interest information represented by historical behavior data, and third user interest information represented by historical behavior data with corresponding behavior weights. The first user interest information includes user interest representations with clear intent obtained based on data objects related to historical query terms in historical behavior data; the second user interest information includes user interest representations with unclear search intent obtained based on data objects under various object categories related to the destination in historical behavior data; and the third user interest information includes user interest information obtained by weighted summation of purchase and click behaviors before and after the target historical behavior data in historical behavior data, configuring corresponding behavior weights for the target historical behavior data. The long-term interest information construction part is used to obtain the long-term interest information through an attention mechanism network by utilizing short-term interest information and long-term user historical behavior data. The long-term and short-term interest fusion part includes using an attention mechanism network to assign weights to long-term interest information and short-term interest information respectively, and then merging the long-term interest information and short-term interest information according to their corresponding weights to obtain user interest information; The recommended data object is provided on the interactive interface.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory, wherein the processor, when executing the computer program, implements the method of any one of claims 1-7.

9. A computer-readable storage medium storing a computer program that, when executed by a processor, implements the method of any one of claims 1-7.

Images