Interest speculation model training method and device, electronic equipment and storage medium

Abstract

The invention provides an interest speculation model training method and device, electronic equipment and a storage medium, relates to the technical field of computers, in particular to the field of artificial intelligence and the like, and can be used for application scenes such as content recommendation and the like. According to the specific implementation scheme, key basic information of a known object, an interest information record of the known object in a first time period and target interest information of the known object in a second time period are obtained; generating cue words according to the key basic information, the interest information record and the predefined structured template; and inputting the cue word into the to-be-trained model, training the to-be-trained model, and obtaining an interest speculation model, so that the interest speculation model outputs target interest information of the known object in the second time period. According to the scheme, the interest exploration efficiency and accuracy can be improved.

Description

Technical Field

[0001] This disclosure relates to the field of computer technology, particularly to fields such as artificial intelligence, and can be used in application scenarios such as content recommendation. Specifically, it relates to training methods, devices, electronic devices, and storage media for interest inference models. Background Technology

[0002] In new integrated recommendation scenarios, current recommendation systems need to continuously explore and recommend new, interesting content that users haven't recently encountered to improve user retention and experience, thus avoiding information cocoons. Existing methods mostly rely on random content selection or simple multi-armed slot machine strategies, lacking personalized and supervised training, resulting in low exploration efficiency. Summary of the Invention

[0003] This disclosure provides a training method, apparatus, electronic device, and storage medium for an interest prediction model.

[0004] According to a first aspect of this disclosure, a training method for an interest inference model is provided, comprising: acquiring key basic information of a known object, a record of the interest information of the known object in a first time period, and target interest information of the known object in a second time period; generating prompt words based on the key basic information, the record of interest information, and a predefined structured template; inputting the prompt words into a model to be trained, training the model to obtain an interest inference model, so that the interest inference model outputs the target interest information of the known object in the second time period.

[0005] According to a second aspect of this disclosure, a model-based interest inference method is provided, comprising: acquiring key basic information of a target object and a record of the target object's interest information in a first time period; generating prompt words based on the key basic information, the record of interest information, and a predefined structured template; inputting the prompt words into a pre-trained interest inference model; and acquiring the target interest information of the target object output by the interest inference model in a second time period.

[0006] According to a third aspect of this disclosure, a training apparatus for an interest inference model is provided, comprising: a first information acquisition module for acquiring key basic information of a known object, a record of interest information of the known object in a first time period, and target interest information of the known object in a second time period; a first prompt word generation module for generating prompt words based on the key basic information, the record of interest information, and a predefined structured template; and a model training module for inputting the prompt words into a model to be trained, training the model to obtain an interest inference model, so that the interest inference model outputs the target interest information of the known object in the second time period.

[0007] According to a fourth aspect of this disclosure, a model-based interest inference device is provided, comprising: a second information acquisition module for acquiring key basic information of a target object and interest information records of the target object in a first time period; a second prompt word generation module for generating prompt words based on the key basic information, interest information records, and a predefined structured template; a model input module for inputting the prompt words into a pre-trained interest inference model; and a model output module for acquiring target interest information of the target object output by the interest inference model in a second time period.

[0008] According to a fifth aspect of this disclosure, an electronic device is provided, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform any of the methods described in the embodiments of this disclosure.

[0009] According to a sixth aspect of this disclosure, a non-transitory computer-readable storage medium is provided storing computer instructions, wherein the computer instructions are used to cause the computer to perform any of the methods according to embodiments of this disclosure.

[0010] According to a seventh aspect of this disclosure, a computer program product is provided, including a computer program that, when executed by a processor, implements any of the methods according to embodiments of this disclosure.

[0011] Using the scheme disclosed herein can improve the efficiency and accuracy of interest exploration.

[0012] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this disclosure, nor is it intended to limit the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0013] The accompanying drawings are provided to better understand this solution and do not constitute a limitation of this disclosure. Wherein:

[0014] Figure 1 This is a flowchart illustrating the training method of the interest inference model according to an embodiment of the present disclosure;

[0015] Figure 2 This is a schematic flowchart of a model-based interest inference method according to an embodiment of the present disclosure;

[0016] Figure 3 This is another schematic flowchart of a model-based interest inference method according to an embodiment of the present disclosure;

[0017] Figure 4 This is a schematic diagram of the low-rank adaptation fine-tuning method according to an embodiment of the present disclosure;

[0018] Figure 5 This is a schematic diagram of the structure of a training device for an interest inference model according to an embodiment of the present disclosure;

[0019] Figure 6 This is a schematic diagram of the structure of a model-based interest inference device according to an embodiment of the present disclosure;

[0020] Figure 7 This is a schematic diagram of a training method for an interest inference model according to an embodiment of this disclosure;

[0021] Figure 8 This is a schematic diagram of a scenario based on a model-based interest inference method according to an embodiment of the present disclosure;

[0022] Figure 9 This is a structural diagram of an electronic device used to implement the training method and / or model-based interest inference method of the interest inference model in the embodiments of this disclosure. Detailed Implementation

[0023] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0024] In this document, the term "and / or" merely describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent three cases: A alone, A and B simultaneously, and B alone. The term "at least one" in this document indicates any combination of at least two of a plurality of elements. For example, including at least one of A, B, and C can mean including any one or more elements selected from the set consisting of A, B, and C. The terms "first" and "second" in this document refer to and distinguish between multiple similar technical terms, not to restrict the order or to limit there to only two. For example, "first feature" and "second feature" refer to two categories / two features; the first feature can be one or more, and the second feature can also be one or more.

[0025] Furthermore, to better illustrate this disclosure, numerous specific details are set forth in the following detailed description. Those skilled in the art will understand that this disclosure can be practiced without certain specific details. In some instances, methods, means, components, and circuits well known to those skilled in the art have not been described in detail in order to highlight the main points of this disclosure.

[0026] Before introducing the technical solutions of the embodiments of this disclosure, the technical terms that may be used in this disclosure will be further explained:

[0027] New integrated scenarios refer to application scenarios that combine multiple emerging technologies, different data sources, business models, or terminal channels to innovatively provide users with personalized recommendation services through cross-border integration. They are a typical manifestation of the upgrade of the recommendation system field towards intelligence, diversification, and scenario-based approaches.

[0028] Information cocoon: refers to the phenomenon in which people are surrounded by similar and homogeneous information for a long time in the process of acquiring information due to factors such as personalized recommendation algorithms, social circles, and interest preferences, forming a relatively closed information environment, which leads to a narrow vision, deepening cognitive biases, and reduced exposure to diverse viewpoints.

[0029] In related technologies, timely exploration and recommendation of new user interests in new integrated recommendation scenarios can bring users a fresh experience and is of great significance for improving user retention and activity. Recommendation systems need to continuously discover and recommend new interest content that users may be interested in but have not recently encountered, thereby improving user experience and preventing users from getting trapped in information cocoons. However, the commonly used methods for exploring new interests in the industry mainly involve randomly selecting content that users have not seen from a candidate pool, or borrowing algorithms from multi-armed slot machines to dynamically record and adjust user feedback on explored content. These exploration methods rely heavily on random selection in the early stages and lack supervised training, resulting in low overall exploration efficiency and difficulty in effectively capturing users' true new interests, thus affecting the recommendation system's ability to improve user time and activity. In addition, existing new interest exploration mechanisms usually share a content pool with all users, randomly exploring the content in the pool, lacking personalized recommendations, and rarely making full use of the reasoning capabilities of large models. This further restricts the efficiency and effectiveness of new interest exploration and makes it difficult to significantly improve overall business metrics. With the development of large model technology, its powerful reasoning and generalization capabilities have brought new opportunities for exploring new interests. In theory, it is possible to improve exploration efficiency by reasoning about changes in user interests and mining potential interests. However, there is currently a lack of large-scale training data in the recommendation field, and the effect of directly applying large models is limited. On the other hand, fine-tuning open-source large models across the entire domain faces high training costs, making it difficult to implement on a large scale.

[0030] In order to at least partially solve one or more of the above-mentioned problems and other potential problems, this disclosure proposes a training method for an interest inference model that can improve the efficiency and accuracy of interest exploration.

[0031] This disclosure provides a method for training an interest inference model. Figure 1This is a flowchart illustrating a training method for an interest inference model according to an embodiment of the present disclosure. This training method can be applied to a training device for the interest inference model. The training device is located in an electronic device. This electronic device includes, but is not limited to, fixed devices and / or mobile devices. For example, fixed devices include, but are not limited to, servers, which can be cloud servers or ordinary servers. Mobile devices include, but are not limited to, content recommendation devices, which can be mobile phones, tablets, in-vehicle terminals, etc. In some possible implementations, the training method for the interest inference model can also be implemented by a processor calling computer-readable instructions stored in memory. Figure 1 As shown, the training method for this interest inference model includes:

[0032] S101. Obtain key basic information of the known object, record the interest information of the known object in the first time period, and the target interest information of the known object in the second time period.

[0033] S102. Generate prompt words based on key basic information, interest information records, and predefined structured templates;

[0034] S103. Input the prompt words into the model to be trained, train the model to obtain the interest inference model, so that the interest inference model can output the target interest information of the known object in the second time period.

[0035] The known object refers to the entity object for which interest inference needs to be performed, including but not limited to the following types: individual user: a personalized push request subject generated based on a user's unique identifier such as a User Identifier (UID); user group: a set of users with common characteristics divided by a clustering algorithm (such as K-means), whose feature labels include at least one of demographic attributes, behavioral preference patterns, or real-time scene status. In this embodiment of the disclosure, in the context of a recommendation system, the known object can be a user in a content recommendation scenario.

[0036] Here, key basic information refers to information that can describe the core attributes of a known object. In the embodiments of this disclosure, key basic information can help the model fully understand the basic characteristics of the object.

[0037] The first time period refers to a historical time range in which the model is used to learn and infer changes in interest.

[0038] Interest information records refer to the data on interest-related behaviors actually generated by a known object within a first time period. In this embodiment of the disclosure, interest information records may include records related to a user's interests, such as browsing, clicking, liking, collecting, forwarding, and searching, within the first time period.

[0039] The second time period refers to the target time range within which the model needs to infer the interest of known objects. In this embodiment, the second time period may immediately follow the first time period, or it may be separated from the first time period by a certain time interval.

[0040] The target interest information refers to the actual interest behavior of known objects or the interest targets defined by business rules during the second time period. In this embodiment, the target interest information can be used as a supervision signal or sample label for model training.

[0041] In this embodiment, known objects can be selected first. For example, users with a certain level of activity and complete data can be selected as known objects based on actual needs. Further, key basic information of the known objects can be obtained. For example, key basic information of the known objects can be obtained directly from business databases, user registration information tables, user profiles, etc. Even further, interest information records of the known objects can be obtained. For example, interest information records can be extracted from records such as behavior logs and recommendation system logs. Specifically, user interest-related behaviors within a first time period can be summarized, including: the type of content viewed, the tags clicked, the identification code of the content collected, the category of likes or shares, etc. Finally, target interest information can be obtained. For example, by analyzing the user's clicks, browsing, collections, and other behaviors within a second time period, the most active content categories and tags can be selected, or generated using business-defined rules. Specifically, the content categories most frequently accessed by the user in the second time period can also be statistically analyzed as target interests. The above are merely illustrative examples and are not intended to limit all possible situations for obtaining the required information; they are simply not exhaustive.

[0042] In this context, a predefined structured template refers to one or more text templates that are designed and fixed in advance during the training of the interest inference model. In this embodiment, the predefined structured template can be used to organize raw data into a natural language description or input format in a structured way that is easily understood by large models.

[0043] The prompt word is a complete text generated by filling in the actual data of a specific user into a predefined structured template. In this embodiment, the prompt word is the actual input content during the training or inference of a large model.

[0044] In this embodiment, a predefined structured template suitable for understanding large models can be designed first. Specifically, the predefined structured template includes key basic information, historical interest dynamics, and prediction targets. Further, user basic attributes, interest behaviors, target labels, and other data can be filled into the predefined structured template to generate prompt words. Exemplarily, a data processing script can be written to automatically fill each user's basic attributes, interest information records, and other data into the predefined structured template. Specifically, the generated prompt words can be batch-processed using a script and saved as text or other formats for subsequent model training. The above is merely an illustrative example and does not limit the possibilities for generating prompt words; it is simply not exhaustive.

[0045] In this embodiment, an open-source large language model can be selected as the model to be trained. Specifically, a large language model that supports instruction fine-tuning or prompt word fine-tuning can also be used. Further, the generated prompt word text is used as input data, and the target interest information is used as output data to construct training samples. The training samples are input into the model to be trained to minimize the loss between the predicted output and the target interest information, and the model's effectiveness is verified. Parameter tuning and optimization are performed until the desired interest inference model is obtained. The above is merely an illustrative example and does not represent all possible scenarios for model training; it is simply not exhaustive.

[0046] The technical solution of this disclosure, by fully leveraging existing user attribute and behavioral data on the platform, significantly reduces the difficulty of acquiring training data without additional annotation. By standardizing input through predefined structured templates, it enhances the understanding of recommendation domain information by large models, enabling better transfer and generalization. By combining information from the first and second time periods, the model can capture the evolution patterns of interests, resulting in more accurate predictions. By introducing individualized historical and attribute information during model training, it achieves fine-grained and highly personalized interest prediction, not only improving the efficiency and accuracy of interest exploration but also helping recommendation systems break free from the limitations of traditional information cocoons, proactively discovering and recommending fresh content to users. This, in turn, increases user dwell time and conversion rates, enhancing user stickiness and activity.

[0047] In some embodiments, generating prompt words based on key basic information, interest information records, and a predefined structured template includes: generating historical interest information of a known object in a first time period based on the interest information records; adding key basic information to a first position in the predefined structured template; adding historical interest information to a second position in the predefined structured template; and generating prompt words based on the content of the first and second positions.

[0048] Historical interest information refers to a summary of the interest behavior of a known object within a first time period. In this embodiment of the disclosure, historical interest information is obtained by statistical analysis, summarization, and processing based on interest information records, and can reflect the object's attention to or preference for any target within the first time period.

[0049] In this embodiment, all interest-related operation data of the user within a first time period can be obtained from channels such as behavior logs and user operation databases. Subsequently, this raw data is aggregated and statistically analyzed to filter out the data that best represents the user's interests, and this data is used as historical interest information. Specifically, the obtained historical interest information can be organized into natural language descriptions, tag lists, or structured short sentences for easy insertion into templates later. The above is merely an illustrative example and does not limit the possibilities for generating historical interest information; it is simply not exhaustive.

[0050] The first position refers to the specific content insertion area reserved within the predefined structured template for key basic information.

[0051] In this embodiment, a predefined structured template can be traversed, and a first position can be determined according to preset identifiers or search rules. Subsequently, corresponding content can be added to the first position based on key basic information. The above is merely an illustrative example and is not intended to limit all possible cases of adding key basic information; it is simply not exhaustive.

[0052] The second position refers to a specific content insertion area reserved within a predefined structured template for key basic information.

[0053] In this embodiment, a predefined structured template can be traversed, and the second position can be determined according to preset identifiers or search rules. Subsequently, corresponding content can be added to the second position based on historical interest information. The above is merely an illustrative example and is not intended to limit all possible scenarios for adding historical interest information; it is simply not exhaustive.

[0054] In this embodiment, a predefined structured template can be used as the pattern specification. Key basic information, historical interest information, and other content from the predefined structured template, already added, can be concatenated according to requirements to generate complete prompt words. The above is merely an illustrative example and does not limit the possibilities for generating prompt words; it is simply not exhaustive.

[0055] Thus, by summarizing and statistically analyzing interest information records, historical interest information can efficiently and accurately reflect users' true interests within a specific time period, providing the model with the most representative input features and reducing noise interference. A unified template structure ensures consistent input format for all samples, suitable for large-scale model understanding and learning, and beneficial for improving model generalization and fitting capabilities. Each prompt word integrates the user's basic attributes and interest behaviors, fully reflecting individual differences and stimulating the personalized reasoning ability of large-scale models, thereby improving the accuracy of interest prediction. Through automated data processing and concatenation, the entire prompt word generation process is efficient and scalable, facilitating large-scale production of training data and rapid adaptation to business changes without cumbersome manual intervention.

[0056] In some embodiments, the predefined structure template further includes: output indication information, which is used at least to indicate the difference between target interest information and historical interest information.

[0057] In this embodiment of the disclosure, the output indication information refers to a portion of content added to a predefined structured template. This content, expressed in natural language or structured language, explicitly instructs the model to actively focus on differences or changes between historical and future interests, enabling more forward-looking inferences about potential new user needs and trends. For example, the output indication information may include: "Inferring new user interests for the next week based on the user's basic information and historical interests" and "The predicted interest points cannot have appeared in historical interests," etc. Specifically, the output indication information may also include other content that needs to be considered by the model, such as the reference data range, prediction result content, and prediction result format. The above is merely an illustrative example and does not constitute a limitation on all possible scenarios for the output indication information; it is simply not exhaustive.

[0058] In this way, by outputting instructional information, the model can proactively consider and infer the natural evolution or changes in user interests. This helps the model discover potential new interests of users, improve the diversity and novelty of recommendations, and avoid the "information cocoon" problem. Clearly distinguishing between historical and target interests helps the model focus on the dynamic changes and development trajectories of interests, thereby more accurately reflecting the true trends of user behavior.

[0059] In some implementations, predefined structured templates may include:

[0060] "Suppose you are a recommendation system expert, you would infer a user's new interests for the coming week based on their basic information and historical interests. Here is the text snippet you would need to answer the question."

[0061] ----------------------

[0062] User basic information: {First position}

[0063] ----------------------

[0064] User's historical interests: {Second position}

[0065] ----------------------

[0066] When answering questions, please follow these rules:

[0067] 1. Please provide a correct, comprehensive, and concise answer, highlighting the key points. Do not deviate from the content of the question and the text fragment and provide an excessive response.

[0068] 2. 'Points of interest' and 'degree of liking' are part of the prediction results, please reflect them in your answer.

[0069] 3. The value corresponding to 'degree of liking' is a floating-point number between 0 and 1. The larger the number, the deeper the degree of liking. Please reflect this in your answer.

[0070] 4. The predicted 'points of interest' cannot have appeared in the historical interests list; please indicate this in your answer.

[0071] The first position is used to add key basic information; the second position is used to add historical interest information.

[0072] In some implementations, the training data used to train the model to be trained may include:

[0073] "input":"",

[0074] "instruction":"\nAssuming you are a recommendation system expert, you would infer a user's new interests for the coming week based on their basic information and historical interests. Below is the text snippet you need to answer the question.\n----------------------\nBasic User Information:……\n----------------------\nUser Historical Interests: ##Interest Point: Food Reviews; Liking Level: 0.59 ##Interest Point: Showing Affection; Liking Level: 0.41;\n\n----------------------\nWhen answering the question, please adhere to the following rules:\n1. Please provide correct, comprehensive, and concise answers, highlighting the key points. Do not deviate excessively from the question and text snippet.\n2. `Interest Points` and `Liking Levels` are part of the prediction result; please reflect this in your answer.\n3. The value corresponding to `Liking Level` is a floating-point number between 0 and 1. The larger the number, the deeper the liking. Please reflect this in your answer.\n4. The predicted `Interest Points` cannot have appeared in the historical interests. Please reflect this in your answer."

[0075] "output":"##Interest Point: Showing off good looks; Liking Level: 0.37##Interest Point: Showing off hairstyle; Liking Level: 0.37##Interest Point: Cute baby; Liking Level: 0.33"

[0076] }".

[0077] It should be noted that the above training data is only an example and is not intended to limit all possible cases of training data; it is simply not exhaustive.

[0078] In some implementations, a low-rank adaptation (Lora) fine-tuning method can be used to load a lightweight model, fine-tune it using training data, and then deploy the fine-tuned model offline to infer new user interests at the user granular level in batches offline.

[0079] This disclosure provides a model-based method for interest inference. Figure 2This is a flowchart illustrating a model-based interest inference method according to an embodiment of the present disclosure. This model-based interest inference method can be applied to a model-based interest inference device. The model-based interest inference device is located in an electronic device. The electronic device includes, but is not limited to, fixed devices and / or mobile devices. For example, fixed devices include, but are not limited to, servers, which can be cloud servers or ordinary servers. Mobile devices include, but are not limited to, content recommendation devices, which can be mobile phones, tablets, in-vehicle terminals, etc. In some possible implementations, the model-based interest inference method can also be implemented by a processor calling computer-readable instructions stored in memory. Figure 2 As shown, this model-based interest inference method includes:

[0080] S201. Obtain key basic information about the target object and record the target object's interest information in the first time period;

[0081] S202. Generate prompt words based on key basic information, interest information records, and predefined structured templates;

[0082] S203. Input the prompt words into the pre-trained interest inference model;

[0083] S204. Obtain the target interest information of the target object in the second time period output by the interest inference model.

[0084] In this embodiment, key basic information can be directly queried from user registration information tables, profiling systems, etc., based on the target object. Alternatively, it can be automatically obtained through database queries, API calls, and other means. Furthermore, interest information records can be extracted from user behavior logs and their corresponding content categories or tags. Specifically, aggregation statistics can be performed based on content type and behavior type to obtain interest information records. The above are merely illustrative examples and are not intended to limit all possible scenarios for obtaining target object information; they are simply not exhaustive.

[0085] In this embodiment, user basic attributes and interest behaviors can be filled into the corresponding positions of a predefined structured template and concatenated into complete natural language prompts. The above is merely an illustrative example and is not intended to limit all possible scenarios for generating prompts; it is simply not exhaustive.

[0086] In this embodiment, an interest inference model can be loaded first, and prompt words can be directly input using an interface or script. The model then performs inference based on the input information. The above is merely an illustrative example and is not intended to limit all possible cases for the input model; it is simply not exhaustive.

[0087] In this embodiment, the target interest information returned by the model can be acquired and organized. Specifically, the target interest information can be text or structured information such as tags and category numbers. The above is merely an illustrative example and is not intended to limit all possible scenarios for acquiring target interest information; it is simply not exhaustive.

[0088] The technical solution of this disclosure, through predefined structured templates and automated data processing, enables large-scale, standardized batch operations for the entire interest inference process, avoiding tedious manual labeling and rule configuration and improving inference efficiency. By combining static basic user information with dynamic historical interests, the interest inference model can more comprehensively grasp individual differences and behavioral preferences, resulting in more personalized and accurate predictions. The predefined structured templates and unified processing flow facilitate expansion to other interest inference, user profiling, and personalized recommendation scenarios, demonstrating strong versatility and reusability. By focusing on historical interest information and future interest changes, the model can capture interest migration, uncover potential new demands, and enhance content distribution and product innovation capabilities.

[0089] In some embodiments, the interest information record includes a first point of interest and the degree to which the target object likes the first point of interest.

[0090] Here, a point of interest refers to a specific topic, tag, or category of interest. In this embodiment of the disclosure, a point of interest may refer to a video tag or video category, etc.

[0091] Here, the first point of interest refers to the main point of interest shown by the user within a first time period. In this embodiment of the disclosure, the first point of interest represents the video tag or video category that the user pays attention to within the first time period.

[0092] Here, the degree of liking refers to the intensity or preference score of a user's liking for a particular point of interest. In this embodiment of the disclosure, the degree of liking can be measured in various ways. For example, a user's viewing time, number of views, likes, favorites, comments, and reposts can be weighted and statistically analyzed, and finally converted into a score or level, which is then used as the degree of liking.

[0093] In this embodiment of the disclosure, user historical behavior data can be collected, and for each type of video tag or category, the relevant behavior volume of the user can be counted, and finally a structured record of interest points and degree of liking can be generated.

[0094] In this way, the level of liking can accurately reflect user interests, and this level can be dynamically updated according to changes in user behavior, making the interest profile more realistic. Furthermore, by comparing changes in liking levels over different time periods, shifts in user interests can be captured in a timely manner, allowing for dynamic adjustments to content delivery strategies. Based on interests and liking levels, users can be segmented into multiple interest groups, supporting differentiated operations.

[0095] In some embodiments, the target interest information includes at least one second point of interest and the degree to which the target object likes the second point of interest, which is different from the first point of interest.

[0096] The second point of interest refers to the video tags or video categories that the model predicts users are most likely to show interest in during the second time period.

[0097] In this embodiment of the disclosure, the second point of interest is a prediction of the future and is required to be different from the first point of interest; that is, it is not a primary point of interest that the user has previously shown, but rather a potential shift in interest or a new point of interest. Furthermore, for each second point of interest, the system not only needs to predict what content tag / category it is, but also needs to provide the user's possible "degree of liking," that is, the intensity, probability, score, or level of preference.

[0098] In this way, instead of simply speculating on users' past preferences, we can proactively explore users' potential new interests, broaden their content experience, and prevent them from being bound by their past interests in the long run, which is conducive to the healthy development of the information ecosystem. By predicting and capturing user interest trends and changes in a timely manner, we can shift from passive recommendation to proactive guidance, thereby increasing user activity and the exposure of new content.

[0099] In some embodiments, the target interest information further includes at least one third point of interest and the target object's liking for the third point of interest, wherein the third point of interest is the same as the first point of interest.

[0100] The third point of interest refers to the content categories or tags that users may continue to show interest in during the second time period, and these points of interest are consistent with the first point of interest in the historical period.

[0101] In this embodiment of the disclosure, the third point of interest is a prediction of the future and is required to be the same as the first point of interest, that is, it refers to the user's main interests that have been previously expressed. Furthermore, for each third point of interest, the system needs to predict the possible changes in the user's level of liking for the same point of interest over a future time period.

[0102] In this way, we can both detect and uncover new user interests, as well as identify and quantify users' persistent core interests, thus constructing a more comprehensive and accurate user interest profile. Since some users have long-term, stable preferences for certain content, only recommending new interests can cause them to miss out on content they truly want. By using third-party interest points, we can continuously recommend content related to users' core interests, which helps maintain user activity and platform stickiness. During training and inference, the model can both detect changes in interests and predict the maintenance and intensity of those interests, improving the model's generalization and interpretability. Simultaneously, based on the ratio of persistent to new interests, we can dynamically adjust content supply, optimize the content pool structure, and improve the overall health of the content ecosystem.

[0103] In some embodiments, the duration of the second time period is greater than the duration of the first time period.

[0104] In this embodiment of the disclosure, the duration of the second time period can be longer than that of the first time period, that is, the time window for predicting future interests is longer than the historical interest analysis window, thereby using a shorter historical window to predict future interest changes over a longer period of time.

[0105] In this way, predicting users' interests over a longer period of time is more in line with the long-term operational needs of actual business, and it is easier to capture the evolution and migration patterns of user interests, so as to prepare for content supply, recommendation strategies, etc. in advance.

[0106] In some embodiments, the duration of the second time period is the same as the duration of the first time period.

[0107] In this embodiment of the disclosure, the duration of the second time period can be the same as that of the first time period, that is, the time window for future interest prediction is the same as the historical interest analysis window, thereby realizing periodic and window-consistent behavior modeling and prediction.

[0108] In this way, the input and output time windows are consistent, facilitating model training, validation, and comparison. Training samples can be generated in batches, and the input and output dimensions, aggregation methods, etc., of the models are completely consistent, making it easy to reproduce and optimize. The same data structure can be used directly to process historical and prediction windows, reducing development and maintenance costs. The prediction target is consistent with the input window, which is more conducive to analyzing and capturing short-term cyclical changes in user interests and behavioral inertia.

[0109] In some embodiments, generating prompt words based on key basic information, interest information records, and a predefined structured template includes: generating historical interest information of the target object in a first time period based on the interest information records; adding key basic information to a first position in the predefined structured template; adding historical interest information to a second position in the predefined structured template; and generating prompt words based on the content of the first and second positions.

[0110] In this embodiment of the disclosure, historical interest information can be automatically concatenated and formatted from structured data into a text description based on interest information records, thereby obtaining historical interest information that is easy to understand and process using natural language. The above is merely an illustrative example and is not intended to limit all possible scenarios for generating historical interest information; it is simply not exhaustive.

[0111] In this embodiment, a predefined structured template can be traversed, and a first position can be determined according to preset identifiers or search rules. Subsequently, corresponding content can be added to the first position based on key basic information. The above is merely an illustrative example and is not intended to limit all possible cases of adding key basic information; it is simply not exhaustive.

[0112] In this embodiment, a predefined structured template can be traversed, and the second position can be determined according to preset identifiers or search rules. Subsequently, corresponding content can be added to the second position based on historical interest information. The above is merely an illustrative example and is not intended to limit all possible scenarios for adding historical interest information; it is simply not exhaustive.

[0113] In this embodiment, key basic information added to the first position, historical interest information added to the second position, and other content from a predefined structured template can be concatenated to generate a complete prompt word. The above is merely an illustrative example and is not intended to limit all possible scenarios for generating prompt words; it is simply not exhaustive.

[0114] In this way, all prompts are generated according to a unified structure, reducing descriptive ambiguity and stylistic differences, and improving the system and model's efficiency in understanding the input. Adding key basic information and historical interests to the first and second positions respectively helps the model distinguish different data, contributing to improved accuracy in downstream prediction and recommendation tasks. Furthermore, through predefined structured templates, different business scenarios and information combinations can be adapted simply by adjusting or adding to the template structure.

[0115] In some embodiments, the predefined structure template further includes: output indication information, which is used at least to indicate the difference between target interest information and historical interest information.

[0116] In this embodiment, the output instruction information guides the downstream model or system to focus on comparing the differences between the target interest information and historical interest information when generating results. Specifically, the output instruction information is essentially a descriptive or commanding text, which is directly incorporated into the final generated prompt as part of a template.

[0117] In this way, by directly describing the output requirements in the template, redundant, irrelevant, or repetitive content is avoided from the model output, allowing the model to focus on key points such as discovering new interests and improving the practical value of the output content. By automatically comparing historical interests with future predicted interests, it is beneficial to analyze interest migration, user growth, and changes in content preferences, accurately capturing new and lost interests, and supporting business actions such as content pool optimization, user segmentation, and precise push notifications.

[0118] Figure 3 Another flowchart illustrating the model-based interest inference method according to an embodiment of this disclosure is shown. Figure 3 As shown, the method may include:

[0119] S301. Obtain user's historical interest suggestions.

[0120] S302. Fine-tune the prompt text based on Lora.

[0121] S303. Perform batch generative reasoning based on the fine-tuning results to obtain the current target interest information.

[0122] S304. Generate online strategies based on target interest information.

[0123] In some implementations, a prompt can be generated first based on the user's historical interest data and basic information. The purpose of the prompt is to provide context for the pre-trained large model, allowing the model to understand user needs and predict interests. For example, the prompt may include basic user information, historical interests, etc.

[0124] Furthermore, LoRa technology can be used to fine-tune the pre-trained model for specific domains. The fine-tuned model can better understand user interests and generate more accurate recommendation results.

[0125] Furthermore, the finely tuned model can be applied to offline scenarios to generate batches of new interest predictions for users. Specifically, the model can predict new points of interest that users may be interested in based on the Prompt input and quantify their level of liking.

[0126] Finally, the predicted set of new interests can be combined with real-time data to dynamically adjust recommendation weights, thereby generating an online strategy. This online strategy can adjust the recommendation logic based on the user's current behavior and the context, resulting in more timely and personalized recommendations and thus improving the user experience.

[0127] Figure 4 A schematic diagram illustrating the principle of the low-rank adaptation fine-tuning method according to an embodiment of this disclosure is shown, as follows: Figure 4 As shown, LoRa fine-tuning techniques can include pre-trained weight parameters and the addition of sidepaths. Specifically, a Prompt of dimension d can be used as input x, and the matrix W ∈ R of the pre-trained weight parameters can be... d×d This means that the input dimension and the input dimension of the weight matrix are the same. Furthermore, the newly added bypass includes matrices A and B, whose sizes are controlled by a preset dimensionality reduction factor r, i.e., A∈R. d×r , B∈R r×d The initial value of matrix A is a Gaussian distribution, i.e., A = N(0, σ). 2), where N represents a Gaussian distribution and σ represents the standard deviation of the Gaussian distribution, which can be used to control the initial value range of matrix A. Specifically, during training, the parameters of matrix A are updated progressively, responsible for capturing newly added semantics or patterns during fine-tuning. Furthermore, the initial value of matrix B is 0, ensuring that the bypass matrix does not introduce additional computational results at the start of training. Specifically, during training, the parameters of matrix B are also updated, working together with matrix A.

[0128] In some implementations, during fine-tuning, only the parameters of matrices A and B can be updated, while keeping the matrix W of the pre-trained weight parameters unchanged, thereby significantly reducing the overhead of parameter updates and avoiding changes to the structure of the pre-trained model.

[0129] In some implementations, during inference, the matrix W of pre-trained weight parameters can be merged with matrices A and B, and inference can be performed through parameter reparameterization to obtain the hidden layer output h, thereby ensuring that no additional computational overhead is introduced.

[0130] This disclosure provides a training apparatus for an interest inference model, such as... Figure 5 As shown, the device may include: a first information acquisition module 501, used to acquire key basic information of a known object, interest information records of the known object in a first time period, and target interest information of the known object in a second time period; a first prompt word generation module 502, used to generate prompt words based on the key basic information, interest information records, and a predefined structured template; and a model training module 503, used to input the prompt words into a model to be trained, train the model to be trained, and obtain an interest inference model, so that the interest inference model outputs the target interest information of the known object in the second time period.

[0131] In some embodiments, the first prompt word generation module 502 includes: a first historical information generation submodule, used to generate historical interest information of a known object in a first time period based on interest information records; a first basic information addition submodule, used to add key basic information to a first position of a predefined structure template; a first historical information addition submodule, used to add historical interest information to a second position of a predefined structure template; and a training prompt word generation submodule, used to generate prompt words based on the content of the first and second positions.

[0132] In some embodiments, the predefined structure template further includes: output indication information, which is used at least to indicate the difference between target interest information and historical interest information.

[0133] The specific functions and examples of each module and sub-module in the training device of the interest inference model in this embodiment can be found in the relevant descriptions of the corresponding steps in the above-described embodiment of the interest inference model training method, and will not be repeated here.

[0134] The training device for the interest inference model in this embodiment can significantly reduce the difficulty of acquiring training data by fully utilizing existing user attribute and behavioral data on the platform without additional annotation. Standardizing input through predefined structured templates enhances the model's understanding of recommendation domain information, enabling better transfer and generalization. By combining information from the first and second time periods, the model can capture the evolution of interests, resulting in more accurate predictions. Introducing individualized historical and attribute information during model training enables fine-grained, highly personalized interest prediction, helping recommendation systems break free from traditional "cocoon" limitations. This not only improves the efficiency and accuracy of interest exploration but also helps recommendation systems overcome the limitations of traditional information cocoons, proactively discovering and recommending fresh content to users, thereby increasing user dwell time and conversion rates, and enhancing user stickiness and activity.

[0135] This disclosure provides a model-based interest inference device, such as... Figure 6 As shown, the device may include: a second information acquisition module 601, used to acquire key basic information of the target object and interest information records of the target object in a first time period; a second prompt word generation module 602, used to generate prompt words based on the key basic information, interest information records and predefined structured templates; a model input module 603, used to input the prompt words into a pre-trained interest inference model; and a model output module 604, used to acquire the target interest information of the target object output by the interest inference model in the second time period.

[0136] In some embodiments, the interest information record includes a first point of interest and the degree to which the target object likes the first point of interest.

[0137] In some embodiments, the target interest information includes at least one second point of interest and the degree to which the target object likes the second point of interest, which is different from the first point of interest.

[0138] In some embodiments, the target interest information further includes at least one third point of interest and the target object's liking for the third point of interest, wherein the third point of interest is the same as the first point of interest.

[0139] In some embodiments, the duration of the second time period is greater than the duration of the first time period.

[0140] In some embodiments, the duration of the second time period is the same as the duration of the first time period.

[0141] In some embodiments, the second prompt word generation module 602 includes: a first historical information generation submodule, used to generate historical interest information of the target object in a first time period based on interest information records; a second basic information adding submodule, used to add key basic information to a first position of a predefined structure template; a second historical information adding submodule, used to add historical interest information to a second position of the predefined structure template; and a predictive prompt word generation submodule, used to generate prompt words based on the content of the first and second positions.

[0142] In some embodiments, the predefined structure template further includes: output indication information, which is used at least to indicate the difference between target interest information and historical interest information.

[0143] The specific functions and examples of each module and submodule in the model-based interest inference device of this disclosure can be found in the relevant descriptions of the corresponding steps in the above-described model-based interest inference method embodiments, and will not be repeated here.

[0144] The model-based interest inference device in this embodiment enables large-scale, standardized operations through predefined structured templates and automated data processing, avoiding tedious manual labeling and rule configuration and improving efficiency. By combining static user information with dynamic historical interests, the interest inference model can more comprehensively grasp individual differences and behavioral preferences, resulting in more personalized and accurate predictions. The predefined structured templates and unified processing flow facilitate expansion to other interest inference, user profiling, and personalized recommendation scenarios, demonstrating strong versatility and reusability. By focusing on historical interest information and future interest changes, the model can capture interest migration, uncover potential new demands, and enhance content distribution and product innovation capabilities.

[0145] This disclosure provides a scenario illustration of a training method for an interest inference model, such as... Figure 7 As shown.

[0146] As previously described, the method for training the interest inference model provided in this disclosure is applied to electronic devices. These electronic devices are intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers.

[0147] Specifically, the electronic device may perform the following operations:

[0148] The system acquires key basic information of a known object, records of the object's interest information in the first time period, and records of the object's target interest information in the second time period. Based on the key basic information, interest information records, and a predefined structured template, it generates prompt words. The prompt words are then input into the model to be trained to obtain an interest inference model, which outputs the target interest information of the known object in the second time period.

[0149] It should be understood that Figure 7 The scene diagrams shown are merely illustrative and not restrictive; those skilled in the art can interpret them based on... Figure 7 Even with various obvious changes and / or substitutions to the examples, the resulting technical solutions still fall within the scope of this disclosure.

[0150] This disclosure provides a scenario illustration of a model-based interest inference method, such as... Figure 8 As shown.

[0151] As previously described, the model-based interest inference method provided in this disclosure is applied to electronic devices. These electronic devices are intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers.

[0152] Specifically, the electronic device may perform the following operations:

[0153] Obtain key basic information of the target object and record the target object's interest information in the first time period; generate prompt words based on the key basic information, interest information records, and predefined structured templates; input the prompt words into a pre-trained interest inference model; obtain the target object's target interest information output by the interest inference model in the second time period.

[0154] It should be understood that Figure 8 The scene diagrams shown are merely illustrative and not restrictive; those skilled in the art can interpret them based on... Figure 8 Even with various obvious changes and / or substitutions to the examples, the resulting technical solutions still fall within the scope of this disclosure.

[0155] The acquisition, storage, and application of user personal information involved in the technical solution disclosed herein comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0156] According to embodiments of this disclosure, this disclosure also provides an electronic device, a readable storage medium, and a computer program product.

[0157] Figure 9A schematic block diagram of an example electronic device 900 that can be used to implement embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0158] like Figure 9 As shown, device 900 includes a computing unit 901, which can perform various appropriate actions and processes based on a computer program stored in read-only memory (ROM) 902 or a computer program loaded from storage unit 908 into random access memory (RAM) 903. RAM 903 may also store various programs and data required for the operation of device 900. The computing unit 901, ROM 902, and RAM 903 are interconnected via bus 904. Input / output (I / O) interface 905 is also connected to bus 904.

[0159] Multiple components in device 900 are connected to I / O interface 905, including: input unit 906, such as keyboard, mouse, etc.; output unit 907, such as various types of monitors, speakers, etc.; storage unit 908, such as disk, optical disk, etc.; and communication unit 909, such as network card, modem, wireless transceiver, etc. Communication unit 909 allows device 900 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0160] The computing unit 901 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 901 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, digital signal processors (DSPs), and any suitable processor, controller, microcontroller, etc. The computing unit 901 performs the various methods and processes described above, such as methods for training interest inference models and / or model-based interest inference methods. For example, in some embodiments, methods for training interest inference models and / or model-based interest inference methods can be implemented as computer software programs tangibly contained in a machine-readable medium, such as storage unit 908. In some embodiments, part or all of the computer program can be loaded and / or installed on device 900 via ROM 902 and / or communication unit 909. When the computer program is loaded into RAM 903 and executed by computing unit 901, one or more steps of the interest prediction model training method and / or model-based interest prediction method described above can be performed. Alternatively, in other embodiments, computing unit 901 can be configured to perform the interest prediction model training method and / or model-based interest prediction method by any other suitable means (e.g., by means of firmware).

[0161] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-chip (SoCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0162] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0163] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory, read-only memory, erasable programmable read-only memory (EPROM), flash memory, optical fiber, compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0164] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a cathode ray tube (CRT) or liquid crystal display (LCD) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0165] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0166] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact via communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other. Servers can be cloud servers, servers in distributed systems, or servers incorporating blockchain technology.

[0167] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this disclosure can be achieved, and this is not limited herein.

[0168] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A training method for an interest inference model, comprising: Acquire key basic information of a known object, records of the known object's interest information in the first time period, and target interest information of the known object in the second time period; Based on the key basic information, the interest information record, and the predefined structured template, generate prompt words; The prompt words are input into the model to be trained, and the model is trained to obtain an interest inference model, so that the interest inference model outputs the target interest information of the known object in the second time period.

2. The method according to claim 1, wherein, The step of generating prompt words based on the key basic information, the interest information record, and the predefined structured template includes: Based on the interest information, the historical interest information of the known object in the first time period is generated; Add the key basic information to the first position of the predefined structure template; Add the historical interest information to the second position of the predefined structure template; The prompt word is generated based on the content of the first position and the second position.

3. The method according to claim 2, wherein, The predefined structure template also includes: output indication information, which is at least used to indicate the difference between the target interest information and the historical interest information.

4. A model-based method for interest inference, comprising: Acquire key basic information about the target object and record the target object's interest information in the first time period; Based on the key basic information, the interest information record, and the predefined structured template, generate prompt words; Input the prompt words into a pre-trained interest inference model; Obtain the target interest information of the target object in the second time period, as output by the interest inference model.

5. The method according to claim 4, wherein, The interest information record includes a first point of interest and the degree to which the target object likes the first point of interest.

6. The method according to claim 5, wherein, The target interest information includes at least one second point of interest and the degree to which the target object likes the second point of interest, wherein the second point of interest is different from the first point of interest.

7. The method according to claim 6, wherein, The target interest information also includes at least one third point of interest and the degree to which the target object likes the third point of interest, wherein the third point of interest is the same as the first point of interest.

8. The method according to any one of claims 4 to 7, wherein, The duration of the second time period is greater than the duration of the first time period.

9. The method according to any one of claims 4 to 7, wherein, The duration of the second time period is the same as the duration of the first time period.

10. The method according to claim 4, wherein, The step of generating prompt words based on the key basic information, the interest information record, and the predefined structured template includes: Based on the interest information, the historical interest information of the target object in the first time period is generated; Add the key basic information to the first position of the predefined structure template; Add the historical interest information to the second position of the predefined structure template; The prompt word is generated based on the content of the first position and the second position.

11. The method according to claim 10, wherein, The predefined structure template also includes: output indication information, which is at least used to indicate the difference between the target interest information and the historical interest information.

12. A training device for an interest inference model, comprising: The first information acquisition module is used to acquire key basic information of a known object, interest information records of the known object in a first time period, and target interest information of the known object in a second time period. The first prompt word generation module is used to generate prompt words based on the key basic information, the interest information record, and the predefined structured template; The model training module is used to input the prompt words into the model to be trained, train the model to obtain an interest inference model, and enable the interest inference model to output the target interest information of the known object in the second time period.

13. A model-based interest inference device, comprising: The second information acquisition module is used to acquire key basic information of the target object and the interest information record of the target object in the first time period. The second prompt word generation module is used to generate prompt words based on the key basic information, the interest information record, and the predefined structured template; The model input module is used to input the prompt words into a pre-trained interest inference model; The model output module is used to obtain the target interest information of the target object in the second time period, which is output by the interest inference model.

14. An electronic device comprising: At least one processor; as well as A memory that is communicatively connected to at least one processor; wherein, The memory stores instructions that can be executed by at least one processor to enable the at least one processor to perform the method of any one of claims 1-11.

15. A non-transitory computer-readable storage medium storing computer instructions, wherein, Computer instructions are used to cause a computer to perform the method according to any one of claims 1-11.

16. A computer program product comprising a computer program stored on a storage medium, wherein the computer program, when executed by a processor, implements the method according to any one of claims 1-11.

Images