User tag generation method, terminal device and computer program product

Abstract

The invention is suitable for the technical field of computers, and provides a user tag generation method, terminal equipment and a computer program product, and the method comprises the steps: obtaining an interaction record between a user and an intelligent agent; analyzing the interaction record, and generating a first label of the user in a first time dimension; generating second tags of the user in different time dimensions based on the first tag and historical tags of the user in different time dimensions; different time dimensions respectively correspond to different durations, the different time dimensions do not include the first time dimension, and the duration corresponding to the first time dimension is the shortest; and generating a target tag of the user based on the first tag and each second tag. According to the method and the device, diversified data can be obtained through the interaction record of the user and the intelligent agent to generate the accurate first tag, and then the final total tag of the user can be generated by combining the first tag of the user and the historical tags under different time dimensions, so that the generation accuracy of the user tag is improved.

Description

Technical Field

[0001] The present application belongs to the field of computer technology, and in particular, relates to a user tag generation method, a terminal device, and a computer program product. Background Art

[0002] User tags are an abstract summary of user information and features, used to describe user features and distinguish user groups. User tags are formed by analyzing user behavior data within the platform and extracting distinctive features.

[0003] The prior art usually generates corresponding user tags based on the user's behavior data on the platform, such as multimedia operation behavior and APP operation behavior, etc. It can be seen that the prior art has the problem of obtaining single data and reducing the accuracy of generating user tags. Summary of the invention

[0004] The embodiments of the present application provide a user tag generation method, a terminal device, and a computer program product to solve the problem in the prior art that the acquired data is single, which reduces the accuracy of user tag generation.

[0005] In a first aspect, an embodiment of the present application provides a method for generating a user tag, comprising:

[0006] Obtain the interaction records between the user and the agent;

[0007] Analyze the interaction record to generate a first label of the user in a first time dimension;

[0008] generating a second tag of the user in the different time dimensions based on the first tag and the historical tags of the user in the different time dimensions; the different time dimensions respectively correspond to different durations, and the first time dimension corresponds to the shortest duration;

[0009] A target tag of the user is generated based on the first tag and each of the second tags.

[0010] Optionally, analyzing the interaction record to generate a first label of the user in a first time dimension includes:

[0011] Obtain a historical tag set in the first time dimension within a historical time period;

[0012] Extracting keywords from the historical tag set to obtain a first prompt word;

[0013] The first prompt word and the interaction record are input into a first large language model for processing to obtain the first label.

[0014] Optionally, the tag generation order of the first time dimension is first priority; and generating the second tag of the user in the different time dimension based on the first tag and the historical tags of the user in the different time dimensions includes:

[0015] Determine the order of generating labels for the different time dimensions;

[0016] When the label generation order of the target time dimension is second, a second label of the target time dimension is generated based on the first label and the historical label of the target time dimension; the target time dimension is any one of the different time dimensions.

[0017] Optionally, after determining the order of generating labels for the different time dimensions, the method further includes:

[0018] When the label generation order of the target time dimension is not the second priority, the second label of the target time dimension is generated based on the second label of the previous priority time dimension and the historical label of the target time dimension; the previous priority time dimension refers to the last time dimension before the label generation order of the target time dimension.

[0019] Optionally, the generating a second label of the target time dimension based on the first label and the historical label of the target time dimension includes:

[0020] Inputting the first label into the second largest language model for processing to obtain an initial label attribute corresponding to the target time dimension;

[0021] Performing a correlation evaluation on the first label and the historical label of the target time dimension to obtain an evaluation result;

[0022] The first label and the historical label are processed based on the evaluation result and the initial label attribute to obtain a second label of the target time dimension.

[0023] Optionally, the processing the first label and the historical label based on the evaluation result and the initial label attribute to obtain the second label of the target time dimension includes:

[0024] If the evaluation result is relevant, inputting the first label and the historical label of the target time dimension into the second largest language model for processing to obtain a target label attribute corresponding to the target time dimension;

[0025] The first label and the historical labels of the target time dimension are filtered based on the target label attribute to obtain a second label of the target time dimension.

[0026] Optionally, the filtering and processing the first label and the historical label of the target time dimension based on the target label attribute to obtain the second label of the target time dimension includes:

[0027] The first tag and the historical tags of the target time dimension are filtered based on the target tag attribute to obtain at least one first subtag and at least one historical subtag that match the target tag attribute; each of the first subtags is any one of the first tags, and each of the historical subtags is any one of the historical tags of the target time dimension;

[0028] If the first target subtag and the second target subtag are conflicting tags, the second target subtag is deleted; the first target subtag is any one of at least one of the first subtags, and the second target subtag is any one of at least one of the historical subtags;

[0029] A second label of the target time dimension is generated according to at least one of the first sub-labels and other sub-labels; the other sub-labels refer to historical sub-labels other than the second target sub-label.

[0030] Optionally, the processing the first label and the historical label based on the evaluation result and the initial label attribute to obtain the second label of the target time dimension includes:

[0031] If the evaluation result is irrelevant, the first label and the historical label of the target time dimension are screened based on the initial label attribute to obtain a second label of the target time dimension.

[0032] In a second aspect, an embodiment of the present application provides a user tag generating device, including:

[0033] A first acquisition unit, used to acquire the interaction record between the user and the agent;

[0034] A first generating unit, configured to analyze the interaction record and generate a first label of the user in a first time dimension;

[0035] A second generating unit is configured to generate a second tag of the user in the different time dimensions based on the first tag and the historical tags of the user in the different time dimensions; the different time dimensions respectively correspond to different time lengths, and the first time dimension corresponds to the shortest time length;

[0036] The third generating unit is configured to generate a target tag for the user based on the first tag and each of the second tags.

[0037] In a third aspect, an embodiment of the present application provides a terminal device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the user tag generation method as described in any one of the first aspects above is implemented.

[0038] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the user tag generation method as described in any one of the above-mentioned first aspects is implemented.

[0039] In a fifth aspect, an embodiment of the present application provides a computer program product. When the computer program product is run on a terminal device, the terminal device can execute the user tag generation method described in any one of the first aspects above.

[0040] Compared with the prior art, the embodiments of the present invention have the following beneficial effects:

[0041] The embodiment of the present application provides a user tag generation method, which obtains the interaction record between the user and the intelligent agent; analyzes the interaction record to generate the first tag of the user in the first time dimension; generates the second tag of the user in different time dimensions based on the first tag and the historical tags of the user in different time dimensions; the durations corresponding to different time dimensions are different, and the duration corresponding to the first time dimension is the shortest; based on the first tag and each second tag, generates the user's target tag. Compared with the single data obtained by the prior art, since the intelligent agent can collect a large amount of data during the interaction with the user, such as the user's preferences, behavioral habits, needs, etc., the present application can obtain diversified data through the interaction record between the user and the intelligent agent, so that the single tag generated by the interaction record is more comprehensive. At the same time, the process of generating the first tag is based on the objective data of the actual interaction record between the user and the intelligent agent, which reduces subjective assumptions, thereby improving the objectivity and accuracy of the first tag. Afterwards, since the second tags of different time dimensions are generated by combining the first tag and the historical tags in different time dimensions, the consistency and comprehensiveness of the tags of different time dimensions are guaranteed. Finally, through the first label of the first time dimension and the second labels corresponding to different time dimensions, the user's behavioral performance in time dimensions of different lengths is comprehensively considered, from the shortest first time dimension to other time dimensions of different lengths, the dynamic changes of user behavior characteristics are fully captured, avoiding the one-sidedness of single time point analysis, so as to fully understand the user's behavior, habits and important information in different time dimensions, thereby ensuring the comprehensiveness of the final user labels generated subsequently, and thus improving the accuracy of user label generation. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0043] Figure 1 This is a flowchart of a method for generating user tags according to an embodiment of the present application;

[0044] Figure 2 is a flowchart of a method for generating user tags according to another embodiment of the present application;

[0045] Figure 3 is a flowchart of a method for generating user tags according to another embodiment of the present application;

[0046] Figure 4 is a flowchart of a method for generating user tags according to another embodiment of the present application;

[0047] Figure 5 It is a structural schematic diagram of a user tag generating device provided by an embodiment of the present application;

[0048] Figure 6 It is a structural diagram of a terminal device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0049] In the following description, specific details such as specific system structures, technologies, etc. are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present application. However, it should be clear to those skilled in the art that the present application may also be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to prevent unnecessary details from obstructing the description of the present application.

[0050] It should be understood that when used in the present specification and the appended claims, the term "comprising" indicates the presence of described features, wholes, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or combinations thereof.

[0051] It should also be understood that the term “and / or” used in the specification and appended claims refers to any and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0052] As used in the specification and appended claims of this application, the term "if" can be interpreted as "when" or "uponce" or "in response to determining" or "in response to detecting", depending on the context. Similarly, the phrase "if it is determined" or "if [described condition or event] is detected" can be interpreted as meaning "uponce it is determined" or "in response to determining" or "uponce [described condition or event] is detected" or "in response to detecting [described condition or event]", depending on the context.

[0053] In addition, in the description of the present application specification and the appended claims, the terms "first", "second", "third", etc. are only used to distinguish the descriptions and cannot be understood as indicating or implying relative importance.

[0054] References to "one embodiment" or "some embodiments" etc. described in the specification of this application mean that one or more embodiments of the present application include specific features, structures or characteristics described in conjunction with the embodiment. Therefore, the statements "in one embodiment", "in some embodiments", "in some other embodiments", "in some other embodiments", etc. that appear in different places in this specification do not necessarily refer to the same embodiment, but mean "one or more but not all embodiments", unless otherwise specifically emphasized in other ways. The terms "including", "comprising", "having" and their variations all mean "including but not limited to", unless otherwise specifically emphasized in other ways.

[0055] In practical applications, user tags are an abstract summary of user information and features, used to describe user features and distinguish user groups. User tags are formed by analyzing user behavior data within the platform and extracting differentiated features.

[0056] The prior art usually generates corresponding user tags based on the user's behavior data on the platform, such as multimedia operation behavior and APP operation behavior, etc. It can be seen that the prior art has the problem of obtaining single data and reducing the accuracy of generating user tags.

[0057] Based on this, the present application proposes a user tag generation method, which can obtain diversified data through the interaction records between users and intelligent agents and improve the accuracy of user tag generation.

[0058] See also Figure 1 , Figure 1 This is a flowchart of a method for generating a user tag provided in an embodiment of the present application. In the embodiment of the present application, the execution subject of the method for generating a user tag is a terminal device. The terminal device includes but is not limited to a notebook, a desktop computer, and a computer.

[0059] like Figure 1As shown, the user tag generation method provided in an embodiment of the present application may include S101 to S104, which are described in detail as follows:

[0060] In S101, the interaction record between the user and the agent is obtained.

[0061] In actual applications, when a user tag needs to be generated, relevant personnel can send a tag generation request to the terminal device.

[0062] In the embodiment of the present application, the terminal device detects the above-mentioned tag generation request may be: a preset operation for the terminal device is detected. The preset operation can be set according to actual needs and is not limited here. Exemplary, the preset operation may be clicking a preset control on the terminal device. Based on this, when the terminal device detects that its preset control is clicked, it indicates that the preset operation is detected, that is, the above-mentioned tag generation request is detected.

[0063] After detecting the above tag generation request, the terminal device can obtain the interaction record between the user and the intelligent agent in real time through the intelligent agent connected to it by wireless communication.

[0064] In practical applications, an agent is an entity with intelligent characteristics such as autonomy, responsiveness, and initiative. It can exist in a software environment and has the ability to perform specific tasks and interact with the outside world.

[0065] Intelligent agents can be presented on a variety of different carriers. They can exist independently in the form of software, or be integrated into various applications as software packages to assist in realizing specific functions. They can also be an important component of the software system, participating in the overall operation of the system and external services.

[0066] From the perspective of hardware carriers, intelligent entities can be attached to a variety of hardware, such as servers, smart phones, and smart wearable devices.

[0067] It should be noted that the intelligent agent has the ability to learn and adapt, and can continuously optimize its own behavior and decision-making methods based on the accumulated experience and new data. At the same time, through in-depth analysis of user data, the intelligent agent can provide personalized service experience for different users. It can also provide services uninterruptedly, without restrictions such as time and region. As long as the hardware carrier and network conditions allow, it can respond to user needs at any time. Therefore, the intelligent agent can collect a large amount of data during the interaction with the user, and this data can accurately represent the user's preferences, behavioral habits, needs, etc.

[0068] Based on this, in the embodiment of the present application, the terminal device can obtain the interaction record between the user and the intelligent agent. Compared with the existing method of obtaining data, since the interaction record between the user and the intelligent agent is the interaction information between the user and the intelligent agent in a specific situation, it has a strong personalized feature, and these records accurately reflect the user's needs, preferences, emotions and the specific scene at that time. At the same time, the interaction record is a dynamic process, which is constantly updated with the dialogue, operation and other behaviors between the user and the intelligent agent. Unlike the data acquisition method of regular collection or batch processing, it can reflect the user's latest ideas and behavior changes in real time. In addition, the interaction record also reflects the two-way interaction between the user and the intelligent agent. The user sends a request, instruction or feedback to the intelligent agent, and the intelligent agent responds according to the user input and may guide the user to provide further information. This two-way communication method is in sharp contrast to the traditional one-way data acquisition (such as collecting environmental data only through sensors). In summary, the terminal device can obtain diversified data related to the user through the interaction record between the user and the intelligent agent.

[0069] In the embodiment of the present application, the above-mentioned interaction record may be all chat records between the user and the agent in the historical time period. The historical time period may be determined according to actual needs and is not limited here. Exemplarily, the end time of the historical time period may be the time when the terminal device detects the tag generation request.

[0070] In S102, the interaction record is analyzed to generate a first label of the user in a first time dimension.

[0071] In the embodiment of the present application, the terminal device can input the interaction record into the first label generation model for analysis, thereby obtaining the first label of the user in the first time dimension. The first label is used to describe the user's daily behavior and habits.

[0072] In some possible embodiments, since the intelligent agent also has reasoning capabilities, the terminal device can analyze the interaction records through the reasoning capabilities of the intelligent agent to obtain the first label of the user in the first time dimension.

[0073] Compared with existing analysis methods, the reasoning function of the intelligent agent can use its internally stored knowledge and previously accumulated experience for reasoning. This knowledge can include rules, cases, semantic understanding and other forms. Unlike the traditional fixed algorithm analysis method, the intelligent agent can flexibly use this knowledge for reasoning according to different situations and problems. At the same time, when faced with incomplete information or uncertain environments, the intelligent agent can make reasonable inferences based on the existing knowledge system. In addition, the reasoning function of the intelligent agent has the characteristics of dynamically adapting to environmental changes. With the acquisition of new data, knowledge and changes in task requirements, the intelligent agent can continuously adjust its reasoning strategy. In summary, through the reasoning function of the intelligent agent, the terminal device can obtain a more accurate first label.

[0074] It should be noted that the first tag may include multiple tags.

[0075] In some possible embodiments, the duration of the first time dimension may be one day. Therefore, the first time dimension may be a day dimension, and its first label may be a day label.

[0076] Exemplarily, the first tag may include the following content:

[0077] 1. Daily schedule: the time the user wakes up and goes to bed every day.

[0078] 2. Eating habits: the user’s daily breakfast, lunch and dinner times and dietary preferences.

[0079] 3. Exercise habits: whether the user exercises every day and the specific time.

[0080] 4. Work / study time: the start and end time of the user’s work every day.

[0081] 5. Entertainment time: the time users spend on entertainment (such as watching TV, playing games, reading, etc.) every day.

[0082] 6. Transportation: The main means of transportation used by users every day.

[0083] 7. Social activities: the time and frequency of users’ daily social activities.

[0084] 8. Shopping habits: users’ daily shopping behaviors and preferences.

[0085] 9. Weather reaction: The user’s reaction to the weather of the day (e.g., likes rainy days, hates hot days).

[0086] 10. Emotional state: the user’s daily emotional fluctuations.

[0087] In an embodiment of the present application, the first label generation model is obtained by training a pre-constructed neural network model. The first label generation model can be obtained by training a pre-constructed neural network model based on a preset sample set. Each sample data in the preset sample set includes a sample interaction record and a sample label corresponding to the sample interaction record. When training the pre-constructed neural network model, the sample interaction record in each sample data is used as the input of the neural network model, and the sample label corresponding to the sample interaction record in each sample data is used as the output of the neural network model. Through training, the neural network model can learn the correspondence between all possible sample interaction records and sample labels, and use the trained neural network model as the first label generation model.

[0088] In practical applications, Large Language Models (LLM) are the first deep learning models trained based on massive text data, and large language models can not only generate natural language text, but also deeply understand the meaning of text and handle various natural language tasks, such as text summarization, question answering, translation, etc. Therefore, in order to further improve the accuracy of generating the first label, the terminal device can generate the first label through the following steps, which are detailed as follows:

[0089] Obtain a historical tag set in the first time dimension within a historical time period;

[0090] Extracting keywords from the historical tag set to obtain a first prompt word;

[0091] The first prompt word and the interaction record are input into a first large language model for processing to obtain the first label.

[0092] In this embodiment, the historical tag set includes multiple historical tags in the first time dimension within the historical time period.

[0093] Afterwards, the terminal device may perform keyword extraction on each history tag in the history tag set to obtain a first prompt word.

[0094] In practical applications, keyword extraction methods include but are not limited to: unsupervised keyword extraction methods and supervised keyword extraction methods.

[0095] Unsupervised keyword extraction specifically refers to a method of extracting keywords without manually annotated corpus, but using certain methods to find more important words in the text as keywords. This method first extracts candidate words, then scores each candidate word, and then outputs the topK candidate words with the highest scores as keywords.

[0096] Therefore, in this embodiment, the terminal device can perform word frequency statistics on the pre-stored historical tag set, that is, record the frequency of occurrence of each word, and determine each word with an occurrence frequency greater than or equal to a first threshold as a historical candidate phrase. The first threshold can be determined according to actual needs and is not limited here.

[0097] Afterwards, the terminal device can filter out words corresponding to the historical candidate phrase from the historical tag set, and determine the words that appear as candidate words corresponding to the test content. Afterwards, the terminal device scores each candidate word, and then outputs the topK candidate words with the highest scores as keywords corresponding to the test content.

[0098] It should be noted that there are different algorithms depending on the scoring strategy, such as keyword extraction based on statistical features (term frequency-inverse document frequency, TF-IDF), keyword extraction based on word graph model (TextRank) and keyword extraction based on topic model (Latent Dirichlet Allocation, LDA) and other algorithms.

[0099] Among them, the idea of ​​the keyword extraction algorithm based on statistical features is to extract the keywords of the document using the statistical information of the words in the document.

[0100] Keyword extraction based on the word graph model first needs to construct a language network graph of the document, then perform network graph analysis on the language, and look for important words or phrases on this graph. These phrases are the keywords of the document.

[0101] The topic-based keyword extraction algorithm mainly uses the properties of topic distribution in the topic model to extract keywords.

[0102] The supervised keyword extraction method treats the keyword extraction process as a binary classification problem. First, candidate words are extracted, and then each candidate word is labeled as either a keyword or not a keyword, and then a keyword extraction classifier is trained. After that, after receiving the historical tag set, the terminal device can extract all the candidate words in the historical tag set, and then use the trained keyword extraction classifier to classify each candidate word, and finally use the candidate word labeled as a keyword as a keyword.

[0103] In this embodiment, after obtaining the first prompt word, the terminal device may input the first prompt word and the interaction record into the first large language model for processing to obtain a first label.

[0104] Large language models include but are not limited to: Llama2 model and Chatgpt model.

[0105] It should be noted that the large language model uses a multi-layer stacked neural network structure, and each layer of the network is composed of multiple neurons. The input text sequence is processed and transformed through a series of network layers to finally obtain the output result of the model. Among them, the large language model can use the Transformer structure as the neural network structure. The Transformer structure is set with a self-attention mechanism.

[0106] In the large language model, the self-attention mechanism is implemented by performing matrix multiplication and softmax activation function operations on the input text sequence. For each position, the large language model calculates the similarity between it and other positions, and performs weighted average on the sequence according to the weight of the similarity. In this way, in different network layers, the large language model can learn different relationships in the sequence, such as the relationship between words, the relationship between sentences, etc.

[0107] It should be noted that the training process of the above-mentioned large language model includes two stages: pre-training and fine-tuning. In the pre-training stage, the large language model uses a large amount of text corpus (historical interaction records in the sample data set) for unsupervised learning to learn the language rules and semantic relationships in the text sequence. In the fine-tuning stage, the large language model uses a small amount of annotated data (such as the first prompt word and the historical label set) for supervised learning to optimize the model parameters.

[0108] Based on this, in this embodiment, when training a pre-built large language model, the terminal device can input a sample data set into the large language model for pre-training and fine-tuning to obtain a trained first large language model.

[0109] From the above, it can be seen that the user tag generation method provided by this embodiment clearly focuses on the historical tag set under the first time dimension within the historical time period. It does not collect all tags aimlessly, but screens according to the set time dimension. The historical tag set obtained is closely related to the relevant activities and events under the time frame, which is more in line with the needs of the actual application scenario. Afterwards, key and representative words are extracted from many historical tags as the first prompt words, and the complicated tag information is highly condensed to highlight the core points, so that the subsequent model can better understand and process these key information, and also reduce the amount of data input into the subsequent large language model, avoiding excessive model processing burden or "noise" interference due to too much trivial information, so that the model can focus more on key information for effective analysis. Finally, relying on the powerful language understanding and generation capabilities of the large language model, and combining the two aspects of information, the refined prompt words and the interaction records, the past key features and real-time interaction situations are comprehensively considered, so that the generated first tag can take into account both historical experience and current interactive characteristics, and then generate tags that meet logic and actual needs, so that valuable first tags can be generated more efficiently, accurately and flexibly.

[0110] In S103, based on the first tag and the historical tags of the user in different time dimensions, a second tag of the user in the different time dimensions is generated; the different time dimensions correspond to different durations, and the first time dimension corresponds to the shortest duration.

[0111] In the embodiment of the present application, the terminal device can generate a second tag of the user in different time dimensions based on the first tag and the historical tags of the user in different time dimensions. The different time dimensions correspond to different durations, and the different time dimensions do not include the first time dimension, which corresponds to the shortest duration.

[0112] Specifically, for any one of the different time dimensions, the terminal device may input the first label and the historical label of the any one time dimension into a trained second label generation model for processing, thereby obtaining the second label of the any one time dimension.

[0113] In some possible embodiments, since the intelligent agent also has reasoning capabilities, the terminal device can analyze the first label and the user's historical labels at different time dimensions through the intelligent agent's reasoning capabilities, thereby generating a second label for the user at different time dimensions.

[0114] It should be noted that the second tags of different time dimensions may include multiple ones.

[0115] In an embodiment of the present application, the second label generation model is obtained by training the pre-constructed first deep learning model. The second label generation model can be obtained by training the pre-constructed first deep learning model based on a preset sample set. Each sample data in the preset sample set includes sample information (including a first label and a historical label of any time dimension) and a sample label corresponding to the sample information. When training the pre-constructed first deep learning model, the sample information in each sample data is used as the input of the first deep learning model, and the sample label corresponding to the sample information in each sample data is used as the output of the first deep learning model. Through training, the first deep learning model can learn the correspondence between all possible sample information and sample labels, and the trained first deep learning model is used as the second label generation model.

[0116] In some possible embodiments, when the duration of the first time dimension is one day, the durations of different time dimensions may include, but are not limited to, one week, one month, one quarter, and one year, and therefore, different time dimensions may include: a week dimension, a month dimension, a quarter dimension, and a year dimension. Accordingly, the second labels of different time dimensions may include: a week label, a month label, a quarter label, and a year label.

[0117] The weekly label is used to describe the user's weekly behavior patterns and can include the following:

[0118] 1. Commuting status: the user’s weekly commuting method, time and frequency.

[0119] 2. Work / study schedule: the user’s weekly work or study plan and tasks.

[0120] 3. Leisure activities: users’ weekly leisure activities (such as sports, watching movies, gatherings, etc.).

[0121] 4. Shopping frequency: the user’s weekly shopping habits (such as going to the supermarket, online shopping, etc.).

[0122] 5. Dietary patterns: the user’s weekly dietary structure and habits (e.g., preference for dining out on weekends).

[0123] 6. Family time: The time and activities that users spend with their families each week.

[0124] 7. Hobbies and Interests: Interest classes or hobby activities that users attend every week (such as painting classes, music classes).

[0125] 8. Health status: The user’s weekly health status and any medical records.

[0126] 9. Social activities: the user’s weekly social activities and frequency (such as friends’ gatherings, community activities).

[0127] 10. Time management: The user’s weekly schedule and management capabilities.

[0128] Monthly labels are used to describe the user's behavior and important events each month. They can include the following:

[0129] Income / Expenditure: The user's monthly income and main expenditures.

[0130] 1. Bill Payment: The user's monthly bill payment status (such as rent, utilities, credit cards).

[0131] 2. Major events: users’ major events or incidents every month (such as birthdays and anniversaries).

[0132] 3. Travel plan: the user’s monthly travel or business trip arrangements.

[0133] 4. Health check: Users’ monthly health check or medical appointment.

[0134] 5. Learning progress: User’s monthly learning progress and achievements (such as exams, project completion).

[0135] 6. Work performance: the user’s monthly work performance and goal completion status.

[0136] 7. Interpersonal relationships: monthly changes in users’ interpersonal relationships (such as making new friends and developing relationships).

[0137] 8. Savings / Investment: User’s monthly savings and investment status.

[0138] 9. Interest development: users’ progress and new attempts in their interests and hobbies every month.

[0139] Quarterly labels are used to describe user trends and changes each quarter and can include the following:

[0140] 1. Seasonal activities: users’ seasonal activities every quarter (such as skiing and seaside vacations).

[0141] 2. Learning / Training: The user’s quarterly learning or training plan.

[0142] 3. Career planning: users’ quarterly career goals and plans.

[0143] 4. Financial status: the user’s quarterly financial status and investment income.

[0144] 5. Health plan: The user’s quarterly health plan and goals (such as weight loss, fitness).

[0145] 6. Family relationships: User’s family relationships and dynamics every quarter.

[0146] 7. Social participation: users’ social activities and participation (such as volunteer activities) every quarter.

[0147] 8. Major purchases: users’ major purchases each quarter (such as home appliances and furniture).

[0148] 9. Psychological changes: users’ psychological state and changes every quarter.

[0149] 10. Environmental adaptation: Users’ adaptation to environmental changes (such as moving, changing jobs) every quarter.

[0150] Year labels are used to describe the user's important information and long-term habits each year. They can include the following:

[0151] 1. Birthday: The user’s birthday and how they celebrate it.

[0152] 2. Anniversary: ​​important anniversaries of users (such as wedding anniversary, love anniversary).

[0153] 3. Annual goals: The user’s annual goals and plans (such as promotion, learning new skills).

[0154] 4. Major events: major events of users every year (such as marriage, childbirth, moving).

[0155] 5. Travel plan: User’s annual travel plan and destination.

[0156] 6. Health status: User’s annual health status and annual physical examination results.

[0157] 7. Job changes: The user’s job changes each year (such as job hopping, promotion).

[0158] 8. Financial planning: User’s annual financial planning and savings goals.

[0159] 9. Interest development: the user’s interest development and new attempts each year.

[0160] 10. Social relationships: annual changes in users’ social relationships and network expansion.

[0161] In the embodiment of the present application, through the first label of the first time dimension and the second labels corresponding to different time dimensions, the user's behavior, habits and important information in different time dimensions can be fully understood, thereby ensuring the comprehensiveness of the final user label generated subsequently.

[0162] In one embodiment of the present application, in order to ensure the consistency and comprehensiveness between the various labels, the terminal device may generate the second labels of different time dimensions in sequence according to the label generation order of the different time dimensions. Based on this, the terminal device may specifically generate the second labels of different time dimensions by: Figure 2Steps S201 to S203 shown generate second labels of different time dimensions, as detailed below:

[0163] In S201, the label generation order of each of the different time dimensions is determined.

[0164] In this embodiment, since different time dimensions correspond to different durations, the terminal device can determine the order of generating labels according to the durations corresponding to the different time dimensions, wherein the order of generating labels in the first time dimension is the first priority.

[0165] It should be noted that the label generation order is positively correlated with the duration, that is, the shorter the duration, the earlier the label generation order, and the longer the duration, the later the label generation order.

[0166] Exemplarily, assuming that different time dimensions include: weekly dimension, monthly dimension, quarterly dimension and annual dimension, the duration corresponding to the weekly dimension is one week, the duration corresponding to the monthly dimension is one month, the duration corresponding to the quarterly dimension is one quarter, and the duration corresponding to the annual dimension is one year. Then, the label generation order of the weekly dimension is the second priority, the label generation order of the monthly dimension is the third priority, the label generation order of the quarterly dimension is the fourth priority, and the label generation order of the annual dimension is the fifth priority.

[0167] In S202, when the label generation order of the target time dimension is second, a second label of the target time dimension is generated based on the first label and the historical label of the target time dimension; the target time dimension is any one of the different time dimensions.

[0168] In this embodiment, when the label generation order of the target time dimension is the second priority, in order to ensure the consistency of the second label of the target time dimension with the first label of the first time dimension, the terminal device can generate the second label of the target time dimension based on the first label and the historical label of the target time dimension. Wherein, the target time dimension is any one of the different time dimensions.

[0169] Specifically, the terminal device may input the first label and the historical label of the target time dimension into a trained second label generation model for processing, thereby obtaining the second label of the target time dimension.

[0170] In practical applications, Large Language Models (LLM) are the first deep learning models trained based on massive text data. Large language models can not only generate natural language text, but also deeply understand the meaning of text and handle various natural language tasks, such as text summarization, question answering, translation, etc. Therefore, in order to further improve the generation accuracy of the second label of the target time dimension, the terminal device can specifically use the following methods: Figure 3Steps S301 to S303 shown generate a second label of the target time dimension, as detailed below:

[0171] In S301, the first label is input into the second largest language model for processing to obtain an initial label attribute corresponding to the target time dimension.

[0172] In this embodiment, since the duration of the target time dimension is greater than the duration of the first time dimension, the duration of the target time dimension can be composed of multiple durations of the first time dimension, that is, the terminal device can generate a second label of the target time dimension based on the first label of the first time dimension. Based on this, the terminal device can input the first label into the second largest language model for processing to obtain the initial label attribute corresponding to the target time dimension. Among them, the initial label attribute is used to describe the properties and characteristics that the second label of the target time dimension may have, that is, the attribute.

[0173] Specifically, the terminal device can perform keyword extraction on the historical tags of the target time dimension to obtain a second prompt word. After that, the terminal device can input the second prompt word and the first tag into the second largest language model for processing to obtain an initial tag attribute corresponding to the target time dimension.

[0174] It should be noted that the training process of the large language model includes two stages: pre-training and fine-tuning. In the pre-training stage, the large language model uses a large amount of text corpus (the first labels in the sample data set) for unsupervised learning to learn the language rules and semantic relationships in the text sequence. In the fine-tuning stage, the large language model uses a small amount of annotated data (such as the second label containing the second prompt word and the target time dimension) for supervised learning to optimize the model parameters.

[0175] Based on this, in this embodiment, when the terminal device trains the pre-built large language model, the sample data set can be input into the large language model for pre-training and fine-tuning to obtain a trained second large language model.

[0176] In S302, a correlation evaluation is performed on the first tag and the historical tag of the target time dimension to obtain an evaluation result.

[0177] In this embodiment, the terminal device can input the first tag and the historical tag of the target actual dimension into the evaluation model for correlation evaluation to obtain an evaluation result, wherein the evaluation result is used to describe whether the first tag and the historical tag of the target time dimension are correlated.

[0178] Evaluation results include, but are not limited to: relevant and irrelevant.

[0179] It should be noted that the evaluation model is obtained by training a pre-constructed first deep learning model. Among them, the evaluation model can be obtained by training a pre-constructed second deep learning model based on a preset sample set. Among them, each sample data in the preset sample set includes sample information (including a first label and a historical label of the target time dimension) and a sample evaluation result corresponding to the sample information. When training the pre-constructed second deep learning model, the sample information in each sample data is used as the input of the second deep learning model, and the sample evaluation result corresponding to the sample information in each sample data is used as the output of the second deep learning model. Through training, the second deep learning model can learn the correspondence between all possible sample information and sample evaluation results, and the trained second deep learning model is used as the evaluation model.

[0180] In S303, the first label and the historical label are processed based on the evaluation result and the initial label attribute to obtain a second label of the target time dimension.

[0181] In this embodiment, when the evaluation result is irrelevant, it means that the first label is irrelevant to the historical label of the target time dimension. Since the duration of the target time dimension can be composed of multiple durations of the first time dimensions, that is to say, the terminal device can generate a second label of the target time dimension based on the first label of the first time dimension. Therefore, the terminal device can directly filter out at least one sub-label that matches the initial label attribute from the first label, and determine the at least one sub-label as the second label of the target time dimension.

[0182] In one embodiment of the present application, in order to improve the comprehensiveness of the second label of the target time dimension and avoid missing sub-labels that are consistent with the attributes of the initial label, when the terminal device detects that the evaluation result is irrelevant, it can still filter at least one sub-label that is consistent with the attributes of the initial label from the first label and the historical labels of the target time dimension, and determine the at least one sub-label as the second label of the target time dimension.

[0183] When the evaluation result is related, it means that the first tag is related to the historical tag of the target time dimension. Therefore, the terminal device can filter out at least one sub-tag that matches the initial tag attribute from the first tag and the historical tag of the target time dimension, and determine the at least one sub-tag as the second tag of the target time dimension.

[0184] From the above, it can be seen that the user tag generation method provided by this embodiment uses the second largest language model to further process the first tag, closely focusing on the target time dimension, ensuring that the generated initial tag attributes are highly relevant to the specific time dimension, avoiding the generalization of general attributes, making the attributes more targeted and practical, and meeting the detailed characterization requirements of tags in a specific time frame. Afterwards, through correlation evaluation, the similarities, differences, and potential causal relationships between the first tag and the historical tags can be accurately found. Finally, the correlation evaluation results and the initial tag attributes are organically combined, and the first tag and the historical tags are systematically processed, making full use of the information obtained in the first two steps. Instead of viewing each data element in isolation, they are optimized and adjusted under an overall framework to form a coherent and coordinated tag generation process, so that the second tag of the target time dimension finally obtained combines the advantages of historical experience, current characteristics, and professional model analysis, and has higher quality and practicality.

[0185] In another embodiment of the present application, in order to improve the accuracy of generating the second label of the target time dimension, when the terminal device detects that the evaluation result is relevant, it can be performed as follows: Figure 4 Steps S401 to S402 shown in the figure obtain the second label of the target time dimension, which is described in detail as follows:

[0186] In S401, if the evaluation result is relevant, the first label and the historical label of the target time dimension are input into the second largest language model for processing to obtain a target label attribute corresponding to the target time dimension.

[0187] In S402, the first tag and the historical tags of the target time dimension are screened based on the target tag attribute to obtain a second tag of the target time dimension.

[0188] In this embodiment, when the terminal device detects that the evaluation result is relevant, it indicates that the first label is related to the historical label of the target time dimension. Therefore, the terminal device can input the first label and the historical label of the target time dimension into the second largest language model for processing to obtain a more comprehensive and accurate target label attribute corresponding to the target time dimension.

[0189] Afterwards, the terminal device may filter at least one sub-tag that matches the target tag attribute from the first tag and the historical tags of the target time dimension, and determine the at least one sub-tag as the second tag of the target time dimension.

[0190] From the above, it can be seen that the user tag generation method provided by this embodiment only performs subsequent in-depth processing when there is a correlation between the first tag and the historical tag, avoiding unnecessary waste of computing resources and interference from irrelevant information, and improving the efficiency and accuracy of the entire process. At the same time, the first tag and the historical tag of the target time dimension are input into the model together to achieve a deep fusion of historical data and current tag information. Afterwards, the target tag attributes are used as the core basis for screening, and this method has a clear goal orientation. It can select the tags that best meet these attribute characteristics from a large number of first tags and historical tags based on the extracted target tag attributes, ensuring that the final second tag set has a high degree of consistency and logic, and is closely aligned with the business needs and data characteristics under the target time dimension, thereby improving the generation quality and accuracy of the second tag.

[0191] In one embodiment of the present application, since there may be conflicting or contradictory sub-labels between the first label and the historical label of the target time dimension, in order to improve the accuracy of generating the second label of the target time dimension, the terminal device may specifically implement step S402 according to the following steps, which are described in detail as follows:

[0192] The first tag and the historical tags of the target time dimension are filtered based on the target tag attribute to obtain at least one first subtag and at least one historical subtag that match the target tag attribute; each of the first subtags is any one of the first tags, and each of the historical subtags is any one of the historical tags of the target time dimension;

[0193] If the first target subtag and the second target subtag are conflicting tags, the second target subtag is deleted; the first target subtag is any one of at least one of the first subtags, and the second target subtag is any one of at least one of the historical subtags;

[0194] A second label of the target time dimension is generated according to at least one of the first sub-labels and other sub-labels; the other sub-labels refer to historical sub-labels other than the second target sub-label.

[0195] In this embodiment, the terminal device can filter at least one first sub-tag and at least one historical sub-tag that match the target tag attribute from the first tag and the historical tags of the target time dimension. Each first sub-tag is any tag in the first tag, and each historical sub-tag is any tag in the historical tags of the target time dimension.

[0196] Afterwards, the terminal device may perform conflict detection on the at least one first subtag and the at least one historical subtag one by one to determine whether there is a conflicting tag between the at least one first subtag and the at least one historical subtag.

[0197] Specifically, the terminal device may input the at least one first subtag and the at least one historical subtag into a conflict detection model for processing to obtain a conflict result between the at least one first subtag and the at least one historical subtag, wherein the conflict result is used to describe whether a certain first subtag and a certain historical subtag conflict.

[0198] It should be noted that the conflict detection model is trained by a pre-built third deep learning model.

[0199] In this embodiment, the conflict detection model can be obtained by training a pre-constructed third deep learning model based on a preset sample set. Each sample data in the preset sample set includes sample information (including multiple first sub-labels and multiple historical sub-labels) and a sample conflict result corresponding to the sample information. When training the pre-constructed third deep learning model, the sample information in each sample data is used as the input of the third deep learning model, and the sample conflict result corresponding to the sample information in each sample data is used as the output of the third deep learning model. Through training, the third deep learning model can learn the correspondence between all possible sample information and sample conflict results, and the trained third deep learning model is used as the conflict detection model.

[0200] In this embodiment, when the terminal device detects that the conflict result is that the first target subtag conflicts with the second target subtag, it can determine that the first target subtag and the second target subtag are conflicting tags. The first target subtag is any one of the at least one first subtag, and the second target subtag is any one of the at least one historical subtag.

[0201] In one embodiment of the present application, when the terminal device detects that there are no conflicting tags in at least one first subtag and at least one historical subtag, the terminal device can directly determine the at least one first subtag and at least one historical subtag as the second tag of the target time dimension.

[0202] In another embodiment of the present application, when the terminal device detects that the first target sub-tag and the second target sub-tag are conflicting tags, since the first tag of the first time dimension is generated after receiving the tag generation request, the generation time of the historical tag of the target time dimension already existed before the time corresponding to the tag generation request, indicating that the generation time of the first tag is later than the historical tag of the target time dimension, that is, the first tag is more effective and accurate than the historical tag of the target time dimension, and therefore, the terminal device can delete the second target sub-tag.

[0203] Afterwards, the terminal device may determine at least one first sub-tag and the remaining sub-tags as the second tag of the target time dimension, wherein the remaining sub-tags refer to the historical sub-tags except the second target sub-tag.

[0204] As can be seen from the above, the user tag method provided in this embodiment selects sub-tags that match the target tag attribute from the first tag and the historical tag based on the clear standard of the target tag attribute. This screening method has high accuracy and pertinence, and can quickly focus on the tag information that is most relevant to the current target, avoiding the waste of resources and analysis complexity on irrelevant or low-relevance tags. At the same time, by screening the first tag and the historical tag together, the association and integration between data of different periods are promoted, and the tag elements with consistency and coherence under the target tag attribute framework are excavated, so that the selected sub-tags contain both the current new information and the effective experience and patterns in the historical data, providing a rich and valuable data foundation for subsequent tag optimization. Afterwards, conflict processing is performed in the selected sub-tag set, and the tag set can be dynamically adjusted according to the actual situation to ensure that the sub-tags used to generate the second tag are coordinated and non-conflicting, thereby ensuring the stability and reliability of the entire tag generation process, and adapting to various conflicts that may occur under different data combinations and business scenarios. Finally, the first sub-label and the remaining historical sub-labels after conflict resolution are comprehensively considered, and the effective label information after screening and conflict resolution is fully utilized. The two are combined to generate the second label, which embodies a comprehensive and systematic label generation method. It not only absorbs the innovation and timeliness of the new label, but also retains the stability and empirical elements in the historical label, so that the final generated second label has higher quality and comprehensiveness.

[0205] In S203, when the label generation order of the target time dimension is not the second priority, the second label of the target time dimension is generated based on the second label of the previous priority time dimension and the historical label of the target time dimension; the previous priority time dimension refers to the last time dimension before the label generation order of the target time dimension.

[0206] In this embodiment, when the terminal device detects that the tag generation order of the target time dimension is not the second priority, in order to ensure the connection and consistency of the user's tags in different time periods, the terminal device can generate the second tag of the target time dimension based on the second tag of the previous priority time dimension and the historical tag of the target time dimension. The previous priority time dimension refers to the last time dimension before the tag generation order of the target time dimension.

[0207] Exemplarily, assuming that the label generation order of the target time dimension is third, the terminal device can generate a second label of the target time dimension based on the second label of the time dimension with the second label generation order and the historical label of the target time dimension.

[0208] In this embodiment, the terminal device can input the second label of the previous time dimension and the historical label of the target time dimension into the trained second label generation model for processing, so as to obtain the second label of the target time dimension.

[0209] In one embodiment of the present application, step S203 in this embodiment is Figure 2 The step S202 in the corresponding embodiment is the same. Please refer to the relevant description in step S202 for details, which will not be repeated here.

[0210] As can be seen from the above, the user label method provided by this embodiment sorts different time dimensions, which reflects a systematic and structured way of thinking. By predetermining the order of each time dimension in the label generation process, a clear framework and logical order are established for the entire label generation process, which is helpful to organize and coordinate complex multi-time dimension data processing work. Afterwards, when the label generation order of the target time dimension is the second priority, both the current first label (representing new information or data features) and the historical label of the target time dimension (containing past experience and rules) are considered. This data fusion method can make full use of the value of historical data and incorporate new information in a timely manner, making the generated second label more comprehensive and timely. Then, when the label generation order of the target time dimension is not the second priority, the second label of the previous priority is used as the basis, which means that the information that has been processed and refined in the previous time dimension is inherited, and then on this basis, the historical label of the current time dimension is combined for further refinement and supplementation, so that the generated second label can continuously enrich and improve its own connotation and extension in the context of a more macroscopic time series, and better adapt to the characteristics and business needs of data at different time scales. In summary, this embodiment can construct a complete label system that includes both macro trend information (gradually accumulated from the early time dimension) and micro detail changes (supplemented by historical labels of each time dimension) through recursive label generation, and also ensures the consistency and comprehensiveness between the labels.

[0211] In S104, a target tag of the user is generated based on the first tag and each of the second tags.

[0212] In the embodiment of the present application, the terminal device can splice the first tag and each second tag to obtain the user's target tag.

[0213] Exemplarily, assuming that the first label is X1, and the second labels are X2, X3, X4 and X5, the target label may be [X1, X2, X3, X4, X5].

[0214] As can be seen from the above, a user tag generation method provided by an embodiment of the present application obtains the interaction record between the user and the agent; analyzes the interaction record to generate the first tag of the user in the first time dimension; generates the second tag of the user in different time dimensions based on the first tag and the historical tags of the user in different time dimensions; the durations corresponding to different time dimensions are different, and the different time dimensions do not include the first time dimension, and the duration corresponding to the first time dimension is the shortest; based on the first tag and each second tag, the user's target tag is generated. Compared with the single data obtained by the prior art, since the agent can collect a large amount of data during the interaction with the user, such as the user's preferences, behavioral habits, needs, etc., the present application can obtain diversified data through the interaction record between the user and the agent, so that the single tag generated by the interaction record is more comprehensive. At the same time, the process of generating the first tag is based on the objective data of the actual interaction record between the user and the agent. It is obtained by analyzing, reducing subjective assumptions, thereby improving the objectivity and accuracy of the first tag. Afterwards, since the second tags of different time dimensions are generated by combining the first tag and the historical tags in different time dimensions, the consistency and comprehensiveness of the tags of different time dimensions are guaranteed. Finally, through the first label of the first time dimension and the second labels corresponding to different time dimensions, the user's behavioral performance in time dimensions of different lengths is comprehensively considered, from the shortest first time dimension to other time dimensions of different lengths, the dynamic changes of user behavior characteristics are fully captured, avoiding the one-sidedness of single time point analysis, so as to fully understand the user's behavior, habits and important information in different time dimensions, thereby ensuring the comprehensiveness of the final user labels generated subsequently, and thus improving the accuracy of user label generation.

[0215] It should be understood that the size of the serial numbers of the steps in the above embodiments does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0216] Corresponding to the user tag generation method described in the above embodiment, Figure 5 The structure diagram of a user tag generation device provided by an embodiment of the present application is shown. For the convenience of explanation, only the part related to the embodiment of the present application is shown. Figure 5 The user tag generation device 500 includes: a first acquisition unit 51, a first generation unit 52, a second generation unit 53 and a third generation unit 54. Wherein:

[0217] The first acquisition unit 51 is used to acquire the interaction record between the user and the agent.

[0218] The first generating unit 52 is used to analyze the interaction record and generate a first label of the user in a first time dimension.

[0219] The second generating unit 53 is used to generate a second tag of the user in the different time dimensions based on the first tag and the historical tags of the user in the different time dimensions; the different time dimensions each correspond to a different duration, the different time dimensions do not include the first time dimension, and the first time dimension corresponds to the shortest duration.

[0220] The third generating unit 54 is configured to generate a target tag for the user based on the first tag and each of the second tags.

[0221] In one embodiment of the present application, the first generating unit 52 specifically includes: an extracting unit, a fourth generating unit and a first processing unit. Among them:

[0222] The extraction unit is used to obtain a set of historical tags in the first time dimension within a historical time period;.

[0223] The fourth generating unit is used to extract keywords from the historical tag set to obtain a first prompt word.

[0224] The first processing unit is used to input the first prompt word and the interaction record into a first large language model for processing to obtain the first label.

[0225] In one embodiment of the present application, the tag generation order of the first time dimension is the first priority; the second generation unit 53 specifically includes: a first determination unit and a fifth generation unit. Among them:

[0226] The first determining unit is used to determine the label generation order of the different time dimensions.

[0227] The fifth generating unit is used to generate a second label of the target time dimension based on the first label and the historical label of the target time dimension when the label generation order of the target time dimension is second; the target time dimension is any one of the different time dimensions.

[0228] In one embodiment of the present application, the user tag generating device 500 may further include: a sixth generating unit.

[0229] The sixth generation unit is used to generate the second label of the target time dimension based on the second label of the previous-order time dimension and the historical label of the target time dimension when the label generation order of the target time dimension is not the second-order; the previous-order time dimension refers to the last time dimension before the label generation order of the target time dimension.

[0230] In one embodiment of the present application, the fifth generation unit specifically includes: a second processing unit, an evaluation unit and a third processing unit.

[0231] The second processing unit is used to input the first label into the second largest language model for processing to obtain an initial label attribute corresponding to the target time dimension.

[0232] The evaluation unit is used to perform a correlation evaluation on the first label and the historical label of the target time dimension to obtain an evaluation result.

[0233] The third processing unit is used to process the first label and the historical label based on the evaluation result and the initial label attribute to obtain a second label of the target time dimension.

[0234] In one embodiment of the present application, the third processing unit specifically includes: a fourth processing unit and a first screening unit.

[0235] The fourth processing unit is used to input the first label and the historical label of the target time dimension into the second large language model for processing if the evaluation result is relevant, so as to obtain the target label attribute corresponding to the target time dimension.

[0236] The first screening unit is used to screen the first label and the historical label of the target time dimension based on the target label attribute to obtain a second label of the target time dimension.

[0237] In one embodiment of the present application, the first screening unit specifically includes: a second screening unit, a deletion unit and a seventh generation unit. Among them:

[0238] The second filtering unit is used to filter the first tag and the historical tags of the target time dimension based on the target tag attributes to obtain at least one first sub-tag and at least one historical sub-tag that are consistent with the target tag attributes; each of the first sub-tags is any one of the first tags, and each of the historical sub-tags is any one of the historical tags of the target time dimension.

[0239] The deleting unit is used to delete the second target subtag if the first target subtag and the second target subtag are conflicting tags; the first target subtag is any one of at least one of the first subtags, and the second target subtag is any one of at least one of the historical subtags.

[0240] The seventh generating unit is used to generate a second label of the target time dimension according to at least one of the first sub-labels and other sub-labels; the other sub-labels refer to historical sub-labels other than the second target sub-label.

[0241] In one embodiment of the present application, the third processing unit specifically includes: a third screening unit. The third screening unit is used to screen the first label and the historical label of the target time dimension based on the initial label attribute to obtain the second label of the target time dimension if the evaluation result is irrelevant.

[0242] It should be noted that the information interaction, execution process, etc. between the above-mentioned devices / units are based on the same concept as the method embodiment of the present application. Their specific functions and technical effects can be found in the method embodiment part and will not be repeated here.

[0243] The technicians in the relevant field can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In practical applications, the above-mentioned function allocation can be completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated in a processing unit, or each unit can exist physically separately, or two or more units can be integrated in one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, which will not be repeated here.

[0244] Figure 6 This is a schematic diagram of the structure of a terminal device provided in an embodiment of the present application. Figure 6 As shown, the terminal device 6 of this embodiment includes: at least one processor 60 ( Figure 6 Only one is shown in the figure) a processor, a memory 61, and a computer program 62 stored in the memory 61 and executable on the at least one processor 60, and when the processor 60 executes the computer program 62, the steps in any of the above-mentioned user tag generation method embodiments are implemented.

[0245] The terminal device may include, but is not limited to, a processor 60 and a memory 61. Those skilled in the art will appreciate that Figure 6 It is only an example of the terminal device 6 and does not constitute a limitation on the terminal device 6. It may include more or fewer components than shown in the figure, or a combination of certain components, or different components. For example, it may also include input and output devices, network access devices, etc.

[0246] The processor 60 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc.

[0247] In some embodiments, the memory 61 may be an internal storage unit of the terminal device 6, such as the memory of the terminal device 6. In other embodiments, the memory 61 may also be an external storage device of the terminal device 6, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the terminal device 6. Further, the memory 61 may also include both the internal storage unit of the terminal device 6 and an external storage device. The memory 61 is used to store an operating system, an application program, a boot loader (BootLoader), data, and other programs, such as the program code of the computer program. The memory 61 may also be used to temporarily store data that has been output or is to be output.

[0248] An embodiment of the present application further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps in the above-mentioned method embodiments can be implemented.

[0249] An embodiment of the present application provides a computer program product. When the computer program product is run on a terminal device, the terminal device can implement the steps in the above-mentioned method embodiments when executing the computer program product.

[0250] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present application implements all or part of the processes in the above-mentioned embodiment method, which can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may at least include: any entity or device capable of carrying the computer program code to the terminal device, a recording medium, a computer memory, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), an electric carrier signal, a telecommunication signal and a software distribution medium. For example, a USB flash drive, a mobile hard disk, a magnetic disk or an optical disk. In some jurisdictions, according to legislation and patent practice, computer-readable media cannot be electric carrier signals and telecommunication signals.

[0251] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0252] The embodiments described above are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, a person skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application, and should all be included in the protection scope of the present application.

Claims

1. A method for generating a user tag, characterized in that: include: Obtain the interaction records between the user and the agent; Analyze the interaction record to generate a first label of the user in a first time dimension; Based on the first tag and the historical tags of the user in different time dimensions, generating a second tag of the user in the different time dimensions; the different time dimensions respectively correspond to different durations, and the first time dimension corresponds to the shortest duration; A target tag of the user is generated based on the first tag and each of the second tags.

2. The user tag generation method according to claim 1, characterized in that: The analyzing the interaction record to generate a first label of the user in a first time dimension includes: Obtain a historical tag set in the first time dimension within a historical time period; Extracting keywords from the historical tag set to obtain a first prompt word; The first prompt word and the interaction record are input into a first large language model for processing to obtain the first label.

3. The user tag generation method according to claim 1, characterized in that: The tag generation order of the first time dimension is the first priority; generating the second tag of the user in the different time dimensions based on the first tag and the historical tags of the user in the different time dimensions includes: Determine the order of generating labels for the different time dimensions; When the label generation order of the target time dimension is second, a second label of the target time dimension is generated based on the first label and the historical label of the target time dimension; the target time dimension is any one of the different time dimensions.

4. The user tag generation method according to claim 3, characterized in that: After determining the label generation order of the different time dimensions, the method further includes: When the label generation order of the target time dimension is not the second priority, the second label of the target time dimension is generated based on the second label of the previous priority time dimension and the historical label of the target time dimension; the previous priority time dimension refers to the last time dimension before the label generation order of the target time dimension.

5. The user tag generation method according to claim 3, characterized in that: The generating a second label of the target time dimension based on the first label and the historical label of the target time dimension includes: Inputting the first label into the second largest language model for processing to obtain an initial label attribute corresponding to the target time dimension; Performing a correlation evaluation on the first label and the historical label of the target time dimension to obtain an evaluation result; The first label and the historical label are processed based on the evaluation result and the initial label attribute to obtain a second label of the target time dimension.

6. The user tag generation method according to claim 5, characterized in that: The processing of the first label and the historical label based on the evaluation result and the initial label attribute to obtain the second label of the target time dimension includes: If the evaluation result is relevant, inputting the first label and the historical label of the target time dimension into the second largest language model for processing to obtain a target label attribute corresponding to the target time dimension; The first label and the historical labels of the target time dimension are filtered based on the target label attribute to obtain a second label of the target time dimension.

7. The user tag generation method according to claim 6, characterized in that: The filtering and processing of the first label and the historical label of the target time dimension based on the target label attribute to obtain the second label of the target time dimension includes: The first tag and the historical tags of the target time dimension are filtered based on the target tag attribute to obtain at least one first subtag and at least one historical subtag that match the target tag attribute; each of the first subtags is any one of the first tags, and each of the historical subtags is any one of the historical tags of the target time dimension; If the first target subtag and the second target subtag are conflicting tags, the second target subtag is deleted; the first target subtag is any one of at least one of the first subtags, and the second target subtag is any one of at least one of the historical subtags; A second label of the target time dimension is generated according to at least one of the first sub-labels and other sub-labels; the other sub-labels refer to historical sub-labels other than the second target sub-label.

8. The user tag generation method according to claim 5, characterized in that: The processing of the first label and the historical label based on the evaluation result and the initial label attribute to obtain the second label of the target time dimension includes: If the evaluation result is irrelevant, the first label and the historical label of the target time dimension are screened based on the initial label attribute to obtain a second label of the target time dimension.

9. A terminal device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the user tag generating method according to any one of claims 1 to 8 is implemented.

10. A computer program product, characterized in that It comprises a computer program, which implements the user tag generating method according to any one of claims 1 to 8 when being executed.

Images