Task decision-making method and device based on large model and computer equipment

Abstract

The invention relates to a task decision-making method and device based on a large model and computer equipment. The method comprises the steps that basic information of a user is acquired, a user portrait is determined based on the basic information, and the user portrait is used for hierarchical division of the user and role description of the user; dialogue information of the user is obtained, and the large model carries out intention recognition and behavior prediction of the user based on the dialogue information and the user portrait; and performing task decision based on the intention recognition result and the behavior prediction result. Based on the method, the decision result can be quickly obtained, and the basic information of the user and the dialogue information of the user are integrated in the decision process, so that the decision result is more reasonable and accurate.

Description

Technical Field

[0001] The present application relates to the field of large model technology, and in particular to a large model-based task decision method, device and computer equipment. Background Art

[0002] In actual user interaction scenarios, a lot of user information is usually included. Collect this information and analyze users based on it, such as user portraits and user stratification; and then make reasonable decisions, such as recommendations, services, feedback, etc. Therefore, the entire process can be roughly divided into two modules: "information collection" and "response decision-making".

[0003] In one solution, you can collect user portrait information through surveys and tracking methods, such as labeled data such as age, gender, city, and education level, and then use some statistical modeling methods to perform some stratification and portrait analysis on users. Further combined with business characteristics, determine some differentiated and personalized responses for users, such as recommendations, services, feedback, etc.

[0004] However, this solution relies more on some labeled structured data in the information collection stage and more on manual statistics in the response decision-making stage. This determines that the solution has major shortcomings in terms of dimension and labor cost. Summary of the invention

[0005] Based on this, it is necessary to provide a method that can improve the efficiency and accuracy of task decision-making for the above technical problems. Based on the information collection and response feedback solution based on the large model (LLM) proposed in this application, relying on the powerful knowledge reserve and ability of the large model, it can well mine text information without a lot of manual intervention, and give personalized and differentiated response feedback based on the different characteristics of users, where the response feedback includes diverse content generation. This solution can provide a more comprehensive dimension and has a relatively large application prospect in some common application scenarios, such as customer voice, user quality inspection, recommendation and other scenarios.

[0006] A task decision method based on a large model, the method comprising:

[0007] Obtaining basic information of the user, and determining a user portrait based on the basic information, wherein the user portrait is used to divide the user into different levels and characterize the user's role;

[0008] The user's conversation information is obtained, and the big model recognizes the user's intention and predicts the user's behavior based on the conversation information and the user portrait;

[0009] Based on the intention recognition result and the behavior prediction result, a task decision is made.

[0010] In some embodiments, the task decision scheme includes one or more, and the task decision scheme includes:

[0011] Determining the relevance of each of the user's existing intentions;

[0012] Conduct comprehensive analysis and task decision-making based on relevant intentions to obtain decision results;

[0013] Make task decisions based on independent and unrelated intentions and obtain decision results.

[0014] In some embodiments, the task decision scheme includes one or more, and the task decision scheme includes:

[0015] The importance of each of the existing intentions to the user is determined based on the basic information of the user.

[0016] The large model makes a task decision based on the importance of the existing intention and obtains a decision result.

[0017] In some embodiments, the task decision scheme includes one or more, and the task decision scheme includes:

[0018] The large model predicts the user's current intention based on the user's basic information, and makes a decision based on the predicted intention and / or the current intention to obtain a decision result.

[0019] In some embodiments, the method further comprises:

[0020] Predicting a recommendation scheme that the user may be interested in based on the user's basic information, and determining a recommendation reason corresponding to the recommendation scheme;

[0021] A decision solution is obtained based on the recommendation solution and the recommendation reason, and is recommended to the user terminal based on the importance of the decision solution.

[0022] In some embodiments, the method further comprises:

[0023] The user's willingness strength is determined, and users and decision plans whose willingness strength meets a preset value are given priority.

[0024] In some embodiments, the method further comprises:

[0025] Obtain user's historical behavior information;

[0026] Making task decisions based on the historical behavior information, the basic information, and key information in the dialogue information;

[0027] The key information in the dialogue information is used as an important consideration point for task decision making, and the obtained task decision plan is developed around the corresponding key point information.

[0028] In some embodiments, the method further comprises:

[0029] Determine the level and role of the user based on the user's basic information;

[0030] Determine the target products that match the user's level and role;

[0031] A decision plan is generated and recommended to the user based on the target product and the characteristics of the target product.

[0032] A task decision device based on a large model, the device comprising:

[0033] An acquisition module is used to acquire basic information of a user and determine a user portrait based on the basic information. The user portrait is used to divide the user into different levels and characterize the user's role.

[0034] A data processing module is used to obtain the user's conversation information. The large model recognizes the user's intention and predicts the user's behavior based on the conversation information and the user portrait.

[0035] A decision module is used to make task decisions based on the intention recognition result and the behavior prediction result.

[0036] A computer device comprises a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the following steps are implemented:

[0037] Obtaining basic information of the user, and determining a user portrait based on the basic information, wherein the user portrait is used to divide the user into different levels and characterize the user's role;

[0038] The user's conversation information is obtained, and the big model recognizes the user's intention and predicts the user's behavior based on the conversation information and the user portrait;

[0039] Based on the intention recognition result and the behavior prediction result, a task decision is made.

[0040] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the following steps:

[0041] Obtaining basic information of the user, and determining a user portrait based on the basic information, wherein the user portrait is used to divide the user into different levels and characterize the user's role;

[0042] The user's conversation information is obtained, and the big model recognizes the user's intention and predicts the user's behavior based on the conversation information and the user portrait;

[0043] Based on the intention recognition result and the behavior prediction result, a task decision is made.

[0044] The above-mentioned task decision method, device, computer equipment and storage medium based on the big model conduct a comprehensive analysis of the user's conversation information and basic information based on the big model to obtain the user's intention and possible behavior, and then automatically make task decisions. This can not only make the decision results more reasonable, but also improve the efficiency of decision-making due to automatic decision-making through the big model. In addition, the user's basic information and conversation information are comprehensively considered in the decision-making process, which makes the decision more accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 A flowchart of a task decision method based on a large model provided in one embodiment;

[0046] Figure 2 A schematic diagram of a process for making multiple task decisions based on a large model provided in one embodiment;

[0047] Figure 3 A structural block diagram of a task decision device based on a large model provided in one embodiment;

[0048] Figure 4 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0049] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0050] The task decision method based on large model provided in this application is

[0051] A task decision method based on a large model, the method comprising the following steps:

[0052] Step 101, obtain basic information of the user, and determine a user portrait based on the basic information, wherein the user portrait is used to divide the user into different levels and characterize the user's role.

[0053] In some embodiments, the basic information data of users is collected, mainly including user portrait data. User portrait data may include age, gender, address, marital status, province, education, social security, etc., which usually represents the medium- and long-term relatively stable basic information of users, usually structured data.

[0054] The basic information of users is used to describe the characteristics of user portraits. This part of information can be used to assist in describing the characteristics of users. For example, hierarchical division of users by "age" and "gender" is a kind of "role" boundary definition. Because different roles will have differences in response decisions.

[0055] The basic information of users (age, gender, address, marital status, etc.) is mainly used to describe the basic characteristics of users. These characteristics may lead to some group and hierarchical distinctions among users, and can usually be described as the common characteristics of a category of users.

[0056] For example, when recommending a car, for a "35-year-old man with a family and children" and a "22-year-old single woman", although the models, brands, configurations, etc. that these two people may be interested in are very similar, considering the differences in basic user information, including "age", "marriage", "family members", "gender", etc., there will be certain differences when responding to decisions.

[0057] In the above embodiment, by using the basic information of the user, the large model (LLM) can be assisted in making some more accurate and reasonable response decisions in the "role portrayal" of the large model (LLM). For example, through the prompt project of the large model (LLM), the basic information of the user is integrated into the understanding and content generation of the large model (LLM) by designing the prompt.

[0058] Step 102: Acquire the user's conversation information, and the big model recognizes the user's intention and predicts the user's behavior based on the conversation information and the user portrait.

[0059] User conversation text data mainly includes user chat records and telephone recording text (e.g., data that can be obtained through ASR translation) obtained through legal channels, such as corporate WeChat, app customer service, service phone, etc., which mainly records direct interactions with users. Specifically, it includes text-related information, including text chat records and voice calls (ASR translated to text).

[0060] In some embodiments, text preprocessing may also be performed on the chat session text, including data cleaning, processing, sentence segmentation, word segmentation, etc.

[0061] In some embodiments, the method provided in the present application also includes data preprocessing of the user's basic information and user behavior data (for example, the user's conversation information), including data cleaning, standardization and structuring. By preprocessing the data, noise and outliers can be eliminated to ensure the quality and accuracy of the data.

[0062] For user text chat data, the existing large model (LLM) technology is mainly used in combination with the prompt method. On the one hand, some important user information is mined from the text data. On the other hand, it is necessary to make reasonable response decisions based on the mined text information and the user's basic (portrait) information, using the understanding and generation capabilities of the large model (LLM).

[0063] In some embodiments, in order to expand the data type, some artificial intelligence methods are introduced, such as NLP technology, to extend information collection and response decisions to unstructured links, and to achieve an automated mode to a certain extent. For example, through the information extraction and classification technology of NLP, information collection can be extended to text data, including some chat dialogues, comments, opinion information, etc., which can all be used as data sources for information mining. At the same time, for example, the text generation technology in NLP technology can realize automatic content generation to a certain extent as a response decision. However, the NLP technology used in this solution is more dependent on training data, usually requiring relatively large labor costs for annotation, and because of the limitations of the model architecture, there is usually a relatively obvious upper limit on the effect. The latest large model (LLM) technology is a general model trained with a large amount of data. In addition to containing a large amount of knowledge and information, this large model (LLM) has great advantages in deep semantics and logical reasoning.

[0064] It should be noted that a large model can be a model with a large number of parameters trained based on a transformer architecture through massive data. The advantage of a large model (LLM) is that, on the one hand, the large model can be used as a fixed knowledge base to achieve certain information query capabilities. On the other hand, the large parameter attribute of the large model (LLM) generates the "emergence" ability of the large model, so the large model (LLM) has strong understanding and reasoning capabilities.

[0065] The large model is based on the transformer architecture, which is a relatively powerful neural network model. It mainly imitates the attention of the human brain. The model can be used to select different information to capture more core information. Among them, the multi-head attention mechanism of the transformer architecture. Through the attention mechanism and similarity calculation, the word weights in the conversation can be output. According to the word weights, the importance of keywords can be obtained, and then effective keywords can be obtained.

[0066] Among them, the expression of the multi-head attention mechanism is:

[0067]

[0068] In the above expression, Q, K, and V all represent input dialogue vectors, where the dialogue vector is obtained from the original dialogue content through text encoding technology. The self-attention in the dialogue vector is further calculated, and the original dialogue vector is transformed into three branches of Q, K, and V through linear operations to better calculate the correlation in the dialogue vector, so as to better highlight the weight of the key words, and then obtain the weight score of the word as the main basis for subsequent keyword determination.

[0069] Step 103: Make a task decision based on the intention recognition result and the behavior prediction result.

[0070] In some embodiments, the large model performs user intention recognition and behavior prediction based on the conversation information and the user portrait, including:

[0071] Based on the big model, the conversation information is mined to obtain the user's intention and the behavior prediction information associated with the intention; wherein the intention label and behavior label are automatically extracted and determined by the big model, for example, they can be automatically extracted by the big model based on the user's conversation information. In some embodiments, these intention labels can be defined by the user in the early stage, and can be defined and mined by the big model (LLM) in the later stage, which is equivalent to adding a user feature dimension.

[0072] On the one hand, the large model (LLM) can mine user information in the text. For example, during a chat, a user said, "I recently heard that you have released a new car that is comparable to the xx model. Can you briefly introduce the main selling points and price of this car to me?"

[0073] Using the large model (LLM) technique, we can extract several intent-related information from this sentence:

[0074] 1. Intent: New car consultation

[0075] 2. Do you want to inquire: Yes

[0076] 3. Interested: Yes

[0077] 4. Is it possible to visit the store: Very likely

[0078] For this embodiment, the large model (LLM) can not only mine the user's intention information, but also mine the user's behavior information, and based on the intention and behavior information, give a predictive judgment, that is, the user is likely to test drive in the store, so an "invitation to test drive in the store" can be further issued to complete automatic decision-making.

[0079] For example, the user said the following sentence:

[0080] "My child is going to school soon, and my wife may need to pick him up from school frequently, so I plan to buy another new energy car, mainly for my wife to use."

[0081] Using the large model (LLM) technology, we can extract some information about the customer profile from this sentence and automatically make some decisions:

[0082] Family status: With children

[0083] Child age: kindergarten or primary school

[0084] Need: Pick up children

[0085] Is it the first car: Not the first car

[0086] Target models: New energy vehicles, sedans

[0087] Model database search matching models: xx, yy, zz

[0088] For this embodiment, the large model (LLM) can not only automatically mine the user's personal information and complete certain mining work, but also automatically match the user's demand conditions to the recommended models in the model library, thereby completing the "decision-making" task.

[0089] In the above embodiment, the user's conversation information and basic information are comprehensively analyzed based on the big model to obtain the user's intentions and possible behaviors, and then automatically make task decisions. This not only makes the decision results more reasonable, but also improves the efficiency of decision-making because the big model is used to make automatic decisions. In the decision-making process, the user's basic information and conversation information are comprehensively considered, which makes the decision more accurate.

[0090] See also Figure 2 In some embodiments, the task decision scheme includes one or more, and the task decision scheme includes: determining the correlation between the intentions of the user; performing comprehensive analysis and task decision based on the correlated intentions to obtain a decision result; performing task decision based on independent and uncorrelated intentions to obtain a decision result.

[0091] Considering that some intentions are related and some are independent, the large model can make task decisions based on the correlation between the user's intentions during the task decision process. For example, correlation analysis can be performed for related intentions, and task decisions can be made by comprehensive analysis based on the relationship between related intentions. Independent decisions can be made for unrelated intentions. In this way, multiple task decision results may be obtained, such as the decision results obtained by comprehensive analysis of multiple related intentions, and the decision results obtained by independent decisions on independent unrelated intentions.

[0092] By analyzing the relevance of intents and combining multiple related intents to make decisions, the decision results are more in line with user needs. And through correlation analysis, more information can be mined, further expanding the user's intent, making it easier to obtain the user's true intent, and even to mine intent that the user is not aware of, making the decision more intelligent and accurate.

[0093] See also Figure 2 In some embodiments, the task decision scheme includes one or more schemes, and the task decision scheme includes: determining the importance of each of the existing intentions to the user based on the basic information of the user. The large model makes task decisions based on the importance of the existing intentions to obtain a decision result.

[0094] For example, if a user has multiple intentions, the importance of each intention can be analyzed. For example, the importance of the intention can be analyzed based on the user's basic information, or the importance of the user's intention can be determined based on the user's conversation information. In this way, when making task decisions, priority decisions can be made for intentions with higher importance to obtain decision results that are more important to the user, and secondary decisions can be made for intentions with lower importance to the user. In this way, the primary and secondary relationships of the decision results can be distinguished, so that the obtained decision results are more in line with the user's needs.

[0095] In some embodiments, the importance of the user's current intention can be obtained based on the analysis of the user's conversation information, can also be obtained based on the user's basic information, or can be obtained based on a comprehensive analysis of the user's conversation information and basic information.

[0096] For example, the importance of the user's intention can also be understood as the priority of the intention. Intentions with higher importance can be understood as having higher priorities, and task decisions can be made based on the higher-important purposes, so that the decision results obtained can be more in line with user needs.

[0097] In some embodiments, different user intentions may have the same priority of importance, in which case the decision results corresponding to intentions of different dimensions can be recommended to the user at the same time. For example, users have the same intention strength for price intention and functional intention, in which case two parallel decision results can be obtained, one is the price decision result obtained by analyzing the price intention, and the other is the functional decision result obtained by analyzing the functional intention. Of course, in some embodiments, a comprehensive decision result can also be obtained based on a comprehensive analysis of multiple intentions, that is, in the decision result, not only the price factor but also the functional factor is taken into account.

[0098] See also Figure 2 In some embodiments, the task decision scheme includes one or more, and the task decision scheme includes: the big model predicts the user's current intention based on the user's basic information, and makes a decision based on the predicted intention and / or the current intention to obtain a decision result.

[0099] For example, some intentions may not be directly expressed by the user, that is, they cannot be directly extracted from the user's conversation information, but they can be mined after comprehensive analysis based on the user's basic information and conversation information. At this time, the predicted intention mined is also the information that the user is more concerned about. At this time, task decisions can be made based on the predicted intention, and then more user needs can be mined and met.

[0100] This application can make task decisions based on the predicted intent, or it can refer to the predicted intent and the user's existing intent for a comprehensive analysis and decision to obtain the final decision plan.

[0101] In some embodiments, the method also includes: predicting recommendation options that the user may be interested in based on the user's basic information, and determining recommendation reasons corresponding to the recommendation options; obtaining a decision option based on the recommendation option and the recommendation reasons, and recommending the decision option to the user terminal based on the importance of the decision option.

[0102] In some embodiments, not only a decision-making plan is required, but also reasons for recommending the plan. The reasons for recommending can be developed around key points of concern to users, thereby making the recommended plan more compatible with users, increasing user acceptance, and making the recommended plan more accurate and effective.

[0103] In some embodiments, the user said the following sentence: "We are currently mainly considering the safety of the car, and the overall expected price is about 150,000 to 200,000 yuan. By the way, we hope that the car has a larger space, but the configuration does not need to be too high."

[0104] Using the large model (LLM) technology, we can further match several car models based on the customer description, recommend models that are slightly higher than the customer's needs, and automatically make several decisions:

[0105] Pure price matching: xx, yy

[0106] Demand mining and matching: zz (Reason: The price of the low-end version is 210,000, which is slightly over the budget, but it has a larger space and can be used if the configuration requirements are not high.), Recommended reason: Although it is slightly over the expectations, it does have a larger space and better performance. There is a promotion going on recently. The original price was 230,000, and it is very cost-effective to buy it now.

[0107] Through the first-level price calculation and combined with the user's demand matching, appropriate expanded vehicle model recommendations were made, slightly exceeding customer needs, but sufficient reasons were given, which was equivalent to an expansion impact.

[0108] Alternatively, a trigger signal of "related recommendation" is given, and the user's basic information and the focus of the current discussion topic are combined to output differentiated and personalized recommended products through the generation capability of the large model (LLM). For example, although the models, brands, and configurations of "35-year-old men with families and children" and "22-year-old single women" may be very similar, "XX brand SUV" and "XX brand sedan" are recommended respectively based on their basic situations. At the same time, in the process of generating content by the large model (LLM), the attributes of the role are fully considered. For example, when recommending "35-year-old men with families and children", important information such as household and children can be emphasized, while when recommending "22-year-old single women", important information such as fashion and coolness can be emphasized, so as to highlight the characteristics of the "role".

[0109] In only some embodiments, the method further includes: determining the user's willingness strength, and giving priority to users and decision-making plans whose willingness strength meets a preset value.

[0110] For example, the user's intention level is determined, and corresponding strategies are recommended based on the intention level; for high intentions, manual intervention can be performed. In this way, the accuracy and success rate of recommendations can be further improved.

[0111] In some embodiments, the large model can on the other hand perform "(assisted) response decision making", which mainly relies on the understanding and generation capabilities of the large model (LLM).

[0112] For example, based on the intent information obtained above, the understanding ability of the large model (LLM) can be further used to determine the basic situation of the user:

[0113] 1. Intention to visit the store: High

[0114] 2. New car consultation: Yes

[0115] 3. Do you want to inquire: Yes

[0116] By judging the user's intention, the user is classified as "high intention", and then further generating capabilities, a reply can be generated: "We are very happy that you are interested in our brand. Regarding your questions about the new car, there will be a dedicated customer service professional to answer you. Transferring, please wait..."

[0117] At the same time, a trigger signal of "transfer to manual" is given to conduct manual intervention for such "high-intent" users to further facilitate the success of the transaction.

[0118] In some embodiments, the method further includes: obtaining historical behavior information of the user; making task decisions based on the historical behavior information, the basic information, and key information in the conversation information; and using the key information in the conversation information as an important consideration point for task decision making, and developing the obtained task decision plan around the corresponding key point information. The recommendation strategy should be developed around the key points and should match the user role identity characteristics.

[0119] In some embodiments, the method further includes: the character information and the vehicle type are matched, and a vehicle of a corresponding type is recommended based on the match.

[0120] Specifically, the integration of roles and the differentiated generation of content mainly rely on the prompt engineering of the large model (LLM). For example, to generate the response decision of "35-year-old male with a family and children", the prompt can be written in a similar template:

[0121] You are a salesperson at a 4S store of XX brand. Now you need to recommend a car to a 35-year-old man with a family and children. This user is more concerned about family use, school pickup, safety, spaciousness, etc. This user is an engineer in the IT industry with a relatively good income, ranging from 300,000 to 500,000 yuan per year. The models currently recommended to you by the store include:

[0122] Model A: A 2022 luxury SUV, priced at xx, with a large interior space, using xx engine, etc.

[0123] Model B: A high-end SUV of 2021, priced at xx, with luxurious interior, etc.

[0124] The above template is used as the prompt template of the large model (LLM) and input into the large model (LLM). By utilizing the powerful understanding and generation capabilities of the large model, a more accurate response decision can be made and a reasonable matching result can be given.

[0125] It should be noted that response decisions include but are not limited to recommendations, but also include multiple dimensions such as replies, quality inspection, user comments, and user screening.

[0126] The method provided in this application uses data of different dimensions, especially uses a large model (LLM) to mine text information, which has high accuracy and low labor cost. In addition, the user's role information is taken into account in the task decision-making process, and the role information is integrated into the decision-making process, so that the decision results are more compatible with the user, and the success rate and accuracy of the decision are improved.

[0127] In some embodiments, the method further includes: determining the level and role of the user based on the user's basic information; determining a target product that matches the user's level and role; and generating a decision plan recommended to the user based on the target product and its characteristics.

[0128] User information is regarded as a kind of "role information". When making response decisions in the large model (LLM), the attributes of the role will be fully considered to achieve personalized responses.

[0129] Examples of role attributes: For example, although the models, brands, and configurations of "35-year-old male with a family and children" and "22-year-old single female" may be similar, "XX brand SUV" and "XX brand sedan" are recommended based on their basic situations. At the same time, in the process of generating content by the large model (LLM), the attributes of the role are fully considered. For example, when recommending "35-year-old male with a family and children", important information such as household and children can be emphasized, while when recommending "22-year-old single female", important information such as fashion and coolness can be emphasized, so as to highlight the characteristics of the "role".

[0130] The association of roles is mainly reflected in the "decision-making" link. For example, when recommending a car model, the definition of the car brand and the matching degree of the role will be considered. For example, the matching degree of family cars and commercial vehicles is different for family car customers and elite commercial customers. In addition, when summarizing and describing the reasons for the recommendation, the description reasons given by the large model (LLM) can also clearly consider the attributes of the role. For example, family car customers are more concerned about the perspective of children. When describing, the large model (LLM) will often express empathy such as "It is more friendly to children...", "I can understand your consideration for children...", etc., so as to achieve a better anthropomorphic effect and get closer to the real marketing scene.

[0131] The large model (LLM) can serve as an important link in the information mining and response decision-making chain, triggering some key signals to achieve relatively fine-grained monitoring and a higher level of automation.

[0132] In the above embodiment, when making task decisions based on the big model, the user's multi-dimensional data information is taken into consideration, and technologies such as data mining, data analysis, and behavior prediction are integrated, so that the final decision-making plan is more accurate and effective.

[0133] For the user's conversation information, we can mine the user's current focus and interests. Based on the user's basic personal information, we can learn the user's deep-seated needs and the reasons for the current needs. This not only allows us to obtain comprehensive and multi-dimensional data information about the user, so that we can get more intelligent and accurate decision-making solutions based on multi-dimensional comprehensive data, but also can mine the user's deep-seated needs and recommend more reasonable solutions to the user.

[0134] It should be understood that, although the steps in the flowchart of the accompanying drawings are displayed in sequence as indicated by the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and these steps can be executed in other orders. Moreover, at least a portion of the steps in the accompanying drawings may include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least a portion of the steps or stages in other steps.

[0135] In one embodiment, Figure 3 As shown, a task decision device based on a large model is provided. The device can adopt a software module or a hardware module, or a combination of the two to become a part of a computer device. The device specifically includes:

[0136] An acquisition module 301 is used to acquire basic information of a user and determine a user portrait based on the basic information. The user portrait is used to divide the user into different levels and characterize the user's role.

[0137] The data processing module 302 is used to obtain the user's conversation information, and the large model performs user intention recognition and behavior prediction based on the conversation information and the user portrait;

[0138] The decision module 303 is used to make a task decision based on the intention recognition result and the behavior prediction result.

[0139] For the specific definition of the task decision device based on the large model, please refer to the definition of the task decision method based on the large model above, which will not be repeated here. Each module in the above device can be implemented in whole or in part by software, hardware and a combination thereof. Each of the above modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to each of the above modules.

[0140] In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as follows: Figure 4 As shown. The computer device includes a processor, a memory and a network interface connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store user-related data. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a task decision method based on a large model is implemented.

[0141] Those skilled in the art will understand that Figure 4 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0142] In one embodiment, a computer device is further provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above method embodiments when executing the computer program.

[0143] In one embodiment, a computer-readable storage medium is provided, storing a computer program, which implements the steps in the above method embodiments when executed by a processor.

[0144] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory or optical memory, etc. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0145] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0146] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the attached claims.

Claims

1. A task decision method based on a large model, It is characterized in that The method comprises: Obtaining basic information of the user, and determining a user portrait based on the basic information, wherein the user portrait is used to divide the user into different levels and characterize the user's role; The user's conversation information is obtained, and the big model recognizes the user's intention and predicts the user's behavior based on the conversation information and the user portrait; Based on the intention recognition result and the behavior prediction result, a task decision is made.

2. The method according to claim 1, It is characterized in that The task decision scheme includes one or more, and the task decision scheme includes: Determining the relevance of each of the user's existing intentions; Conduct comprehensive analysis and task decision-making based on relevant intentions to obtain decision results; Make task decisions based on independent and unrelated intentions and obtain decision results.

3. The method according to claim 1, It is characterized in that The task decision scheme includes one or more, and the task decision scheme includes: The importance of each of the existing intentions to the user is determined based on the basic information of the user. The large model makes a task decision based on the importance of the existing intention and obtains a decision result.

4. The method according to claim 1, It is characterized in that The task decision scheme includes one or more, and the task decision scheme includes: The large model predicts the user's current intention based on the user's basic information, and makes a decision based on the predicted intention and / or the current intention to obtain a decision result.

5. The method according to claim 1, It is characterized in that The method further comprises: Predicting a recommendation scheme that the user may be interested in based on the user's basic information, and determining a recommendation reason corresponding to the recommendation scheme; A decision solution is obtained based on the recommendation solution and the recommendation reason, and is recommended to the user terminal based on the importance of the decision solution.

6. The method according to claim 1, It is characterized in that The method further comprises: The user's willingness strength is determined, and users and decision plans whose willingness strength meets a preset value are given priority.

7. The method according to any one of claims 1 to 6, It is characterized in that The method further comprises: Obtain user's historical behavior information; Making task decisions based on the historical behavior information, the basic information, and key information in the dialogue information; The key information in the dialogue information is used as an important consideration point for task decision making, and the obtained task decision plan is developed around the corresponding key point information.

8. The method according to claim 1, It is characterized in that The method further comprises: Determine the level and role of the user based on the user's basic information; Determine the target products that match the user's level and role; A decision plan is generated and recommended to the user based on the target product and the characteristics of the target product.

9. A task decision-making device based on a large model, It is characterized in that The device comprises: An acquisition module is used to acquire basic information of a user and determine a user portrait based on the basic information. The user portrait is used to divide the user into different levels and characterize the user's role. A data processing module is used to obtain the user's conversation information. The large model recognizes the user's intention and predicts the user's behavior based on the conversation information and the user portrait. A decision module is used to make task decisions based on the intention recognition result and the behavior prediction result.

10. A computer device comprising a memory and a processor, wherein the memory stores a computer program. It is characterized in that When the processor executes the computer program, the steps of the method according to any one of claims 1 to 8 are implemented.

Images