Human mobility behavior generation method and system based on large language model

Through the human mobile behavior generation method of the large language model, the use of character portraits and thinking chain mechanisms to generate mobile intentions and map them to the real geographical location, solving the problem that traditional methods are difficult to personalize the generation of human mobile behavior, and achieving accurate and efficient mobility behavior simulation.

CN118673102BActive Publication Date: 2024-11-05TSINGHUA UNIVERSITY
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411164470.1
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-08-23
Publication Date
2024-11-05
Estimated Expiration
2044-08-23

AI Technical Summary

Technical Problem

Traditional methods are difficult to achieve personalized generation of human mobile behavior data, and it is difficult to use detailed population attributes to fit the movement laws of real urban populations, making it difficult to achieve accurate, efficient, semantically perceived human mobile behavior generation.

Method used

The human mobile behavior generation method based on the large language model, specify the role through character portraits, use the thinking chain mechanism of the large language model to generate high-level mobile intent, and use the mechanism model to map the intention template to the real geographical location to achieve accurate and efficient Semantically perceived human movement behavior generation.

Benefits of technology

It realizes the generation of accurate, efficient and semantic perception of human mobile behavior, can replace traditional high-cost transportation surveys, comprehensively simulate the mobile travel situation in cities, and is suitable for cities with different population densities and facilities distribution.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118673102B_ABST
    Figure CN118673102B_ABST
Patent Text Reader

Abstract

The present invention relates to the field of artificial intelligence technology, and provides a method and system for generating human mobile behavior based on a large language model. The method determines context-aware prompt words based on a character portrait database; based on the context-aware prompt words, a complete mobile intention template is generated by reasoning using a thinking chain mechanism of a large language model; the complete mobile intention template is input into a pre-trained mechanism model to obtain a human mobile behavior mapped to a real geographical location output by the mechanism model; the mechanism model is used to map the complete mobile intention template to a real geographical location based on a moving distance corresponding to the complete mobile intention template. The present invention uses character portraits to specify the role played by a large language model, and then generates a high-level mobile intention through step-by-step reasoning of the large language model. Finally, the mechanism model is used to map the abstract intention template to a real geographical location, so as to achieve accurate, efficient, and semantically perceptually rich generation of human mobile behavior.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a method and system for generating human mobility behavior based on a large language model. Background Art

[0002] Human mobility behavior is closely related to urban issues such as traffic congestion, energy consumption distribution, and epidemic spread. Accurate modeling of human mobility behavior plays a very important role in traffic control, energy supply, health policy, urban planning, etc. For example, during large-scale events, how to make temporary adjustments to traffic and how to formulate emergency plans; for example, communication network optimization, base station density setting, and dynamic adjustment of base station working power. The traditional way to collect large-scale crowd mobility behavior data is through family travel surveys or portable records of personal sensors on smartphones. Both methods make it difficult to achieve personalized generation of human behavior data and to use more detailed population attributes to fit the mobility patterns of real urban populations. Therefore, how to achieve accurate, efficient, and semantically perceptual generation of human mobility behavior is a challenging research problem. Summary of the invention

[0003] The present invention provides a method and system for generating human mobility behavior based on a large language model. Character portraits are used to specify the roles played by the large language model, and high-level mobility intentions are generated through step-by-step reasoning of the large language model. Finally, a mechanism model is used to map the abstract intention template to the real geographic location, thereby achieving accurate, efficient, and semantically-aware generation of human mobility behavior.

[0004] The present invention provides a method for generating human mobile behavior based on a large language model, comprising: determining context-aware prompt words based on a character portrait database; the character portrait database is used to specify the human role played by a pre-trained large language model; based on the context-aware prompt words, using the thinking chain mechanism of the large language model, inferring and generating a complete mobile intention template; the complete mobile intention template is the mobile behavior of the specified role at different times within a preset time period; inputting the complete mobile intention template into a pre-trained mechanism model to obtain the human mobile behavior mapped to the real geographical location output by the mechanism model; wherein the mechanism model is used to map the complete mobile intention template to the real geographical location based on the moving distance corresponding to the complete mobile intention template.

[0005] According to a method for generating human movement behavior based on a large language model provided by the present invention, the method determines context-aware prompt words based on a character portrait database, including: determining character portrait information based on the character portrait database; determining the context-aware prompt words based on the character portrait information, a small sample example and a preceding movement intention; the small sample example includes a character portrait example and a time-intention template corresponding to the character portrait example; the preceding movement intention is the movement intention of a human character at a time before the current movement intention; and the character portrait example is used to characterize feature information of the example character.

[0006] According to a method for generating human movement behavior based on a large language model provided by the present invention, the time-intention template corresponding to the character portrait example includes: the movement intentions corresponding to the example character at different times in the example time period, and based on the connection between different movement intentions, explaining the reason for the next movement intention.

[0007] According to a method for generating human mobile behavior based on a large language model provided by the present invention, based on the context-aware prompt words, the thinking chain mechanism of the large language model is used to infer and generate a complete mobile intention template, including: decomposing the reasoning generation process of the complete mobile intention template into steps of generating a plurality of intentions, and in each intention generation step, providing all the results of the preceding mobile intention as the reasoning basis of the context-aware thinking chain; wherein the generation process of each intention includes: inputting the context-aware prompt words into the large language model to obtain the current mobile intention output by the large language model; the current mobile intention is the mobile behavior of the specified character at the current moment; based on the current mobile intention, sampling from a preset duration distribution to determine the duration corresponding to the current mobile intention.

[0008] According to a method for generating human movement behavior based on a large language model provided by the present invention, the mechanism model adopts a gravity model; the gravity model satisfies the formula:

[0009]

[0010] in, For Region i and Region j Migration traffic between K is the normalized coefficient, m i For Region i POI density, m j For Region j POI density, r ij For Region i and Regionj The distance between.

[0011] According to a method for generating human mobility behavior based on a large language model provided by the present invention, the mechanism model is used to map the complete mobility intention template to a real geographic location based on the moving distance corresponding to the complete mobility intention template, including: for each mobility intention in the complete mobility intention template, POI sampling is performed based on a previous position to determine the moving distance; the moving distance is the distance from the previous position to the current position; the previous position is the real geographic location corresponding to the previous mobility intention, and the current position is the real geographic location corresponding to the current mobility intention; based on the moving distance, the current position is mapped to the current mobility intention.

[0012] The present invention also provides a human mobility behavior generation system based on a large language model, comprising: a determination module, used to determine context-aware prompt words based on a character portrait database; the character portrait database is used to specify the human role played by the pre-trained large language model; a generation module, used to infer and generate a complete mobility intention template based on the context-aware prompt words using the thinking chain mechanism of the large language model; the complete mobility intention template is the mobility behavior of the specified role at different times within a preset time period; a mapping module, used to input the complete mobility intention template into the pre-trained mechanism model to obtain the human mobility behavior mapped to the real geographic location output by the mechanism model; wherein the mechanism model is used to map the complete mobility intention template to the real geographic location based on the moving distance corresponding to the complete mobility intention template.

[0013] The present invention also provides an electronic 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 program, the method for generating human movement behavior based on a large language model as described above is implemented.

[0014] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any of the above-described methods for generating human movement behavior based on a large language model.

[0015] The present invention also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the computer program implements any of the above-mentioned methods for generating human movement behavior based on a large language model.

[0016] The present invention provides a method and system for generating human mobile behavior based on a large language model. The method determines context-aware prompt words based on a character portrait database; based on the context-aware prompt words, the thinking chain mechanism of the large language model is used to infer and generate a complete mobile intention template; the complete mobile intention template is input into a pre-trained mechanism model to obtain the human mobile behavior mapped to the real geographical location output by the mechanism model; the mechanism model is used to map the complete mobile intention template to the real geographical location based on the moving distance corresponding to the complete mobile intention template. The present invention uses character portraits to specify the role played by the large language model, and then generates high-level mobile intentions through step-by-step reasoning of the large language model. Finally, the mechanism model is used to map the abstract intention template to the real geographical location, which can achieve accurate, efficient, and semantically perceptual human mobile behavior generation. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0018] Figure 1 It is a flow chart of a method for generating human mobility behavior based on a large language model provided by the present invention.

[0019] Figure 2 It is a schematic diagram of the principle of a method for generating human movement behavior based on a large language model provided by the present invention.

[0020] Figure 3 It is a schematic diagram of the principle of the context-aware thinking chain provided by the present invention.

[0021] Figure 4 It is a structural schematic diagram of a human movement behavior generation system based on a large language model provided by the present invention.

[0022] Figure 5 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0023] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0024] Please refer to Figure 1 , Figure 1 A schematic flow chart of a method for generating human movement behavior based on a large language model provided by the present invention.

[0025] The present invention provides a method for generating human movement behavior based on a large language model, comprising:

[0026] 101: Determine context-aware prompt words based on a character portrait database; the character portrait database is used to specify the human role played by the pre-trained large language model;

[0027] 102: Based on context-aware prompt words, the thinking chain mechanism of the large language model is used to infer and generate a complete movement intention template; the complete movement intention template is the movement behavior of the specified character at different times within a preset time period;

[0028] 103: Input a complete mobile intention template into a pre-trained mechanism model to obtain a human movement behavior output by the mechanism model that is mapped to a real geographic location; wherein the mechanism model is used to map the complete mobile intention template to a real geographic location based on a movement distance corresponding to the complete mobile intention template.

[0029] In order to solve the technical problems existing in the prior art, the present invention proposes a method for generating human mobile behavior based on the reasoning ability and role-playing ability of large language models (LLM). First, the human role played by the pre-trained large language model is specified based on the character portrait database to realize the personalized generation of human mobile behavior, thereby determining the context-aware prompt words. The context-aware prompt words are used as the input of the large language model, and a high-level complete mobile intention template is generated through the step-by-step reasoning (thinking chain mechanism) of the large language model. This is a shift in the generation paradigm, which restates the mobility generation as a step-by-step reasoning problem. Finally, the pre-trained mechanism model is used to map the abstract and complete mobile intention template to the human mobile behavior in the real geographical location, which can realize accurate, efficient, and semantically aware human mobile behavior generation.

[0030] Specifically, the definition of the mobile behavior problem is: a mobile behavior data consists of a sequence of points that are continuous in time: , each point It can be expressed as . Indicates a spatial location, which can be a discrete ID of a region or a longitude and latitude coordinate. Indicates access time. Represents the intention type to come to this place. Given a real-world mobile behavior dataset ,in is a sequence of mobile behaviors. The target generates new mobile behavior data , so that the generated data shows similar characteristics and movement patterns as the original real data, and can provide practical value for downstream applications.

[0031] As a universal modeling framework for urban trajectories, the present invention can be flexibly applied to cities with different population densities and facility distributions. It extracts universal mobility patterns from a small amount of trajectory data provided by telecom operators or software service providers, which can replace traditional high-cost traffic travel surveys to fully simulate the mobility situation of the city. First, from a small amount of real human mobility data (hundreds of items) provided by telecom operators, the duration distribution of various intentions and the parameters of the fitted gravity model are statistically analyzed. Since the final generated result is implemented at a specific location in the city, it is also necessary to obtain specific information about POIs in the city from the map service and classify them to correspond to each intention. In addition, demographic data in the city, including occupation, gender, education, consumption level, etc., are required to be specifically specified during LLM role-playing. After the pre-data is prepared, the present invention can be used to generate urban population mobility behavior.

[0032] Please refer to Figure 2 , Figure 2 A schematic diagram of the principle of a method for generating human movement behavior based on a large language model provided by the present invention.

[0033] Please refer to Figure 3 , Figure 3 A schematic diagram of the principle of the context-aware thinking chain provided by the present invention.

[0034] As a preferred embodiment, context-aware prompt words are determined based on a character portrait database, including: determining character portrait information based on the character portrait database; determining context-aware prompt words based on the character portrait information, a small sample example and a preceding movement intention; the small sample example includes a character portrait example and a time-intention template corresponding to the character portrait example; the preceding movement intention is the movement intention of a human character at a time before the current movement intention; the character portrait example is used to characterize the feature information of the example character.

[0035] As a preferred embodiment, the time-intention template corresponding to the character portrait example includes: the movement intentions corresponding to the example character at different times in the example time period, and explains the reason for the next movement intention based on the connection between different movement intentions.

[0036] In this embodiment, sampling is performed from demographic data, the role-playing identity is specified for LLM, and a small sample example and initial time information are provided. The next movement intention (current movement intention) is recursively generated through LLM step-by-step reasoning, and then the duration of this intention is sampled from the distribution of duration corresponding to the current movement intention to obtain the "time-intention" point. The generated results are gradually supplemented and improved until the expected time range ends.

[0037] Specifically, the character portrait information is determined through the character portrait database. As part of the context-aware prompt words, the character portrait information will be provided to the LLM in the form of specified role-playing and participate in each generation step as a global prompt word. The character portrait information includes occupation, gender, income level, education level, etc. In order to more clearly capture the pattern of how the character portrait affects the mobile behavior, a small sample example (e.g., 8 examples) is provided. The small sample example includes a character portrait example extracted from a real data set and a time-intention template corresponding to the character portrait example. The character portrait example is used to characterize the characteristic information of the example role, such as occupation, gender, income level, education level, etc. At each time event point of the provided example (different moments in the example time period), the thinking process behind each mobile intention of the example role is manually annotated to find the connection between the current schematic intention and the previous schematic intention, and explain the reason for making the current decision (next mobile intention).

[0038] An example of this is:

[0039] Here's a high school teacher's schedule for a day on Saturday:

[0040] [[Eating, (06:11, 7:28)], (Reason: As a high school teacher, he is used to getting up early and likes to eat breakfast after getting up early.)

[0041] [Dealing with daily chores, (07:48, 08:32)], (Reason: After breakfast on weekends, he usually goes out to deal with some daily affairs that he has no time to deal with on weekdays.)

[0042] [Go home, (9:00, 11:00)], (Reason: After finishing all the things, I decided to go home and maybe have lunch at home.)

[0043] [Seeking medical treatment, (13:44, 17:03)], (Reason: Today is Saturday and I don’t have to go to work, so I decided to go to the hospital this afternoon to check out some of my recent minor illnesses)

[0044] [Go home, (19:00, 23:59)], (Reason: saw a doctor in the afternoon and rested at home in the evening.) ]

[0046] As a preferred embodiment, based on context-aware prompt words, a thinking chain mechanism of a large language model is used to infer and generate a complete mobile intention template, including: decomposing the reasoning generation process of the complete mobile intention template into steps of generating several intentions, and in each intention generation step, providing all the results of the previous mobile intention as the reasoning basis of the context-aware thinking chain; wherein the generation process of each intention includes: inputting context-aware prompt words into the large language model to obtain the current mobile intention output by the large language model; the current mobile intention is the movement behavior of the specified character at the current moment; based on the current mobile intention, sampling from a preset duration distribution to determine the duration corresponding to the current mobile intention.

[0047] In this embodiment, a daily behavior template is generated given a context-aware prompt word: , including the intention and the corresponding time period Two parts. For example, [["sleep", "(00:00, 08:33)"], ["go to work", "(09:47, 17:49)"], ["eat", "(18:45, 19:49)"], ["shopping", "(20:01, 20:35)"], ["go home", "(21:40, 23:59)"]]. In order to ensure that the generated events are logically arranged in time and interrelated, (for example, if a person works long hours from morning to night, then he is likely to participate in sports or leisure activities after finishing work, such as watching a movie.) The context-aware chain of thought (COT) mechanism is introduced. LLM is not required to generate a complete mobile intention template at one time. Instead, the overall generation process is decomposed into several steps to generate separate intentions. At each generation step, the entire result of the previous mobile intention is provided as the basis for context-aware chain of thought reasoning:

[0048] .

[0049] This Markov chain generation process can significantly enhance the rationality and accuracy of LLM reasoning.

[0050] When asking the LLM, the LLM is prompted to think based on the previous generation results (context-aware prompt words) and make a decision like the example provided (current movement intention). In addition, the current time and date information are further incorporated into the reasoning as contextual prompt words to improve the consistency between the generation results and time. (For example, at 10 o'clock in the evening, most people will not choose to go hiking, but go home.)

[0051] In addition, the distribution of the number of daily intents is calculated from 200 real data, and samples are taken from them before the generation process starts as the limit of the total number of intents to better fit the distribution of real data. There is no need to directly fit the distribution of trajectory data, so there will be no privacy leakage. Still using these 200 data, the duration distribution of each type of intent is also calculated. Once LLM determines the type of the next intent in a certain generation step, it samples from the duration distribution of the intent and calculates the start and end times to generate a complete time-intent point.

[0052] As a preferred embodiment, the mechanism model adopts a gravity model; the gravity model satisfies the formula:

[0053] ;

[0054] in, For Region i and Region j Migration traffic between K is the normalized coefficient, m i For Region i POI density, m j For Region j POI density, r ij For Region i and Region j The distance between.

[0055] As a preferred embodiment, the mechanism model is used to map the real geographic location of the complete mobile intention template based on the moving distance corresponding to the complete mobile intention template, including: for each mobile intention in the complete mobile intention template, POI sampling is performed based on the previous position to determine the moving distance; the moving distance is the distance from the previous position to the current position; the previous position is the real geographic location corresponding to the previous mobile intention, and the current position is the real geographic location corresponding to the current mobile intention; based on the moving distance, the current position is mapped to the current mobile intention.

[0056] In this embodiment, in order to map the complete mobile intention template to the real geographic location, the home and work locations can be directly sampled from the addresses identified by the real data, but for other intentions, how to correspond it to one of the millions of POI locations is a challenge. Considering both performance and implementation cost, this embodiment selects the gravity model in the mechanism model to solve this problem. The gravity model emphasizes the importance of distance in human migration.

[0057] Considering nonlinear factors, the migration traffic volume between two areas can be expressed as: .in, It is a region i and Region j The distance between area i POI density, m j For Region j POI density. represents a decreasing function of distance.

[0058] For each movement intention in the complete movement intention template, the gravity model calculates the probability distribution and samples a specific POI location. Specifically, when the template is Go to When the mechanism model is based on the last location (Pre-order position) Divide the map for probabilistic sampling and select the desired , select the appropriate geographic location (current location) to transform an abstract mobile intention behavior into a specific geographic trajectory.

[0059] In the experiment, the density of POI (Point of Interest) was used instead of the population of the area. It is believed that the higher the density of POI, the more attractive it is to individuals. The map around the previous location (previous location) is divided into concentric circles at intervals of 1 km. For all POIs in a ring, Set to the POI density in the ring (number / ring area), Set to , are fitted by a small amount of data. Assuming that the upper limit of the distance of a single trip for most people is 10km, the POI search is limited to a circle with a radius of 10km.

[0060] By simulating the mobility behavior of a typical urban population, the final result should be able to reproduce the urban mobility patterns at a macro level, and be consistent with the results of traditional traffic surveys and passive trajectory recording based on positioning.

[0061] The human mobility behavior generation system based on a large language model provided by the present invention is described below. The human mobility behavior generation system based on a large language model described below and the human mobility behavior generation method based on a large language model described above can refer to each other.

[0062] Please refer to Figure 4 , Figure 4 A schematic diagram of the structure of a human movement behavior generation system based on a large language model provided by the present invention.

[0063] The present invention also provides a human mobility behavior generation system based on a large language model, comprising: a determination module 1, used to determine context-aware prompt words based on a character portrait database; the character portrait database is used to specify the human role played by the pre-trained large language model; a generation module 2, used to infer and generate a complete mobility intention template based on the context-aware prompt words using the thinking chain mechanism of the large language model; the complete mobility intention template is the mobility behavior of the specified role at different times within a preset time period; a mapping module 3, used to input the complete mobility intention template into the pre-trained mechanism model to obtain the human mobility behavior mapped to the real geographic location output by the mechanism model; wherein the mechanism model is used to map the complete mobility intention template to the real geographic location based on the moving distance corresponding to the complete mobility intention template.

[0064] Figure 5 An example of a physical structure diagram of an electronic device is shown in FIG. Figure 5 As shown, the electronic device may include: a processor 510, a communication interface 520, a memory 530 and a communication bus 540, wherein the processor 510, the communication interface 520 and the memory 530 communicate with each other through the communication bus 540. The processor 510 may call the logic instructions in the memory 530 to execute the method for generating human mobile behavior based on the large language model, the method comprising: determining context-aware prompt words based on a character portrait database; the character portrait database is used to specify the human role played by the pre-trained large language model; based on the context-aware prompt words, using the thinking chain mechanism of the large language model, reasoning and generating a complete mobile intention template; the complete mobile intention template is the mobile behavior of the specified role at different times within a preset time period; inputting the complete mobile intention template into the pre-trained mechanism model to obtain the human mobile behavior mapped to the real geographical location output by the mechanism model; wherein the mechanism model is used to map the complete mobile intention template to the real geographical location based on the moving distance corresponding to the complete mobile intention template.

[0065] In addition, the logic instructions in the above-mentioned memory 530 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0066] On the other hand, the present invention also provides a computer program product, which includes a computer program, which can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the human mobility behavior generation method based on the large language model provided by the above-mentioned methods, the method including: determining context-aware prompt words based on a character portrait database; the character portrait database is used to specify the human role played by the pre-trained large language model; based on the context-aware prompt words, using the thinking chain mechanism of the large language model, reasoning and generating a complete mobile intention template; the complete mobile intention template is the movement behavior of the specified role at different times within a preset time period; inputting the complete mobile intention template into the pre-trained mechanism model to obtain the human mobility behavior mapped to the real geographic location output by the mechanism model; wherein the mechanism model is used to map the complete mobile intention template to the real geographic location based on the moving distance corresponding to the complete mobile intention template.

[0067] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the human mobility behavior generation method based on the large language model provided by the above-mentioned methods, the method comprising: determining context-aware prompt words based on a character portrait database; the character portrait database is used to specify the human role played by the pre-trained large language model; based on the context-aware prompt words, using the thinking chain mechanism of the large language model, inferring and generating a complete mobile intention template; the complete mobile intention template is the movement behavior of the specified role at different times within a preset time period; inputting the complete mobile intention template into the pre-trained mechanism model to obtain the human mobility behavior mapped to the real geographic location output by the mechanism model; wherein the mechanism model is used to map the complete mobile intention template to the real geographic location based on the moving distance corresponding to the complete mobile intention template.

[0068] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0069] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0070] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these 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 invention.

Claims

1. A method for generating human movement behavior based on a large language model, characterized in that: include: Determine context-aware prompt words based on the character portrait database; The character portrait database is used to specify the human roles played by the pre-trained large language model; Based on the context-aware prompt words, using the thought chain mechanism of the large language model, inferring and generating a complete movement intention template; The complete movement intention template is the movement behavior of the specified character at different times within a preset time period; Inputting the complete movement intention template into the pre-trained mechanism model to obtain the human movement behavior output by the mechanism model mapped to the real geographic location; The mechanism model is used to map the complete movement intention template to a real geographical location based on the movement distance corresponding to the complete movement intention template; The prompt words based on the context perception use the thought chain mechanism of the large language model to infer and generate a complete movement intention template, including: Decomposing the reasoning generation process of the complete mobile intention template into steps of generating several intentions, in each intention generation step, providing all results of the preceding mobile intention as the reasoning basis of the context-aware thinking chain; The generation process of each intent includes: Inputting the context-aware prompt word into the large language model to obtain a current movement intention output by the large language model; the current movement intention is a movement behavior of a specified character at a current moment; Based on the current movement intention, sampling from a preset duration distribution to determine a duration corresponding to the current movement intention; The mechanism model is used to map the complete movement intention template to a real geographical location based on a movement distance corresponding to the complete movement intention template, including: For each movement intention in the complete movement intention template, POI sampling is performed based on the previous position to determine the movement distance; the movement distance is the distance from the previous position to the current position; the previous position is the real geographical location corresponding to the previous movement intention, and the current position is the real geographical location corresponding to the current movement intention; Based on the moving distance, the current position mapping is performed on the current moving intention.

2. The method for generating human movement behavior based on a large language model according to claim 1, characterized in that: The determining of context-aware prompt words based on the character portrait database includes: Determining character portrait information based on the character portrait database; Based on the character portrait information, a small sample example and a preceding movement intention, the context-aware prompt word is determined; the small sample example includes a character portrait example and a time-intention template corresponding to the character portrait example; the preceding movement intention is the movement intention of a human character at a time before the current movement intention; the character portrait example is used to characterize the feature information of the example character.

3. The method for generating human movement behavior based on a large language model according to claim 2, characterized in that: The time-intention template corresponding to the character portrait example includes: The example character's corresponding movement intentions at different times during the example time period, and based on the connection between different movement intentions, explain the reason for the next movement intention.

4. The method for generating human movement behavior based on a large language model according to claim 1, characterized in that: The mechanism model adopts a gravity model; the gravity model satisfies the formula: ; in, For Region i and Region j Migration traffic between K is the normalized coefficient, m i For Region i POI density, m j For Region j POI density, r ij For Region i and Region j The distance between.

5. A human movement behavior generation system based on a large language model, characterized in that: include: A determination module, used for determining context-aware prompt words based on a character portrait database; The character portrait database is used to specify the human roles played by the pre-trained large language model; A generation module, configured to generate a complete movement intention template by reasoning based on the context-aware prompt words and using the thought chain mechanism of the large language model; The complete movement intention template is the movement behavior of the specified character at different times within a preset time period; A mapping module, used for inputting the complete movement intention template into the pre-trained mechanism model to obtain the human movement behavior output by the mechanism model mapped to the real geographical location; The mechanism model is used to map the complete movement intention template to a real geographical location based on the movement distance corresponding to the complete movement intention template; The prompt words based on the context perception use the thought chain mechanism of the large language model to infer and generate a complete movement intention template, including: Decomposing the reasoning generation process of the complete mobile intention template into steps of generating several intentions, in each intention generation step, providing all results of the preceding mobile intention as the reasoning basis of the context-aware thinking chain; The generation process of each intent includes: Inputting the context-aware prompt word into the large language model to obtain a current movement intention output by the large language model; the current movement intention is a movement behavior of a specified character at a current moment; Based on the current movement intention, sampling from a preset duration distribution to determine a duration corresponding to the current movement intention; The mechanism model is used to map the complete movement intention template to a real geographical location based on a movement distance corresponding to the complete movement intention template, including: For each movement intention in the complete movement intention template, POI sampling is performed based on the previous position to determine the movement distance; the movement distance is the distance from the previous position to the current position; the previous position is the real geographical location corresponding to the previous movement intention, and the current position is the real geographical location corresponding to the current movement intention; Based on the moving distance, the current position mapping is performed on the current moving intention.

6. An electronic 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 program, the method for generating human movement behavior based on a large language model as described in any one of claims 1 to 4 is implemented.

7. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for generating human movement behavior based on a large language model as described in any one of claims 1 to 4 is implemented.

8. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the method for generating human movement behavior based on a large language model as described in any one of claims 1 to 4 is implemented.