A dynamic adjustment method for multi-link plot based on large language model

Abstract

The present invention relates to a dynamic adjustment method of a multi-link plot based on a large language model, and a script large language model for generating a script is constructed to realize the generation of an unexecuted alternating script of each relevant user based on the user's behavior data and environmental factors in the plot scene, and to replace the remaining script of the initial script that the user has not executed. In the present invention, as the user executes each version of the script (including the first execution of the initial script and the execution of the alternating script as the previous version of the script), a new version of the alternating script is continuously generated to replace the initial script and iteratively alternating the script, that is, to continuously update the script or the remaining script that the user has not executed.

Description

Technical Field

[0001] The present invention relates to the technical field of automatic scenario generation, and more specifically, to a dynamic adjustment method for a multi-link plot based on a large language model. Background Art

[0002] Entertainment products or entertainment activities based on plot advancement include online entertainment products (such as online games, stand-alone games), offline entertainment products (such as offline games, play-execution of plot tasks, etc., mainly script killing, escape room, etc., and tourism products that apply plot tasks to travel itineraries, etc.). Usually, a script includes several plot segments, and the same plot point can be set with several plot segments, and the script is generated according to actual needs. Among them, whether it is an offline entertainment product or an offline entertainment product, the plot segments set by the plot point are pre-set and selected when they need to be selected to form a script.

[0003] In actual implementation, a scene usually only has one or several scripts of different difficulty levels. Each script can include multiple branch plots. The corresponding branch plots can be selected according to the completion of the link plots at different plot points. For example, in the escape room, different clues are obtained at the same location, and different branch plots will be unfolded accordingly. However, no matter how many branch plots are set in the script, the link plots contained in the script are fixed (including fixed order and fixed content). In other words, the execution of the script is actually the selection of branch plots, rather than modifying the script.

[0004] The same is true for online entertainment products. Even though online entertainment products are more likely to set up more different plot points at the plot points and more likely to achieve real-time selection of plot points, in fact, online entertainment products should be regarded as a larger script, and the real-time selection of plot points experienced is essentially still the selection of different branch plots in the script.

[0005] It can be seen that whether it is an online entertainment product or an offline entertainment product, the script cannot be changed during the process, that is, the order and content of the plot included in the script cannot be changed.

[0006] In the prior art, Chinese invention patent application No. 201810879832.3 discloses a method for generating tourism information, comprising: obtaining a user portrait of a user, wherein the user portrait includes user characteristics of the user; generating tourism information for the user based on the user characteristics, wherein the tourism information includes at least one of the following: tourism content recommended to the user, and corresponding reasons for the recommendation.

[0007] According to another aspect of the embodiment of the invention patent application, a method for generating tourism information is also provided, including: detecting an operation, wherein the operation object is tourism; in response to the operation, obtaining a user portrait of a user corresponding to the operation, wherein the user portrait includes user characteristics of the user; based on the user characteristics, generating tourism information for the user, wherein the tourism information includes at least one of the following: tourism content recommended to the user, and corresponding reasons for the recommendation; pushing tourism information to the user.

[0008] In the embodiment of the invention patent application, a method of automatically generating travel information based on a user portrait is adopted. After obtaining the user portrait of the user, the client generates travel information for the user based on the user characteristics included in the user portrait, wherein the user portrait includes the user characteristics of the user, and the travel information includes at least one of the following: travel content recommended to the user, and the corresponding reason for the recommendation.

[0009] In the above process, travel information can be generated according to the user characteristics included in the user portrait, without the need to manually write travel information, thereby improving the efficiency of generating travel information. In addition, since the user portrait can well reflect the user characteristics, the use of the user portrait can generate travel information that better meets the user's needs, thereby improving the user's experience of using the software.

[0010] It can be seen that the solution provided by the invention patent application can achieve the purpose of automatically generating tourism information, thereby improving the efficiency of generating tourism information, and further solving the technical problems that the related technology cannot automatically generate tourism information, resulting in low efficiency of tourism information generation and lack of personalization of tourism information.

[0011] However, the essence of the invention patent application is conditional screening, which has a coarse screening granularity and cannot achieve the "personalization" in conventional perception. Its automatic generation of tourism information is actually a pre-made plan suitable for a large group of people, and is actually the process and result of customers choosing from several options.

[0012] In the prior art, Chinese invention patent application 202210245166.4 discloses a method for operating a cultural tourism self-service cabinet system based on script-killing, wherein step 1: receiving tourist answer data, and analyzing the tourist's script preference type based on the tourist answer data. After the user arrives at the scenic spot self-service cabinet, click on the screen homepage of the human-computer interaction module of the self-service cabinet system, tap the screen to enter the answer module test, and the user answers according to the prompts. Through the answer data, the information processing module will analyze and recommend the script type that the user may like.

[0013] Step 2: Based on the analyzed script preference type, push the corresponding category of scripts. According to the analysis results of step 1, the information visualization module will push the corresponding category of scripts for visitors to view.

[0014] Step 8: Issue plot guidance instructions to guide tourists to other self-service counters in turn. Other self-service counters provide corresponding script clues according to the plot, thereby guiding tourists to traverse all clues of the corresponding plot line and finally complete the plot. After tourists get clothes and props, the information visualization module of the self-service counter system will display the plot guidance instructions, that is, guide tourists to the next clue location according to the plot, that is, other networked self-service counters set up at the corresponding scenic spots in the scenic area.

[0015] The technical solution of the above-mentioned invention patent application is a typical product for selecting a script and guiding the game through the script.

[0016] If the idea of ​​offline games and games guided by scripts is expanded within a reasonable scope and applied to a larger geographical area, the freedom of users will be greatly restricted when they play offline games and games based on scripts with fixed plot content and sequence. When the geographical scope is expanded from a small closed space to a large space, especially a large space that requires users to move between multiple small spaces at a certain distance, whether it is a large closed space or a large open space, the experience of playing script-killing games or escape rooms is not as good as that of a small closed space.

[0017] Another important aspect is that when the user needs to move within or between multiple small spaces, when an emergency occurs in a small space that the user has not reached or between two adjacent small spaces, the user needs to avoid the location of the emergency to avoid danger, which not only ensures the user's offline game and play schedule, but also ensures the user's personal and property safety. The existing technology does not currently exist to modify the script that has been determined and obtained based on emergency events, so that the user can naturally and unconsciously avoid emergency events through the modified plot, which not only ensures the user's personal and property safety, but also ensures that the user can smoothly play offline games and play.

[0018] In the prior art, Chinese invention patent application 202111020022.0 discloses a scene interaction method. Its embodiment 1 records that the first interaction information sent by any interactive device is obtained, wherein the first interaction information includes a user identifier; matching in the user database based on the user identifier to determine the user information corresponding to the user identifier; matching in the script database based on the first interaction information and the user information to determine the second interaction information corresponding to the first interaction information; sending the second interaction information to the interactive device, wherein the interactive device is used to output the second interaction information and collect the user's next first interaction information. The technical solution of the above-mentioned invention embodiment realizes personalized recommendation of scripts by matching the user database and the script database, so that users can get personalized plot guidance, thereby improving the user's immersive experience.

[0019] Its embodiment 2 records that the script can be understood as the development outline of the story, which can be used to determine the development direction of the story. Specifically, by filtering the script in the script database through user information, a story script with strong relevance to the user can be selected instead of a general script, so that the user can be better immersed in the plot. For example, the script is filtered in the script database according to the degree of completion and the number of completed interactions of the user in the current experience in the user information. For example, if the degree of completion and the number of completed interactions of user 1 are much greater than those of other users, user 1 will be matched with a more difficult story script to ensure that all users are in the plot.

[0020] The technical solution of the above invention patent application also records a story script determination unit, which is used to screen the scripts in the script database based on the user information and determine the story scripts that match the user information;

[0021] The "personalization" described in the above invention patent application is actually to select the most suitable script for the current user from the scripts preset in the script database through conditional screening to achieve the so-called "personalization". However, the scripts in the script database must be limited, and the combination methods must be exhaustive. Theoretically speaking, it is not a true "personalization". The "personalization" described is actually personalization for a large group of people, not for each person. It is essentially the process and result of selecting from a number of preset scripts.

[0022] Moreover, it is obvious that the technical solution of the above-mentioned invention patent application cannot generate new scripts for emergency events to guide users to avoid emergency events.

[0023] To sum up, the existing technology has the deficiency of being unable to truly achieve "personalization". Furthermore, it is impossible to "personalize" each user's script in real time for emergency events, so that each user can naturally and unconsciously avoid emergency events in his or her own script, thereby ensuring the personal and property safety of users and ensuring that users can smoothly carry out offline games and play. Summary of the invention

[0024] The purpose of the present invention is to overcome the shortcomings of the prior art and provide a dynamic adjustment method for multi-link plots based on a large language model, describe the behavioral data and global factors of all users, and use the large language model to generate changing scripts adapted to each user. The scripts that influence each other serve as the plot background, and the "personalized" requirements of each user for the script are met in real time; it can also guide users to optimally handle various precautions during the entertainment process through "personalized" changing scripts.

[0025] The technical solution of the present invention is as follows:

[0026] A method for dynamically adjusting a multi-stage plot based on a large language model, comprising: constructing a large language model for a script; obtaining an initial script including a multi-stage plot; obtaining behavioral data of a user executing the stage plot, and parsing and generating a behavioral semantic text describing the meaning of the behavioral data; inputting the behavioral semantic text and the initial script into the large language model for the script, generating an alternative script that has not been executed by the user, and replacing the remaining script of the initial script that has not been executed by the user.

[0027] As a preferred method, the link plot of the current link currently executed by the user is obtained; the behavior data of the user executing the link plot of the current link is obtained, and a behavior semantic text describing the meaning of the behavior data is parsed and generated; the behavior semantic text and the link plot of the current link are input into the script language model to generate the user's execution result of the link plot of the current link and the evaluation of the execution result; the execution result evaluation is the difference between the user's execution result of the link plot of the current link and the preset result corresponding to the link plot of the current link;

[0028] The script language model is used again to determine whether the difference affects the plot of the subsequent link not executed by the user. If so, the behavioral semantic text, the user's execution result of the plot of the current link and the initial script are input into the script language model to generate an alternative script or a complete alternative script that the user has not executed to replace the remaining script or the complete initial script of the initial script that the user has not executed.

[0029] Preferably, the initial scripts of the current user and other users are obtained; the behavioral data of the current user executing the plot and the behavioral data of other users executing the plot are obtained, and a comprehensive semantic text describing the meaning of the behavioral data of the current user and the meaning of the behavioral data of other users is parsed and generated; the comprehensive semantic text and the initial scripts of the current user and other users are input into the script language model to generate replacement scripts that have not been executed by the current user and other users respectively, and replace the remaining parts of the initial scripts that have not been executed by the current user and other users respectively.

[0030] As a preferred method, the link plot of the current link currently being executed by the current user and other users is obtained; the behavior data of the current user and other users in executing the link plot of the current link is obtained, and a comprehensive semantic text describing the meaning of the behavior data of the current user and the meaning of the behavior data of other users is generated by parsing; the comprehensive semantic text and the link plot of the current link currently being executed by the current user and other users are input into the script language model to generate the execution results and execution result evaluations of the link plot of the current link by the current user and other users respectively;

[0031] The script language model is used again to determine whether the difference affects the plot of the subsequent links that have not been executed by the current user and other users. If so, the comprehensive semantic text, the execution results of the current user and other users on the plot of the current link, and the initial scripts of the current user and other users are input into the script language model to generate an alternative script or a complete alternative script that has not been executed by the current user and other users, which replaces the remaining script or the complete initial script of the initial script that has not been executed by the current user and other users.

[0032] Preferably, the behavior data of the current user, the behavior data of other users, and the environmental factors in the plot scenes are obtained and parsed to generate a comprehensive semantic text describing the meaning of the behavior data of the current user, the behavior data of other users, and the meaning of the environmental factors in the plot scenes; the comprehensive semantic text and the initial script of the current user and the initial script of other users are input into the script large language model to generate an alternating script that the current user has not executed to replace the initial script that the current user has not executed; the alternating script that the current user has not executed contains story descriptions of other users and environmental factors that the current user is about to see; wherein, if the same other users and environmental factors are respectively used as story descriptions of alternating scripts that have not been executed by different users, then in the alternating scripts that have not been executed by different users, the alternating scripts of different users are matched respectively to form corresponding description content.

[0033] Preferably, the script large language model is used to determine the difference, specifically, to determine whether the difference affects the plot of the subsequent link not executed by the user. If so, the behavioral semantic text, the plot of the current link and the plot of the subsequent link are input into the script large language model to generate a new version of the plot of the subsequent link not executed by the user, replacing the plot of the subsequent link in the initial script not executed by the user.

[0034] Preferably, the script language model is used to determine the differences. Specifically, it is also determined whether the plot of one or more links associated with the current link or the preset result that have not been executed by the user is affected. If so, the behavioral semantic text, the current link plot and the associated one or more links plot are input into the script language model to generate a new version of the associated one or more links plot to replace the associated one or more links plot in the initial script that has not been executed by the user.

[0035] Preferably, the manual background partially adjusts the initial script or the alternative script that has not been executed by the user, and generates an alternative script that has not been executed by the user based on the script large language model corresponding to the content of the adjusted part, wherein the alternative script that has not been executed by the user includes the content of the adjusted part, and the unexecuted part of the plot before the content of the adjusted part, or the unexecuted part of the plot after the content of the adjusted part.

[0036] Preferably, when obtaining an initial script including a multi-stage plot, only the preset results of each stage are obtained; the environmental factors of the subsequent stages are obtained, and the environmental semantic text describing the meaning of the environmental factors is parsed and generated; the environmental semantic text, the preset results of the previous stage, and the preset results of the subsequent stage are input into the script language model to generate the initial stage plot of the subsequent stage.

[0037] Preferably, the behavior data of the user executing the previous link is obtained, and a behavior semantic text describing the meaning of the behavior data is generated by parsing; the behavior semantic text, the preset result of the previous link, and the preset result of the subsequent link are input into the script language model to generate the user's execution result of the previous link and the evaluation of the execution result; the evaluation of the execution result is the difference between the user's execution result of the previous link and the preset result of the previous link;

[0038] The script language model is used again to determine whether the difference affects the preset result of the subsequent link. If so, the behavioral semantic text, the execution result of the previous link and the preset result of the subsequent link are input into the script language model to generate a new version of the link plot of the subsequent link that has not been executed by the user, replacing the initial link plot of the subsequent link that has not been executed by the user; wherein the preset result of the new version of the link plot of the subsequent link is consistent with the preset result of the initial link plot.

[0039] Preferably, the user does not obtain the corresponding link plot in advance, and the corresponding link plot is output to the user in advance through interaction within a set time before the user executes the link plot; when the remaining part of the initial script that has not been executed by the user is replaced by the remaining part of the alternative script, the remaining part of the initial script that has not been executed by the user is updated without the user noticing, thereby completing the dynamic adjustment of the link plot.

[0040] Preferably, when the user is executing the plot of the current link and the new version of the plot of the next link is updated, the guide words for connecting with the new version of the plot of the next link are automatically written based on the user's completion of the plot of the current link. Through narration or the role played by the staff, the user is guided to complete the unfinished part of the plot of the current link and connect it with the new version of the plot of the next link.

[0041] Preferably, an update rule is set to trigger the update of the link plot of the subsequent link not executed by the user. When the user's behavior data and the execution result of the link plot trigger the update rule, the link plot of the subsequent link not executed by the user is adapted to the update rule and modified accordingly.

[0042] Preferably, historical events of the areas corresponding to the plot points, plot points and / or user-to-plot points corresponding to the link plot are identified, wherein the historical events include but are not limited to plot points, plot points and plot points, user-to-plot points, and events occurring to the user at plot points, plot points and plot points and / or user-to-plot points; when the link plot of the subsequent link not executed by the user is related to the behavior data and historical events, based on the influence of the behavior data combined with the historical events on the link plot of the subsequent link not executed by the user, the link plot of the subsequent link not executed by the user is adaptively modified based on the behavior data combined with the historical events, and the link plot of the corresponding subsequent link not executed by the user is updated.

[0043] Preferably, based on the user's authorization, events occurring in the user's daily activities are identified; correspondingly, the historical events also include events occurring in the user's daily activities; when the initial script is related to the historical events, before the user executes the initial script, based on the impact of the historical events on the initial script, the initial script is adaptively modified based on the historical events to generate an alternative script.

[0044] Preferably, when the plot points corresponding to the link plots of the subsequent links not executed by the user, the events between the plot points, and the events occurring between the user and the plot points are emergency events, the link plots of the subsequent links not executed by the user are adaptively modified, and the updated link plots of the subsequent links not executed by the user delete the plot points corresponding to the location of the emergency event, as well as the corresponding part of the link plots.

[0045] The beneficial effects of the present invention are as follows:

[0046] The method for dynamically adjusting a multi-link plot based on a large language model described in the present invention constructs a large language model for generating a script, so as to realize generation of an unexecuted alternating script for each relevant user based on the user's behavior data and environmental factors in the plot scene, and to replace the remaining part of the initial script that the user has not executed. In the present invention, as the user executes each version of the script (including the first execution of the initial script and the execution of the alternating script as the previous version of the script), a new version of the alternating script is continuously generated to replace the initial script and iterate the alternating script, that is, to continuously update the script or the remaining part of the script that the user has not executed.

[0047] In the present invention, the generation of a new version of the alternating script is continuously carried out following the behavior of the current user, the actual completion of the initial script or the current alternating script by the current user, the behavior of other users, and the actual completion of the initial script or the current alternating script by other users. When the manual background adjusts part of the content of the initial script or the current alternating script that has not been executed by the user, the script language model is used to generate a new version of the alternating script that has not been executed by the user based on the content of the adjusted part, wherein the difference between the new version of the alternating script and the current alternating script as the previous version of the script is the generation and replacement of the unexecuted part of the plot before and / or after the content of the adjusted part.

[0048] In the present invention, the scripts of different users can influence each other. Overall, the scripts of a large number of users serve as the plot background for each other (influence each other), and perform cross-script fusion and interaction to form a plot network. With the addition of new users, the withdrawal of original users, and the execution of the scripts of all users, the plot network is automatically advanced. Correspondingly, the unexecuted part of the plot of each user is also generated, and finally the "personalized" script update adapted to each user is realized. The customization scheme of the prior art is only equivalent to the "obtaining the initial script including multiple plots" recorded in the present invention. Under the technical guidance of the prior art, no matter how detailed the screening conditions are set, it is impossible to achieve the "personalized" script update adapted to each user as described in the present invention.

[0049] In the present invention, the script language model is first used to compare and evaluate the execution results of the user on the link plot, and then based on the evaluation, it is determined whether to generate a new version of the plot, which can take into account both execution efficiency and meeting the needs of "personalization".

[0050] In the present invention, after the user generates the initial script, the user does not obtain the corresponding link plot in advance; within the set time before the user executes the link plot, the corresponding link plot is output to the user in advance through interaction; then, without the user's awareness, the corresponding link plot that has not been executed by the user is updated, and the real-time update of the unexecuted part of the initial script or the alternative script is completed; further, the present invention also realizes the update of the corresponding link plot that has not been executed by the user without the user's awareness, and the real-time update of the unexecuted part of the initial script or the alternative script is completed; on the other hand, when an emergency event occurs at a plot point, through dynamic customization and narration, that is, combined with the deletion and guidance of the script, it is ensured that the user can avoid the emergency event, ensure the user's safety, and ensure a smooth trip.

[0051] The present invention obtains data such as historical events and user-related data of all users before and during travel, which is equivalent to knowing what the user has encountered in the past, what will be encountered in the future, what happened in the past and what will happen in the future in the area covered by the initial script, the user's real-time status and other information, and uses this as the basis for modifying the initial script or the alternate script that the user has not executed. On the one hand, the present invention realizes that the user's initial script or alternate script (combining the deletion and modification of the link plot and the generation of the guiding plot, including writing narration, or narration by the role played by the staff) can ensure that the user is guided to respond to and avoid emergency events; on the other hand, it realizes updating the corresponding initial script or alternate script that the user has not executed without the user's awareness. DETAILED DESCRIPTION

[0052] The present invention is further described in detail below with reference to the embodiments.

[0053] In order to solve the shortcomings of the prior art, such as the rough customization conditions, the insufficient personalization of the scheme, and the inability to make real-time modifications based on the actual situation after the trip, the present invention provides a dynamic adjustment method for multi-link plots based on a large language model. The present invention constructs a large language model for generating scripts and plots, based on the description of the behavioral data and global factors of all users (such as environmental factors in the plot scene, including but not limited to the real-time status of facilities and items in the actual scene and virtual scene, events that occur in real time, etc.), and uses the large language model of the script to continuously generate the alternating scripts adapted to each user. At the same time, the scripts that affect each other are the background of the plot, and the "personalized" requirements of each user for the script are met in real time. Based on the continuous iterative update of the script, the present invention can also guide users to optimally handle various matters needing attention in the entertainment process in real time through the "personalized" alternating script, that is, users execute real-time alternating scripts, and can naturally and imperceptibly perform optimal processing on various matters needing attention, such as avoiding emergency events, handling emergency events, and assisting in the processing of non-emergency matters needing attention. The present invention is applicable to the iterative update of the scripts of online entertainment products and offline entertainment products. The basis for generating new scripts is the behavioral data and global factors of all users. The behavioral data and global factors of all users are converted into the input of the script large language model, and the output of the script large language model is the generated alternating script. Correspondingly, when the present invention is implemented in online entertainment products, by obtaining various parameters of each object, including the behavioral data of all virtual characters and the global factors of all objects of the virtual scene, and converting them into inputs that meet the script large language model, the script large language model is used to generate alternating scripts, and the virtual characters are guided to execute alternating scripts; when the present invention is implemented in offline entertainment products, the behavioral data of all real users in the monitoring area and the global factors in the actual scene are obtained, and converted into inputs that meet the script large language model, the script large language model is used to generate alternating scripts, and the users are guided to execute alternating scripts. In the present invention, the user includes a virtual character and a real user. When it is implemented, according to the implementation in online entertainment products and offline entertainment products, the user is determined to be a virtual character or a real user, the plot scene is a virtual scene or an actual scene, the environmental factor is a virtual environmental factor or an actual environmental factor, the event is a virtual event or an actual event, and the same applies to others.

[0054] The dynamic adjustment method of multi-link plot based on large language model described in the present invention, constructs script large language model, is used to generate script, when specific implementation, script is composed of each link plot, script large language model can also be by generating part of link plot, and then generate alternating script. In the present invention, for each user, the corresponding initial script is obtained in advance, and the initial script includes multi-link plot. When the user executes the initial script for the first time, that is, executes the link plot corresponding to the online entertainment product or offline entertainment product online or offline, obtains the behavior data of the user executing the link plot, and converts the behavior data into the input that satisfies the script large language model, that is, parses the behavior data, and then generates the behavior semantic text describing the meaning of the behavior data. The behavior semantic text and the initial script are input into the script large language model, and the script large language model generates the alternating script that the user has not executed based on the input behavior semantic text and the initial script, replaces the remaining part of the initial script that the user has not executed, and completes a dynamic adjustment of the initial script. The present invention is based on what the user "did" (i.e., behavior data) in the process of executing the link plot, combined with what should be "done" (i.e., the initial script), and automatically generates what the user needs to "do" next (i.e., the alternating script). In the present invention, as the user executes each version of the script (including the first execution of the initial script and the execution of the altered script as the previous version of the script), a new version of the altered script is continuously generated to replace the initial script and iterate the altered script, that is, the script or the remaining part of the script that has not been executed by the user is continuously updated to complete the continuous and dynamic adjustment of the altered script.

[0055] Since the alternating script is an adaptive modification of the script language model based on the initial script, the alternating script is to replace some of the initial script's link plots when the user's behavior data in executing the link plots has a certain degree of influence on the link plots that have not been executed in the initial script. When the user's state after executing several link plots cannot achieve the optimal match with the link plots that have not been executed in the initial script, the generated alternating script maintains the optimal match with the user's real-time state. Therefore, the alternating script is usually less different from the initial script; but as the user executes the link plots, the alternating script is continuously iterated, and even if the generated updated script is significantly different from the initial script, it can still maintain the optimal match with the user's real-time state. However, the difference between the alternating script and the initial script is not necessarily positively correlated with the number of iterations. The link plots contained in the alternating script are generated based on the user's completion of the link plots and the user's behavior data. The link plots contained in the alternating script are related to the user's completion of the link plots and the user's behavior data, and are not related to the number of iterations.

[0056] In the present invention, generating an alternative script is to adaptively modify some of the plots in the initial script or the previous version of the alternative script. The adaptive modification is specifically to rewrite part of the content based on the unexecuted plots in the initial script corresponding to the user, based on the plots in the segments that the user has not executed, for the user's behavior data at the plot points corresponding to the segments, between plot points and / or between the user and the plot points, the user's completion of the current segment plot, emergency events, the location of the occurrence of emergency events, and the corresponding processing solutions, to obtain an alternative script that has been partially rewritten and not executed by the user, so that in the alternative script that has been partially rewritten and not executed by the user, for the user's behavior data, the completion of the current segment plot, emergency events, the occurrence of emergency events, and the corresponding processing solutions Position, corresponding processing scheme, there is adapted plot content; or, based on the user's behavior data at the plot point, between the plot point and the plot point, and / or between the user and the plot point, the unexecuted link plot in the user's corresponding initial script, the emergency event, the location of the emergency event, and the corresponding processing scheme, rewrite all the unexecuted link plots in the user's corresponding initial script, and obtain all the rewritten user-unexecuted alternating scripts, so that in all the rewritten user-unexecuted alternating scripts, there is adapted plot content for the user's behavior data, the completion status of the current link plot, the emergency event, the location of the emergency event, and the corresponding processing scheme. In the present invention, part of the content is rewritten based on the link plot that the user has not executed, or all the link plots that the user has not executed are rewritten with reference to the link plot that the user has not executed. The corresponding user-unexecuted alternating scripts are all preset plot contents that do not belong to the script database. The partial content rewriting and the full rewriting do not directly quote, intercept, or apply the preset plot content in the script database, that is, they are newly generated plot contents. In special cases, there are alternate scripts that have not been executed by the user. For the user's behavior data, the completion status of the current link plot, emergency events, the location of the emergency events, and the corresponding treatment plan, there is a possibility that the adapted plot content is the same or similar to the preset plot content in the script database, but in fact, the newly generated corresponding treatment plan's plot content is not directly selected from the script database to quote, intercept, and apply the preset plot content. In the present invention, adaptive modifications are made to the link plots that have not been executed by the user, and when generating the alternate scripts that have not been executed by the user, the preset plot content in the script database is not searched, matched, and other operations are not performed. In specific implementation, the preset plot content in the script database is only used to generate the initial script and as a sample for training the large language model of the plot, and does not directly participate in the generation of the alternate script.

[0057] In specific implementation, the behavioral data includes but is not limited to the user's location information, posture data, action data, language data, and gaze point; collecting as much behavioral data of the user as possible is more conducive to accurately judging the behavioral results corresponding to the user's behavioral data, and then, it is conducive to more accurately judging the user's execution results of the link plot, and finally, improving the adaptability of modifying the unexecuted link plot.

[0058] Based on this, the unexecuted alternate scripts mentioned above are specifically based on or refer to the unexecuted segments in the initial script corresponding to the user, partially rewrite the content, completely rewrite the content of the segments that the user has not executed, generate the unexecuted alternate scripts, and replace the remaining parts of the initial scripts that the user has not executed. The same is true for the iterative updates of the alternate scripts.

[0059] Furthermore, the present invention also considers the influence of each user's behavior data on the unexecuted segments of the plot by other users, and generates related users (the related users include the current user and other related users that can be affected by the user's behavior data) based on the behavior data of the current user and other users. When the initial script or the alternating script for a certain user is dynamically customized, the user himself is the current user. Furthermore, when the initial script or the alternating script for all users is dynamically customized at the same time, all users are each other's background, just like everyone in the real world is carrying out daily activities at the same time, and some people are in a different situation. Historical activities will have an impact in the future. When the present invention is implemented in online entertainment products, such as offline travel, offline games, etc., it can naturally narrow the scope of users' activities and behaviors, making it easier to judge the impact. When the present invention is implemented in online entertainment products, such as online games, stand-alone games, etc., compared with offline entertainment products, online entertainment products are more likely to judge the impact based on software programs) A new version of the unexecuted alternating script is used to replace the initial script or the previous version of the alternating script that the user has not executed; wherein, the difference between the new version of the alternating script and the initial script or the previous version of the alternating script is the adaptive modification made to the initial script or the alternating script. In the present invention, the adaptive modification of the link plot is continuously carried out following the behavior of the current user, the actual completion of the link plot by the current user, the behavior of other users, and the actual completion of the link plot by other users.

[0060] Similar to generating an alternation script for the current user based only on the behavior data of the current user, generating an alternation script for each user related to the current user and other users based on the behavior data of the current user and other users, specifically: obtaining the initial scripts of the current user and other users; obtaining the behavior data of the current user executing the plot and the behavior data of other users executing the plot, and parsing to generate a comprehensive semantic text describing the meaning of the behavior data of the current user and the meaning of the behavior data of other users; inputting the comprehensive semantic text and the initial scripts of the current user and other users into the script language model, generating alternation scripts that have not been executed by the current user and other users respectively, and replacing the remaining parts of the initial scripts that have not been executed by the current user and other users respectively.

[0061] Furthermore, the present invention also considers that environmental factors in the plot scene (including online and offline facilities, the real-time status of items, real-time events, etc.) and the behavior data of other users may affect the behavior data of the current user, thereby affecting the execution result of the current user on the link plot, and finally affecting the update iteration of the unexecuted link plot. In this embodiment, the environmental factors in the plot scene are also detected, and the environmental factors, the behavior data of the current user and the behavior data of other users are combined. If the link plot that the current user and other users have not executed is related to the environmental factors, the behavior data of the current user and the behavior data of other users, based on the influence of the environmental factors, the behavior data of the current user and the behavior data of other users on the link plot that the current user and other users have not executed, the link plot that the current user and other users have not executed is adaptively modified based on the behavior data, and the corresponding link plot that the current user and other users have not executed is updated. Wherein, the environmental factors include but are not limited to the real-time status of the plot points corresponding to the link plot, the equipment and items between the plot points and / or between the user and the plot point; for example, whether the equipment is busy, whether other users cause the change of the working state of the equipment, whether the item is damaged, and so on. Therefore, the present invention can not only make adaptive modifications to the environmental factors in the plot scene corresponding to the unexecuted part of the link plot, but also guide the user to execute the link plot more smoothly with higher time utilization through the modification of the link plot; moreover, the initial scripts or alternating scripts of different users can form mutual influence. From the overall point of view, the initial scripts or alternating scripts of a large number of users are the plot backgrounds of each other (influence each other), and cross-script fusion interaction is carried out to form a plot network together. With the addition of new users, the withdrawal of original users, and the execution of the initial scripts or alternating scripts of all users, the plot network is automatically advanced. Correspondingly, the unexecuted part of the link plot of each user is also adaptively modified, and finally the diversified "personalized" dynamic customization is realized. The customization scheme of the prior art is only equivalent to the "obtaining the initial script including the multi-link plot" recorded in the present invention. Under the technical guidance of the prior art, no matter how detailed the screening conditions are set, the diversified "personalized" dynamic customization described in the present invention cannot be realized.

[0062] When considering the impact of the current user's behavior data, other users' behavior data, and environmental factors on the plot of all relevant users that have not been executed, specifically, obtain and parse the current user's behavior data, other users' behavior data, and environmental factors in the plot scene, generate a comprehensive semantic text that describes the meaning of the current user's behavior data, the meaning of other users' behavior data, and the meaning of environmental factors in the plot scene, input the comprehensive semantic text and the current user's initial script and other users' initial script into the script large language model, generate an alternation script that the current user has not executed, and replace the initial script that the current user has not executed; the alternation script that the current user has not executed contains story descriptions of other users and environmental factors that the current user is about to see; wherein, if the same other users and environmental factors are respectively used as story descriptions of alternation scripts that have not been executed by different users, then in the alternation scripts that have not been executed by different users, the alternation scripts of different users are matched respectively to form corresponding description content.

[0063] In order to take into account efficiency, adaptability and "personalization", the present invention determines whether to generate an alternative script by judging the execution result of the user on the link plot, and then generates an alternative script when it is judged that there is a necessity to generate an alternative script. Among them, the script large language model is used to perform the judgment of the two stages. Specifically, the link plot of the current link currently executed by the user (in this embodiment, only the current user is considered); the behavior data of the user executing the link plot of the current link is obtained, and the behavior semantic text describing the meaning of the behavior data is parsed and generated. The behavior semantic text and the link plot of the current link are input into the script large language model to generate the user's execution result and execution result evaluation of the link plot of the current link; wherein, the execution result evaluation is the difference between the user's execution result of the link plot of the current link and the preset result corresponding to the link plot of the current link. At this point, the script large language model is used to output the execution result and the execution result evaluation.

[0064] Further, the script language model is used again to determine whether the difference has an impact on the plot of the subsequent link that the user has not executed. If so, the behavior semantic text, the user's execution result of the link plot of the current link, and the initial script are input into the script language model to generate an alternate script or a complete alternate script that the user has not executed, replacing the remaining script or the complete initial script of the initial script that the user has not executed. Since the user executes the link plot of the current link, the link plot of the next link is usually most directly affected. In this embodiment, the script language model is used to determine the difference. Specifically, it is determined whether the difference has an impact on the link plot of the next link that the user has not executed. If so, the behavior semantic text, the link plot of the current link, and the link plot of the next link are input into the script language model to generate a new version of the link plot of the next link that the user has not executed, replacing the link plot of the next link in the initial script that the user has not executed.

[0065] Furthermore, the present invention also considers the impact of the user's execution of the link plot of the current link on the link plots of other links except the link plot of the subsequent link (i.e., the link plots of one or more subsequent links associated with the preset result of the current link). In this embodiment, the script large language model is used to determine the difference. Specifically, it is also determined whether it affects the link plot of one or more links associated with the link plot of the current link or the preset result that the user has not executed. If so, the behavioral semantic text, the link plot of the current link and the link plot of the associated one or more links are input into the script large language model to generate a new version of the link plot of the associated one or more links to replace the link plot of the associated one or more links in the initial script that the user has not executed.

[0066] Among them, the script large language model is used to determine whether the difference has an impact on the plot of the subsequent link not executed by the user. Specifically, a temporary script is generated based on the difference, and the script large language model is used to generate the plot of the latter link in the temporary script and the temporary execution result based on the temporary script, and the plot difference between the plot of the latter link in the temporary script and the plot of the latter link not executed by the user is generated, and the result difference between the temporary execution result and the preset result of the latter link not executed by the user is generated; if the plot difference and the result difference both meet expectations, that is, the conclusion of the plot difference and the result difference output by the script large language model is that no iteration is required, then it is determined that the difference has no impact on the plot of the subsequent link not executed by the user; otherwise, it is determined that there is an impact, that is, the conclusion of one or both of the plot difference or the result difference is that iteration is required, then an iterative script is generated.

[0067] Further, in response to the situation where users influence the unexecuted segments of other users based on their own behavior data, the present invention determines whether to generate an alternating script for each user for the current user and other users by judging the execution results of the current user and other users on their own segments. When it is determined that there is a necessity to generate an alternating script, an alternating script is generated. Among them, the script large language model is used to perform the judgment of the two stages. Similar to the aforementioned embodiment that only considers the current user, in this embodiment, specifically, the segment plot of the current segment currently executed by the current user and other users is obtained; the behavior data of the current user and other users executing the segment plot of the current segment is obtained, and a comprehensive semantic text describing the meaning of the behavior data of the current user and the behavior data of other users is parsed and generated; the comprehensive semantic text and the segment plot of the current segment currently executed by the current user and other users are input into the script large language model to generate the execution results and execution result evaluation of the segment plot of the current segment by the current user and other users.

[0068] Furthermore, the script language model is used again to determine whether the difference affects the plot of the subsequent links that have not been executed by the current user and other users. If so, the comprehensive semantic text, the execution results of the current user and other users on the plot of the current link, and the initial scripts of the current user and other users are input into the script language model to generate an alternative script or a complete alternative script that has not been executed by the current user and other users, replacing the remaining script or the complete initial script of the initial script that has not been executed by the current user and other users.

[0069] It can be seen that in the present invention, only the script language model can be used to complete the judgment, generation and other operations, which greatly improves the execution efficiency and reduces the execution cost. At the same time, the script language model can be used repeatedly. Each time the script language model is used, the input and output can be used as training samples to be screened to optimize the script language model. Then, the script language model can be set as a self-optimization optimization mechanism according to the implementation requirements and when the computing power allows.

[0070] Based on the script large language model, the present invention can also adjust key nodes through the manual background, and then use the script large language model to perform global automatic modification, which can be used as one of the auxiliary means. Specifically, the manual background partially adjusts the initial script or the replacement script that has not been executed by the user, and based on the script large language model, generates the replacement script that has not been executed by the user corresponding to the content of the adjustment part, and the replacement script that has not been executed by the user includes the content of the adjustment part, and the unexecuted part of the plot before the content of the adjustment part, or the unexecuted part of the plot after the content of the adjustment part. Among them, the difference between the new version of the replacement script and the current replacement script as the previous version of the script is the generation and replacement of the unexecuted part of the plot before and / or after the content of the adjustment part.

[0071] The script big language model can also be used to modify the initial script or the iterative script that has been acquired but not yet executed. That is, after the initial script or the iterative script is generated, as the new users increase, the initial script or the iterative script will increase accordingly. The initial script or the iterative script generated previously serves as the background of the initial script or the iterative script generated later and is reflected in the initial script or the iterative script generated later. The increase of the initial script or the iterative script generated later also affects the initial script or the iterative script generated previously. In this way, the script big language model is used to adaptively modify the initial script or the iterative script generated previously, and automatically write it to generate a new version of the iterative script to replace the initial script or the iterative script generated previously. Based on this, all previously generated initial scripts or iterative scripts are continuously updated until the corresponding users start to execute the link plots, or the later generated initial scripts or iterative scripts do not affect the previously generated initial scripts or iterative scripts (the previously generated initial scripts or iterative scripts remain updatable, and when the later generated initial scripts or iterative scripts affect the previously generated initial scripts or iterative scripts, the previously generated initial scripts or iterative scripts are iterated again).

[0072] The script language model can also be used to automatically write the guiding words that connect the previous and next segments of the plot. Through narration or the roles played by the staff, it can guide the user to complete the unfinished part of the current segment and connect it with the next segment.

[0073] As another auxiliary means for dynamically customizing the unexecuted part of the initial script, in the present invention, a more explicit trigger mechanism can be set, specifically, an update rule for triggering the update of the link plot of the subsequent link that the user has not executed is set, and when the user's behavior data and the execution result of the link plot trigger the update rule, the link plot of the subsequent link that the user has not executed is adapted to the update rule and modified accordingly. The update rule can be implemented as a set execution requirement. After the trigger, the link plot of the unexecuted subsequent link is adaptively modified according to the acquired execution requirement, and a new version of the link plot is automatically generated. The present invention is different from the script of the prior art in which multiple parallel branch plots are pre-set. In the initial script or the replacement script of the present invention, parallel branch link plots are not pre-set. If the link plot that the user has not executed is not modified, the corresponding link plot is still executed according to the initial script or the replacement script; and when the link plot that the user has not executed is modified and a new version of the replacement script is generated, the user executes the corresponding link plot according to the new version of the replacement script. In the whole process, the initial script or the replacement script does not set parallel branch link plots, but a new version of the link plot is automatically generated by the script large language model.

[0074] After automatically generating a new version of the plot and guide words based on the script language model, obtain the behavioral data of the user executing the new version of the plot and the guidance of the guide words, and parse the behavioral data to generate semantic text describing the meaning of the behavioral data as a training sample for the optimization training of the script language model.

[0075] The script language model of the present invention is also used to generate an initial script. Specifically, when obtaining an initial script including a multi-link plot, first, only the preset result of each link is obtained; then, the environmental factors of the subsequent link (including the subsequent link when each link is the current link, or the subsequent one or more links associated with the current link) are obtained, and the environmental semantic text describing the meaning of the environmental factors is parsed and generated; finally, the environmental semantic text, the preset result of the previous link, and the preset result of the subsequent link are input into the script language model to generate the initial link plot of the subsequent link. That is, in this embodiment, based on the preset result of each link and the environmental factors of each link, the link plot of each link is generated backwards, and then, it is no longer necessary to set the script database as the basic option, and the corresponding generated initial script can be based on the user's user association data (when implemented in offline entertainment products, the user association data includes but is not limited to clothing, decorations, children or pets carried, supplies carried by travel, and props carried for executing link plots) to generate a "personalized" initial script adapted to the user.

[0076] Furthermore, the present invention can also be implemented as follows: when obtaining the initial script, only the plot of the first link is generated first, and then based on the behavior data of the user executing the plot of the first link and the execution result of the plot of the first link, the plot of the next link or several subsequent links is generated, and so on. When the plot of each link is generated link by link, the plot of the next link is an alternating script adapted to the user when it is generated. When the link plots of several subsequent links are generated in advance, as the user executes the link plots step by step, the link plots of several subsequent links that have been generated are continuously iterated, and the link plots of several subsequent links are added as the initial script; for example, before the user executes the link plot of the first link, the link plots of N links are pre-generated as the initial script (at this time, the initial script does not include the link plots corresponding to the preset results of all links). After the user executes the link plot of the first link, a new version of the link plots from the second to the Nth links is generated as the replacement script. At the same time, the corresponding link plots from the N+1th to the N+nth links are newly generated as the initial script. At this time, in the generated script, the link plots of the Nth link and before are the replacement scripts, and the link plots of the N+1th link and thereafter are the initial scripts. In the present invention, the segment plots of the initial script and the alternating script are generated based on the script big language model. Specifically, the behavior data of the user executing the previous segment is obtained, and a behavior semantic text describing the meaning of the behavior data is parsed and generated; the behavior semantic text, the preset result of the previous segment, and the preset result of the subsequent segment are input into the script big language model to generate the user's execution result of the previous segment and the evaluation of the execution result; the execution result evaluation is the difference between the user's execution result of the previous segment and the preset result of the previous segment.

[0077] Further, the script language model is used again to determine whether the difference affects the preset result of the subsequent link. If so, the behavior semantic text, the execution result of the previous link and the preset result of the subsequent link are input into the script language model to generate a new version of the link plot of the subsequent link that the user has not executed, and replace the initial link plot of the subsequent link that the user has not executed. Among them, priority is given to ensuring that the preset result of the new version of the link plot of the subsequent link is consistent with the preset result of the initial link plot, that is, the link plot of the iterative link does not change the preset result of the link; until as the user executes the link plot, based on the behavior data of the current user and other users and environmental factors, it is determined that the preset result corresponding to the first version of the link plot of the subsequent link generated in real time cannot be consistent with the preset result corresponding to the initial link plot, then the preset result of the corresponding link is iterated to the preset result corresponding to the first version of the link plot.

[0078] One of the purposes of the present invention is to achieve that when the plot of the link that the user has not executed is updated, it will not be perceived or noticed by the user, and to avoid the user being spoiled first and then having the bad experience of broken expectations. In the present invention, after obtaining the initial script or the preset result, the user does not obtain the corresponding link plot in advance; within the set time before the user executes the link plot, the corresponding link plot is output to the user in advance through interaction; then, when the remaining part of the initial script that the user has not executed is replaced by the remaining part of the alternative script, the remaining part of the initial script that the user has not executed is updated without the user noticing, thereby completing the dynamic adjustment of the link plot.

[0079] When the user is executing the plot of the current link, and based on the behavior data that has occurred and the user's completion of the current link plot, the next link plot can be determined. If the user is still executing the current link plot and the new version of the link plot of the next link is updated, then based on the user's completion of the link plot of the current link, guide words that connect with the new version of the link plot of the next link are automatically written, and the user is guided to complete the unfinished part of the link plot of the current link through narration or the role played by the staff, and connect with the new version of the link plot of the next link. In fact, using guide words to guide the user to complete the unfinished part of the current link plot is also regarded as the dynamic customization of the unexecuted part of the current link plot that the user is executing by the present invention, such as skipping certain plots of the unexecuted part of the current link plot, or guiding the user to execute certain plots different from the unexecuted part of the current link plot.

[0080] The present invention also obtains the behavioral data of all users during the trip, the environmental factors in the plot scene, and the historical events within the monitoring range (including the events that occurred during the user's trip, and the events that occurred in the daily activities before the trip), which is equivalent to knowing what the user did in the past, what he encountered in the past, what he will encounter in the future, what happened in the plot scene in the past, and what will happen in the plot scene in the future, and uses this as the basis for modifying the plot of the link that the user has not executed, and automatically writes a narration explanation, or narrates the role played by the staff to guide the user to complete the unfinished part of the current link plot and connect with the next link plot. On the one hand, the present invention realizes real-time "recommendation" and "how to play" to users; on the other hand, it realizes updating the corresponding user's unexecuted link plot without the user's feeling, and completing the dynamic customization of the unexecuted part of the initial script or the alternate script; on the other hand, when an emergency event occurs at the plot point corresponding to the link plot in the offline entertainment product, through dynamic customization and narration, that is, combined with the deletion and guidance of the initial script or the alternate script, it ensures that the user can avoid the emergency event, ensure the user's safety, and travel smoothly.

[0081] In specific implementation, the present invention deploys an event recognition system, the monitoring range of which covers plot points, areas between plot points and / or areas between users and plot points, and is used to perform event recognition in the areas. Specifically, historical events corresponding to plot points, areas between plot points and / or areas between users and plot points corresponding to link plots are identified, and the historical events include but are not limited to plot points, areas between plot points and plot points, areas between users and plot points within the monitoring range, and events that occur to users at plot points, areas between plot points and plot points, and / or areas between users and plot points; when the link plot that the user has not executed is related to behavioral data and historical events, based on the influence of behavioral data combined with historical events on the link plot that the user has not executed, the link plot that the user has not executed is adaptively modified based on behavioral data combined with historical events, and the corresponding link plot that the user has not executed is updated to generate an alternating script.

[0082] Furthermore, the monitoring scope of the event recognition system also covers the area corresponding to the user's daily activities, and based on the user's authorization, identifies events occurring in the user's daily activities; correspondingly, the historical events also include events occurring in the user's daily activities; when the initial script or the altered script is related to historical events, before the user executes the initial script or the altered script, based on the impact of the historical events on the initial script or the altered script, the initial script or the altered script is adaptively modified based on the historical events to generate a new version of the altered script.

[0083] Based on this, for the current user, the identified historical events are what the current user did and encountered in the past; at the same time, the historical events of other users and what happened in the plot scene in the past are also identified. If the initial script or the altered script of other users will intersect with the current user in the future and serve as each other's background, the historical events of other users are what the current user will encounter in the future; for the events corresponding to the plot scene in which the current user and other users will interact in the future, they are what will happen in the plot scene in the future.

[0084] The present invention can also modify the link plot of the subsequent link that the user has not executed to adapt to the emergency event, so that when the user executes the link plot of the subsequent link, the user naturally avoids the place where the emergency event occurs; or guides the user to avoid the place where the emergency event occurs by automatically writing guiding words. Specifically, when the plot point corresponding to the link plot of the subsequent link that the user has not executed, the event between the plot point and the plot point, and the event between the user and the plot point are emergency events, when the link plot of the subsequent link that the user has not executed is adaptively modified, the updated link plot that the user has not executed deletes the plot point corresponding to the location where the emergency event is located, as well as the corresponding part of the link plot. Alternatively, based on the emergency event, automatically write guiding words that connect with the next link plot, and guide the user to avoid the emergency event through narration or the role played by the staff.

[0085] The above embodiments are only used to illustrate the present invention, and are not used to limit the present invention. As long as they are based on the technical essence of the present invention, changes and modifications to the above embodiments will fall within the scope of the claims of the present invention.

Claims

1. A method for dynamically adjusting a multi-link plot based on a large language model, characterized in that: Build a large script language model; obtain the initial script including multiple plot segments; The behavioral data of the user's execution of the plot is obtained, and the behavioral semantic text that describes the meaning of the behavioral data is parsed and generated; the behavioral semantic text and the initial script are input into the script language model to generate an alternative script that the user has not executed, replacing the remaining script of the initial script that the user has not executed.

2. The method for dynamically adjusting a multi-link plot based on a large language model according to claim 1, characterized in that: Obtain the link plot of the current link currently executed by the user; obtain the behavior data of the user executing the link plot of the current link, and parse and generate a behavior semantic text describing the meaning of the behavior data; input the behavior semantic text and the link plot of the current link into the script language model to generate the user's execution result of the link plot of the current link and the evaluation of the execution result; the execution result evaluation is the difference between the user's execution result of the link plot of the current link and the preset result corresponding to the link plot of the current link; The script language model is used again to determine whether the difference affects the plot of the subsequent link not executed by the user. If so, the behavioral semantic text, the user's execution result of the plot of the current link and the initial script are input into the script language model to generate an alternative script or a complete alternative script that the user has not executed to replace the remaining script or the complete initial script of the initial script that the user has not executed.

3. The method for dynamically adjusting a multi-link plot based on a large language model according to claim 1, characterized in that: Obtain the initial scripts of the current user and other users; obtain the behavior data of the current user executing the plot of the link and the behavior data of other users executing the plot of the link, and parse and generate a comprehensive semantic text describing the meaning of the behavior data of the current user and the meaning of the behavior data of other users; The comprehensive semantic text and the initial scripts of the current user and other users are input into the script language model to generate the replacement scripts that have not been executed by the current user and other users, and replace the remaining scripts of the initial scripts that have not been executed by the current user and other users.

4. The method for dynamically adjusting a multi-link plot based on a large language model according to claim 3, characterized in that: Obtain the plot of the current link currently being executed by the current user and other users; obtain the behavior data of the current user and other users in executing the plot of the current link, and parse and generate a comprehensive semantic text describing the meaning of the behavior data of the current user and the meaning of the behavior data of other users; Input the comprehensive semantic text and the current link plot currently executed by the current user and other users into the script language model to generate the execution results and execution result evaluations of the current link plot by the current user and other users respectively; The script language model is used again to determine whether the difference affects the plot of the subsequent links that have not been executed by the current user and other users. If so, the comprehensive semantic text, the execution results of the current user and other users on the plot of the current link, and the initial scripts of the current user and other users are input into the script language model to generate an alternative script or a complete alternative script that has not been executed by the current user and other users, which replaces the remaining script or the complete initial script of the initial script that has not been executed by the current user and other users.

5. The method for dynamically adjusting a multi-link plot based on a large language model according to claim 3, characterized in that: Obtain and parse the behavior data of the current user, the behavior data of other users, and the environmental factors in the plot scene, generate a comprehensive semantic text that describes the meaning of the behavior data of the current user, the meaning of the behavior data of other users, and the meaning of the environmental factors in the plot scene, input the comprehensive semantic text and the initial script of the current user and the initial script of other users into the script large language model, generate an alternating script that the current user has not executed, and replace the initial script that the current user has not executed; the alternating script that the current user has not executed contains the story description of other users and environmental factors that the current user is about to see; wherein, if the same other users and environmental factors are respectively used as story descriptions of alternating scripts that have not been executed by different users, then in the alternating scripts that have not been executed by different users, the alternating scripts of different users are matched respectively to form corresponding description content.

6. The method for dynamically adjusting a multi-link plot based on a large language model according to claim 2 or 4, characterized in that: The script big language model is used to determine the differences. Specifically, whether the differences affect the plot of the subsequent link that has not been executed by the user. If so, the behavioral semantic text, the plot of the current link, and the plot of the subsequent link are input into the script big language model to generate a new version of the plot of the subsequent link that has not been executed by the user, replacing the plot of the subsequent link in the initial script that has not been executed by the user.

7. The method for dynamically adjusting a multi-link plot based on a large language model according to claim 6, characterized in that: The script language model is used to determine the differences. Specifically, it is also determined whether the plot of one or more links associated with the current link or the preset result that have not been executed by the user is affected. If so, the behavioral semantic text, the current link plot and the associated one or more links plot are input into the script language model to generate a new version of the associated one or more links plot to replace the associated one or more links plot in the initial script that has not been executed by the user.

8. The method for dynamically adjusting a multi-link plot based on a large language model according to claim 1, characterized in that: The manual backend partially adjusts the initial script or the replacement script that has not been executed by the user, and based on the script large language model, generates an replacement script that has not been executed by the user corresponding to the content of the adjusted part. The replacement script that has not been executed by the user includes the content of the adjusted part, and the unexecuted part of the plot before the content of the adjusted part, or the unexecuted part of the plot after the content of the adjusted part.

9. The method for dynamically adjusting a multi-link plot based on a large language model according to any one of claims 1 to 5, characterized in that: When obtaining an initial script including a multi-link plot, only the preset results of each link are obtained; the environmental factors of the subsequent link are obtained, and the environmental semantic text describing the meaning of the environmental factors is parsed and generated; the environmental semantic text, the preset results of the previous link, and the preset results of the subsequent link are input into the script language model to generate the initial link plot of the subsequent link.

10. The method for dynamically adjusting a multi-link plot based on a large language model according to claim 9, characterized in that: Obtain the behavior data of the user executing the previous link, and parse and generate a behavior semantic text describing the meaning of the behavior data; input the behavior semantic text, the preset result of the previous link, and the preset result of the next link into the script language model to generate the user's execution result of the previous link and the evaluation of the execution result; The execution result evaluation is the difference between the user's execution result of the previous link and the preset result of the previous link; The script language model is used again to determine whether the difference affects the preset result of the subsequent link. If so, the behavioral semantic text, the execution result of the previous link and the preset result of the subsequent link are input into the script language model to generate a new version of the link plot of the subsequent link that has not been executed by the user, replacing the initial link plot of the subsequent link that has not been executed by the user; wherein the preset result of the new version of the link plot of the subsequent link is consistent with the preset result of the initial link plot.

11. The method for dynamically adjusting a multi-link plot based on a large language model according to claim 1, characterized in that: The user does not obtain the corresponding link plot in advance. Within the set time before the user executes the link plot, the corresponding link plot is output to the user in advance through interaction; when the remaining part of the initial script that has not been executed by the user is replaced by the remaining part of the alternative script, the remaining part of the initial script that has not been executed by the user is updated without the user noticing, thereby completing the dynamic adjustment of the link plot.

12. The method for dynamically adjusting a multi-link plot based on a large language model according to claim 11, characterized in that: When the user is executing the plot of the current link and the new version of the plot of the next link is updated, the guiding words for connecting with the new version of the plot of the next link are automatically written based on the user's completion of the plot of the current link. Through narration or the role played by the staff, the user is guided to complete the unfinished part of the plot of the current link and connect it with the new version of the plot of the next link.

13. The method for dynamically adjusting a multi-link plot based on a large language model according to claim 1, characterized in that: Set the update rules that trigger the update of the link plots of the subsequent links that the user has not executed. When the user's behavior data and the execution results of the link plots trigger the update rules, the link plots of the subsequent links that the user has not executed will be adapted to the update rules and modified accordingly.

14. The method for dynamically adjusting a multi-link plot based on a large language model according to claim 1, characterized in that: Identify historical events of plot points corresponding to the segment plot, areas between plot points and / or between a user and a plot point, wherein the historical events include but are not limited to plot points, areas between plot points and plot points, between a user and a plot point, and events that occur to a user at a plot point, between plot points and plot points, and / or between a user and a plot point; When the plot of the subsequent link not executed by the user is related to the behavioral data and historical events, based on the influence of the behavioral data combined with the historical events on the plot of the subsequent link not executed by the user, the plot of the subsequent link not executed by the user is adaptively modified based on the behavioral data combined with the historical events, and the corresponding plot of the subsequent link not executed by the user is updated.

15. The method for dynamically adjusting a multi-link plot based on a large language model according to claim 14, characterized in that: Based on the user's authorization, events occurring in the user's daily activities are identified; correspondingly, the historical events also include events occurring in the user's daily activities; when the initial script is related to historical events, before the user executes the initial script, based on the impact of historical events on the initial script, the initial script is adaptively modified based on the historical events to generate an alternative script.

16. The method for dynamically adjusting a multi-link plot based on a large language model according to claim 14, characterized in that: When the plot points corresponding to the link plots of the subsequent links not executed by the user, the events between the plot points, and the events between the user and the plot points are emergency events, the link plots of the subsequent links not executed by the user are adaptively modified, and the updated link plots of the subsequent links not executed by the user delete the plot points corresponding to the location of the emergency event, as well as the corresponding part of the link plots.