Interaction prediction method and device, electronic equipment and storage medium

Abstract

The present disclosure relates to an interaction prediction method, device and electronic equipment. The method comprises: obtaining to-be-predicted interaction data of a target user, the to-be-predicted interaction data comprising: historical interaction data, to-be-interacted content and a predicted interaction behavior for the to-be-interacted content; inputting the to-be-predicted interaction data into an interaction prediction model, performing structured interaction psychological reasoning on the target user for the predicted interaction behavior based on the historical interaction data and the to-be-interacted content, obtaining an interaction psychological thought chain of the target user, and performing execution prediction on the predicted interaction behavior based on the interaction psychological thought chain to obtain an interaction behavior prediction result; wherein the interaction psychological thought chain is used to prompt potential interaction decision logic of the target user for the predicted interaction behavior. The embodiments of the present disclosure can effectively improve the prediction performance of the interaction prediction model, thereby improving the accuracy of user interaction behavior prediction.

Description

Technical Field

[0001] This disclosure relates to the field of artificial intelligence technology, and in particular to an interactive prediction method, apparatus, electronic device, and storage medium. Background Technology

[0002] In recent years, with the development of large language models, user simulators based on large language models have gradually become a new technological approach. Compared with traditional user simulators based on rule-based, statistical models, or single-task prediction models, large language models can uniformly handle multiple input modalities and task forms, such as binary classification behavior prediction, continuous numerical prediction, and natural language speech generation. Therefore, they are particularly suitable for complex business environments such as short video platforms.

[0003] In related technologies, existing large language models typically make predictions based on the surface similarity between a user's historical behavior and the content to be interacted with, which fails to effectively model the underlying psychological state behind the actual behavior, resulting in poor model prediction performance and low accuracy in predicting user interaction behavior. Summary of the Invention

[0004] This disclosure provides an interaction prediction model training method, interaction prediction method, device, electronic device, and storage medium to at least solve the technical problems in related technologies, such as large language models typically predicting based on the surface similarity between users' historical behavior and the current content to be interacted with, which fails to effectively model the underlying psychological state behind real behavior, resulting in poor model prediction performance and low accuracy in predicting user interaction behavior. The technical solution of this disclosure is as follows: According to a first aspect of the present disclosure, an interactive prediction method is provided, comprising: Obtain the target user's interaction data to be predicted, which includes: historical interaction data, content to be interacted with, and predicted interaction behavior for the content to be interacted with; The interaction data to be predicted is input into the interaction prediction model. Based on the historical interaction data and the content to be interacted with, the target user is subjected to structured interaction psychological reasoning for the predicted interaction behavior to obtain the interaction psychological thought chain of the target user. Based on the interaction psychological thought chain, the predicted interaction behavior is executed to obtain the interaction behavior prediction result. The interactive psychological thought chain is used to prompt the target user regarding the potential interactive decision-making logic of the predicted interactive behavior.

[0005] In an optional embodiment, the step of performing structured interactive psychological reasoning on the target user based on the historical interaction data and the content to be interacted with, to obtain the target user's interactive psychological thought chain, includes: Based on the historical interaction data, the psychological state of the target user is inferred to obtain the current psychological state data of the target user; Based on the current psychological state data and the content to be interacted with, an interaction matching result is obtained; Based on the historical interaction data and the content to be interacted with, decision pattern recognition is performed on the target user for the predicted interaction behavior to obtain decision pattern data. The current psychological state data, the interaction matching results, and the decision-making pattern data are used as the interactive psychological thought chain.

[0006] In an optional embodiment, the interaction data to be predicted further includes: user profile information. Before performing interaction matching based on the current psychological state data and the content to be interacted with to obtain the interaction matching result, the method further includes: Based on the user profile information, interaction features are extracted from the target user to obtain the interaction feature data of the target user; The interaction matching based on the current psychological state data and the content to be interacted with, to obtain the interaction matching result, includes: Based on the interaction feature data and the current psychological state data, the interaction matching analysis is performed on the content to be interacted with to obtain the interaction matching result; After identifying decision pattern data for the target user's predicted interaction behavior based on the historical interaction data and the content to be interacted with, the method further includes: Based on the interaction feature data, the current psychological state data, the interaction matching results, and the decision pattern data, the predicted interaction behavior is executed to obtain an execution prediction summary text. The step of using the current psychological state data, the interaction matching result, and the decision-making pattern data as the interactive psychological thought chain includes: The interaction feature data, the current psychological state data, the interaction matching result, the decision-making pattern data, and the execution prediction summary text are used as the interaction psychological thought chain. In an optional embodiment, the interaction prediction model is trained as follows: Obtain sample interaction data of sample users, the sample interaction data including: sample historical interaction data, sample content to be interacted with, sample predicted interaction behavior for the sample content to be interacted with, and the execution result of the labeled behavior corresponding to the sample predicted interaction behavior; The sample interaction data is input into the teacher's language model. Based on the sample's historical interaction data and the sample's content to be interacted with, the teacher performs structured interactive psychological reasoning on the sample user to predict the interaction behavior of the sample, and obtains the sample user's labeled interactive psychological thought chain. Based on the sample interaction data, the labeled interaction mental thought chain, and the labeled behavior execution results, the student language model is trained for user interaction prediction to obtain the interaction prediction model.

[0007] In an optional embodiment, the method further includes: Based on the prediction task type corresponding to the predicted interaction behavior of the sample, an interactive psychological reasoning instruction corresponding to the prediction task type is constructed; the interactive psychological reasoning instruction is used to indicate the structured reasoning format for the potential interactive psychology of the sample user. The step of inputting the sample interaction data into the teacher's language model, and performing structured interactive psychological reasoning on the sample user based on the sample's historical interaction data and the sample's content to be interacted with, to obtain the sample user's labeled interactive psychological thought chain includes: The sample interaction data and the interaction psychological reasoning instructions are input into the teacher language model. Based on the sample historical interaction data and the sample content to be interacted with, structured interaction psychological reasoning is performed on the sample user to predict the interaction behavior of the sample, and the labeled interaction psychological thought chain of the sample user is obtained.

[0008] In an optional embodiment, the step of constructing the interactive psychological reasoning instruction corresponding to the prediction task type based on the sample prediction interaction behavior includes: If the prediction task type is a category prediction type, the preset mental reasoning instruction that does not carry the execution result of the labeled behavior will be used as the interactive mental reasoning instruction.

[0009] In an optional embodiment, the step of constructing the interactive psychological reasoning instruction corresponding to the prediction task type based on the sample prediction interaction behavior includes: If the prediction task type is a numerical prediction type, set the numerical range prompt information corresponding to the execution result of the annotation behavior; The numerical range prompt information is added to the preset mental reasoning instructions to obtain the interactive mental reasoning instructions.

[0010] In an optional embodiment, the step of constructing the interactive psychological reasoning instruction corresponding to the prediction task type based on the sample prediction interaction behavior includes: If the prediction task type is text generation, the execution result of the annotation behavior is added to the preset mental reasoning instruction to obtain the interactive mental reasoning instruction. The interactive mental reasoning instruction is used to instruct the generation of a structured interactive mental thought chain that conforms to the execution result of the annotation behavior.

[0011] In an optional embodiment, the step of inputting the sample interaction data into the teacher's language model, and based on the sample's historical interaction data and the sample's content to be interacted with, performing structured interactive psychological reasoning on the sample user to predict the interaction behavior of the sample, and obtaining the sample user's labeled interactive psychological thought chain includes: A target temperature parameter is determined, which is used to adjust the probability distribution of the interactive psychological thought chain output by the teacher's language model; The sample interaction data is input into the teacher's language model corresponding to the target temperature parameter. Based on the sample's historical interaction data and the sample's content to be interacted with, multiple interactive psychological inferences are performed on the sample user to predict the interaction behavior of the sample, thereby obtaining multiple candidate interactive psychological thought chains of the sample user. Based on the prediction task type corresponding to the predicted interaction behavior of the sample, the quality of the various candidate interaction mental thought chains is screened to obtain the labeled interaction mental thought chain.

[0012] In an optional embodiment, the step of training the student language model for user interaction prediction based on the sample interaction data, the labeled interaction mental thought chain, and the labeled behavior execution results to obtain the interaction prediction model includes: The sample interaction data is input into the student language model. Based on the sample historical interaction data and the sample content to be interacted with, the sample user is subjected to structured interaction psychological reasoning for the predicted interaction behavior of the sample user. The predicted interaction psychological thought chain of the sample user is obtained. Based on the predicted interaction psychological thought chain, the predicted interaction behavior of the sample is executed to obtain the predicted behavior execution result. Based on the labeled interactive mental thought chain, the labeled behavior execution result, the predicted interactive mental thought chain, and the predicted behavior execution result, the interactive prediction loss information is determined. Based on the interaction prediction loss information, the student language model is updated to obtain the interaction prediction model.

[0013] According to a second aspect of the present disclosure, an interactive prediction apparatus is provided, comprising: The predictable data acquisition module is used to acquire the predictable interaction data of the target user. The predictable interaction data includes: historical interaction data, content to be interacted with, and predicted interaction behavior for the content to be interacted with. The interaction prediction module is used to input the interaction data to be predicted into the interaction prediction model, and based on the historical interaction data and the content to be interacted with, to perform structured interaction psychological reasoning on the target user for the predicted interaction behavior, to obtain the interaction psychological thought chain of the target user, and to perform execution prediction on the predicted interaction behavior based on the interaction psychological thought chain, to obtain the interaction behavior prediction result. The interactive psychological thought chain is used to prompt the target user regarding the potential interactive decision-making logic of the predicted interactive behavior.

[0014] In an optional embodiment, the interaction prediction module includes: The first reasoning unit is used to perform psychological state reasoning on the target user based on the historical interaction data to obtain the target user's current psychological state data. The second reasoning unit is used to perform interactive matching based on the current psychological state data and the content to be interacted with, and obtain the interactive matching result. The third reasoning unit is used to identify the decision pattern of the target user for the predicted interaction behavior based on the historical interaction data and the content to be interacted with, and to obtain decision pattern data. The reasoning and summarizing unit is used to take the current psychological state data, the interaction matching results, and the decision-making pattern data as the interactive psychological thought chain.

[0015] In an optional embodiment, the interaction data to be predicted further includes: user profile information, and the device further includes: The fourth inference unit is used to extract interaction features of the target user based on the user profile information to obtain the interaction feature data of the target user. The second reasoning unit is further configured to perform interaction matching analysis on the content to be interacted with based on the interaction feature data and the current psychological state data, and obtain the interaction matching result; The device further includes: The fifth reasoning unit is used to perform execution prediction on the predicted interactive behavior based on the interaction feature data, the current psychological state data, the interaction matching result, and the decision pattern data, and obtain the execution prediction summary text. The reasoning and summarizing unit is further used to take the interaction feature data, the current psychological state data, the interaction matching result, the decision pattern data, and the execution prediction summary text as the interaction psychological thought chain.

[0016] In an optional embodiment, the interactive prediction model is trained using the following method: The sample data acquisition module is used to acquire sample interaction data of sample users. The sample interaction data includes: sample historical interaction data, sample content to be interacted with, sample predicted interaction behavior for the sample content to be interacted with, and the execution result of the labeled behavior corresponding to the sample predicted interaction behavior. The interactive psychological reasoning module is used to input the sample interaction data into the teacher's language model, and based on the sample's historical interaction data and the sample's content to be interacted with, to perform structured interactive psychological reasoning on the sample user for the predicted interactive behavior of the sample, and obtain the sample user's labeled interactive psychological thought chain; the labeled interactive psychological thought chain is used to prompt the sample user for the potential interactive decision logic of the sample's predicted interactive behavior. The model training module is used to train the student language model for user interaction prediction based on the sample interaction data, the labeled interaction mental thought chain, and the labeled behavior execution results, so as to obtain the interaction prediction model.

[0017] In an optional embodiment, the apparatus further includes: The instruction construction module is used to construct an interactive psychological reasoning instruction corresponding to the prediction task type based on the prediction interactive behavior of the sample; the interactive psychological reasoning instruction is used to indicate a structured reasoning format for the potential interactive psychology of the sample user. The interactive psychological reasoning module is further configured to: input the sample interaction data and the interactive psychological reasoning instructions into the teacher language model; and, based on the sample historical interaction data and the sample content to be interacted with, perform structured interactive psychological reasoning on the sample user to predict the interactive behavior of the sample, thereby obtaining the labeled interactive psychological thought chain of the sample user.

[0018] In an optional embodiment, the instruction building module includes: The first instruction construction unit is used to, if the prediction task type is a category prediction type, use a preset mental reasoning instruction that does not carry the execution result of the labeled behavior as the interactive mental reasoning instruction.

[0019] In an optional embodiment, the instruction building module includes: The second instruction construction unit is used to set numerical range prompt information corresponding to the execution result of the annotation behavior if the prediction task type is numerical prediction type; and to add the numerical range prompt information to the preset psychological reasoning instruction to obtain the interactive psychological reasoning instruction.

[0020] In an optional embodiment, the instruction building module includes: The third instruction construction unit is used to add the execution result of the annotation behavior to the preset mental reasoning instruction if the prediction task type is text generation type, so as to obtain the interactive mental reasoning instruction. The interactive mental reasoning instruction is used to instruct the generation of a structured interactive mental thought chain that conforms to the execution result of the annotation behavior.

[0021] In an optional embodiment, the interactive psychological reasoning module includes: A temperature parameter determination unit is used to determine a target temperature parameter, which is used to adjust the probability distribution of the interactive psychological thought chain output by the teacher's language model. The candidate thought chain sampling unit is used to input the sample interaction data into the teacher language model corresponding to the target temperature parameter, and based on the sample historical interaction data and the sample to be interacted with, to perform multiple interaction psychological inferences on the sample user for the sample predicted interaction behavior, and obtain multiple candidate interaction psychological thought chains of the sample user. The thought chain screening unit is used to perform quality screening on the various candidate interactive psychological thought chains based on the prediction task type corresponding to the sample predicted interactive behavior, so as to obtain the labeled interactive psychological thought chain.

[0022] In an optional embodiment, the model training module includes: The thought chain prediction unit is used to input the sample interaction data into the student language model, and based on the sample historical interaction data and the sample content to be interacted with, to perform structured interaction psychological reasoning on the sample user for the predicted interaction behavior of the sample, to obtain the predicted interaction psychological thought chain of the sample user, and based on the predicted interaction psychological thought chain, to perform execution prediction on the sample predicted interaction behavior, and to obtain the predicted behavior execution result. The loss information determination unit is used to determine interactive prediction loss information based on the annotation interaction mental thought chain, the annotation behavior execution result, the prediction interaction mental thought chain, and the prediction behavior execution result. The model update unit is used to update the student language model based on the interaction prediction loss information to obtain the interaction prediction model.

[0023] According to a third aspect of the present disclosure, an electronic device is provided, comprising: a processor; and a memory for storing processor-executable instructions; wherein the processor is configured to execute the instructions to implement any one of the interaction prediction methods of the present disclosure.

[0024] According to a fourth aspect of the present disclosure, a computer-readable storage medium is provided such that, when instructions in the storage medium are executed by a processor of an electronic device, the electronic device is enabled to perform any one of the interaction prediction methods described in the embodiments of the present disclosure. According to a fifth aspect of the present disclosure, a computer program product comprising instructions is provided that, when run on a computer, causes the computer to perform any of the methods described in the interaction prediction methods of the present disclosure.

[0025] The technical solutions provided by the embodiments of this disclosure have at least the following beneficial effects: This application, in a user interaction behavior prediction scenario, acquires the target user's interaction data to be predicted. This data includes historical interaction data, content to be interacted with, and predicted interaction behaviors for the content. The data is input into an interaction prediction model. Based on the historical interaction data and the content to be interacted with, a structured interaction psychology reasoning process is performed on the target user regarding the predicted interaction behaviors, resulting in the target user's interaction psychology thought chain. This thought chain serves to prompt the target user's potential interaction decision logic for the predicted interaction behaviors. Based on this thought chain, the model performs execution prediction on the predicted interaction behaviors, obtaining the interaction behavior prediction result. By designing a structured interaction psychology reasoning format, the model performs thought chain reasoning on the user's potential interaction psychology before outputting the prediction result. ， This mechanism makes the previously unobservable latent psychological state of users explicit, so that the model no longer relies solely on the similarity of surface behavior for prediction. Instead, it first restores the user's latent psychological state and then deduces the user's behavior, which can effectively improve the predictive performance of the interaction prediction model and thus improve the accuracy of predicting user interaction behavior.

[0026] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0027] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure, and are not intended to unduly limit this disclosure.

[0028] Figure 1 This is a schematic diagram illustrating an application environment according to an exemplary embodiment; Figure 2 This is a flowchart illustrating an interactive prediction method according to an exemplary embodiment; Figure 3 This is a flowchart illustrating another interactive prediction method according to an exemplary embodiment; Figure 4This is a flowchart illustrating another interactive prediction method according to an exemplary embodiment; Figure 5 This is a flowchart illustrating a model training method according to an exemplary embodiment; Figure 6 This is a flowchart illustrating another model training method according to an exemplary embodiment; Figure 7 This is a flowchart illustrating another model training method according to an exemplary embodiment; Figure 8 This is a flowchart illustrating another model training method according to an exemplary embodiment; Figure 9 This is a block diagram illustrating an interactive prediction device according to an exemplary embodiment; Figure 10 This is a block diagram illustrating an electronic device for interactive prediction according to an exemplary embodiment. Detailed Implementation

[0029] To enable those skilled in the art to better understand the technical solutions of this disclosure, the technical solutions in the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings.

[0030] It should be noted that the terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this disclosure are used to distinguish similar different contents and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this disclosure described herein can be implemented in orders other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this disclosure as detailed in the appended claims.

[0031] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for display, data used for analysis, etc.) involved in this disclosure are all information and data authorized by the user or fully authorized by all parties.

[0032] First, some nouns or terms that appear in the description of the embodiments of this disclosure shall be interpreted as follows: Large Language Model (LLM): Also known as a large model, natural language model, or large-scale language model, it refers to a natural language processing model with a large number of parameters and training data. The training process of a large language model typically employs unsupervised learning, that is, training the model using a large-scale text corpus to learn the probability distribution and rules of language. During training, the large language model usually uses a language model as the objective function, optimizing the model parameters by maximizing the predicted probability of the next word.

[0033] Token: In LLM, a token represents the smallest unit of information that the model can understand and generate. The definition of a token may vary in different contexts, but they are generally the basic units for multimodal data analysis and processing. Tokens are assigned numerical values ​​or identifiers, arranged in sequences or vectors, and are input into or output from the model; they are the linguistic building blocks of the model.

[0034] Please see Figure 1 , Figure 1 This is a schematic diagram illustrating an application environment according to an exemplary embodiment, which may include a terminal 100 and a server 200.

[0035] In an optional embodiment, terminal 100 can be used to provide services such as interaction prediction to any user. Specifically, terminal 100 can be, but is not limited to, electronic devices such as smartphones, desktop computers, tablets, laptops, smart speakers, digital assistants, augmented reality (AR) / virtual reality (VR) devices, and smart wearable devices, or software running on the aforementioned electronic devices, such as applications. Optionally, the operating system running on the electronic device can be, but is not limited to, Android, iOS, Linux, Windows, etc.

[0036] In an optional embodiment, server 200 can provide background services to terminal 100. Specifically, server 200 can be a standalone physical server, a server cluster or distributed system consisting of multiple physical servers, or a cloud server providing cloud computing services.

[0037] In addition, it should be noted that, Figure 1 The example shown is merely one application environment provided by this disclosure. In practical applications, other application environments may also be included, such as more terminals.

[0038] In the embodiments described in this specification, the terminal 100 and the server 200 can be directly or indirectly connected through wired or wireless communication, and this disclosure does not impose any restrictions.

[0039] Figure 2 This is a flowchart illustrating an interactive prediction model training method according to an exemplary embodiment, such as... Figure 2 As shown, the method may include the following steps: S201, Obtain the target user's interaction data to be predicted. The interaction data to be predicted includes: historical interaction data, content to be interacted with, and predicted interaction behaviors for the content to be interacted with.

[0040] In one specific embodiment, the target user can be a user on a preset platform or in a preset application. The preset application can refer to any social application, and for example, it can include short video social applications, instant messaging social applications, etc. This application does not limit this.

[0041] In a specific embodiment, the interaction data to be predicted can be the input data of the interaction prediction model, that is, the basis for the interaction prediction model to make interaction predictions for the target user. Specifically, the interaction data to be predicted may include: historical interaction data, content to be interacted with, and predicted interaction behaviors for the content to be interacted with.

[0042] In one specific embodiment, historical interaction data can be the interaction data generated by the target user before the current prediction task (i.e., the user interaction prediction task corresponding to the predicted interaction behavior). Historical interaction data can include a sequence of historical interaction behaviors. Specifically, the system can collect the target user's actual user interaction trajectory within a preset time window before the current prediction task to generate historical interaction data. For example, the preset time window can be three months, but it can also be set to one week, one month, six months, or other periods according to actual business needs. Optionally, to adapt to user interaction prediction tasks with long contexts, a maximum context length can be set, such as 32K tokens, and historical interaction behaviors that are closest to or most relevant to the current prediction point (i.e., the predicted interaction behavior) can be extracted in chronological order.

[0043] For example, real user interaction trajectories may include: behavioral events in multiple scenarios, which may optionally include, but are not limited to: video browsing scenarios, such as exposure, clicks, viewing time, completion, likes, comments, forwards, favorites, and following; live streaming scenarios, such as entering the live streaming room, dwell time, interaction, sending gifts, clicking on products, adding to cart, and placing orders; e-commerce shopping scenarios, such as product exposure, clicks, favorites, adding to cart, placing orders, payments, refunds, and inquiries; customer service dialogue scenarios, such as user inquiries, customer service replies, user follow-up questions, complaints, urging delivery, and expressions regarding returns and exchanges; advertising recommendation scenarios, such as ad exposure, clicks, conversions, negative feedback, and dwell time; and search scenarios, such as search terms, click results, dwell time, and rewriting queries.

[0044] In a specific embodiment, the content to be interacted with can be the interactive content of a scenario to be predicted. Specifically, the scenario to be predicted can include, but is not limited to, interactive scenarios such as video browsing, live streaming, e-commerce shopping, customer service dialogue, advertising recommendation, and search. For example, the interactive content in a video browsing scenario can include the video to be viewed; the interactive content in a live streaming scenario can include the live streaming room; the interactive content in an e-commerce shopping scenario can include products; the interactive content in a customer service dialogue scenario can include customer service replies; the interactive content in an advertising recommendation scenario can include advertisements to be recommended; and the interactive content in a search scenario can include the current search terms.

[0045] In one specific embodiment, the predicted interactive behavior can be the interactive events that the target user can generate in response to the content to be interacted with. For example, taking the video to be viewed in a video browsing scenario as the interactive content to be interacted with, the predicted interactive behavior can include, but is not limited to, the viewing behavior, liking behavior, or commenting behavior that the target user can generate in response to the video to be viewed.

[0046] In an optional embodiment, the interaction data to be predicted may further include: user profile information. Specifically, the user profile information is used to extract the stable attributes and long-term interaction preferences of the target user. Optionally, the user profile information may include, but is not limited to, information such as age and gender.

[0047] S202, Input the interaction data to be predicted into the interaction prediction model, Based on historical interaction data and the content to be interacted with, perform structured interaction psychological reasoning on the target user for the predicted interaction behavior, obtain the interaction psychological thought chain of the target user, and based on the interaction psychological thought chain, perform execution prediction on the predicted interaction behavior to obtain the interaction behavior prediction result. Among them, the interactive psychological thought chain is used to prompt target users about the potential interactive decision-making logic for predicting interactive behavior.

[0048] Specifically, the interaction prediction model can be a language model with the ability to predict user interaction behavior. The interaction prediction model can be used to perform structured interaction psychological reasoning on the target user based on historical interaction data and content to be interacted with, to obtain the target user's interaction psychological thought chain, and based on the interaction psychological thought chain, to perform execution prediction on the predicted interaction behavior, and obtain the interaction behavior prediction result.

[0049] In one specific embodiment, the interactive mental thought chain is used to prompt the target user regarding the potential interactive decision logic for predicting interactive behavior. Illustratively, to facilitate the separation of the interactive mental thought chain output by the model and the interactive behavior prediction result, preset separator labels can be used to separate the interactive mental thought chain and the interactive behavior prediction result. That is, the interactive mental thought chain can be the text content between two preset separator labels output by the interactive prediction model. For example, the preset separator labels can be tags.

[0050] In a specific embodiment, the interaction behavior prediction result is the final output answer of the interaction prediction model. The interaction behavior prediction result can be the prediction result for the user to perform the predicted interaction behavior, that is, the simulation of the user performing the predicted interaction behavior.

[0051] In an optional embodiment, such as Figure 3 As shown, the above-mentioned structured interactive psychological reasoning based on historical interaction data and content to be interacted with, which predicts the interactive behavior of the target user, can include the following interactive psychological thought chain: S301, based on historical interaction data, infer the psychological state of the target user to obtain the target user's current psychological state data.

[0052] Specifically, the target user's current psychological state data is used to characterize the target user's recent psychological state. Optionally, recent psychological states may include, but are not limited to, data such as recent preference drift, current interaction activity density, fatigue level, and emotional tendency. Illustratively, historical interaction data may include historical interaction behavior sequences. The interaction prediction model can infer the target user's dynamic psychological state at the current moment based on the target user's historical interaction behavior sequences. This step analyzes not only long-term interests but also recent preference drift, current activity density, fatigue level, and emotional tendency. For example, a user who has long enjoyed comedy short dramas but has recently been continuously watching live streams about baby products may have experienced recent consumer interest drift; a user who has recently been rapidly scrolling through a large number of videos may be in a state of high fatigue or low patience; a user who continuously inquires about logistics in a customer service conversation may be in a state of impatience or anxiety. This step requires the model to explicitly determine the user's current state, rather than simply reiterating historical behavior.

[0053] S302, perform interaction matching based on the current psychological state data and the content to be interacted with, and obtain the interaction matching result.

[0054] Specifically, the interaction matching results are used to characterize the matching between the target user's recent psychological state and the content to be interacted with. For example, the interaction prediction model can perform interaction matching between the aforementioned current psychological state data and the content to be interacted with, such as determining whether the current content, product, advertisement, search results, or customer service context matches the user's current mood and fatigue state.

[0055] S303, based on historical interaction data and content to be interacted with, identifies decision patterns of target users for predicted interaction behaviors, and obtains decision pattern data.

[0056] Specifically, decision pattern data is used to indicate the target user's decision pattern for predicted interactive behavior. Optionally, decision pattern data can include: rational decision pattern data or random decision pattern data. Rational decision pattern data indicates that the target user's decision pattern for predicted interactive behavior is a rational decision pattern, while random decision pattern data indicates that the target user's decision pattern for predicted interactive behavior is a random decision pattern. Illustratively, a rational decision pattern refers to behavior that can be explained by user preferences, state, content matching, and contextual goals, such as a user buying a product because the price is right, or skipping it because the content doesn't match. A random decision pattern refers to behavior caused by curiosity, attention drift, accidental clicks, emotional fluctuations, or illogical jumps. It is understandable that, since real user behavior is not always completely rational, many short video views, ad clicks, and live stream behaviors are random. If the model forces a rational explanation for all behaviors, it will produce bias. Therefore, the model is required to explicitly acknowledge randomness when judging it as random decision-making and not force the construction of false causal explanations.

[0057] S304 uses current psychological state data, interaction matching results, and decision-making pattern data as an interactive psychological thought chain.

[0058] In an optional embodiment, the interaction data to be predicted further includes: user profile information, which is used before performing interaction matching based on the current psychological state data and the content to be interacted with to obtain the interaction matching result, such as... Figure 4 As shown, the above method also includes: S401, Based on user profile information, extract interaction features of the target user to obtain the target user's interaction feature data.

[0059] Specifically, interaction feature data is used to characterize the target user's stable attributes and long-term interaction preferences. Optionally, interaction feature data may include, but is not limited to: user age group, family role, consumption habits, entertainment preferences, expression style, shopping decision-making habits, long-term interest in or avoidance of certain types of content, etc. The purpose of this step is to provide a global context for subsequent behavior prediction and prevent the model from treating users as average users.

[0060] The interaction matching results obtained by matching the current psychological state data with the content to be interacted with can include: S402, based on interaction feature data and current psychological state data, perform interaction matching analysis on the content to be interacted with to obtain interaction matching results.

[0061] Specifically, the model can match the target user's interaction feature data with the content type of the content to be predicted to obtain a first matching result, and match the target user's current psychological state data with the content state data of the content to be predicted to obtain a second matching result. Based on the first and second matching results, an interaction matching result is obtained. Optionally, the content state data can be collected based on the content type of the content to be predicted. For example, if the content type is video, the content state data may include, but is not limited to, video duration and language; if the content type is product, the content state data may include, but is not limited to, inventory and model number. To illustrate, if the content to be interacted with is a video to be viewed, even if a video belongs to a type that the user has long liked, if the duration is too long and the user is currently fatigued, the probability of completing the video may be low; if the content to be interacted with is a product, even if a product matches the user's recent shopping intentions, if the inventory, color, or logistics are uncertain, the user may continue to inquire rather than place an order immediately.

[0062] After identifying decision-making patterns for predicted interactive behaviors of target users based on historical interaction data and content to be interacted with, and obtaining decision-making pattern data, the above method may further include: S403, based on interaction feature data, current psychological state data, interaction matching results and decision-making pattern data, perform execution prediction on the predicted interaction behavior and obtain the execution prediction summary text; Specifically, the execution prediction summary text can be a summary text of the thought chain that maintains semantic consistency with the interaction behavior prediction result. For example, if the predicted interaction behavior is to play the video completely, the corresponding execution prediction summary text could be "Considering the above factors, I judge that the user will not play the video completely." The corresponding interaction behavior prediction result could be "No." This design is intended to reduce the problem of inconsistency between the reasoning chain and the final answer.

[0063] The aforementioned use of current psychological state data, interaction matching results, and decision-making pattern data as an interactive psychological thought chain can include: S404 uses interaction feature data, current psychological state data, interaction matching results, decision-making pattern data, and execution prediction summary text as an interactive psychological thought chain.

[0064] In an optional embodiment, the structured reasoning format of the interactive psychological thought chain can be expanded or tailored according to the business scenario. For example, in the advertising recommendation scenario, reasoning dimensions such as "commercial intent recognition" or "price sensitivity judgment" can be added; in the customer service scenario, reasoning dimensions such as "problem urgency judgment" or "user dissatisfaction judgment" can be added; and in the search scenario, reasoning dimensions such as "query intent evolution judgment" can be added.

[0065] In the above embodiments, by explicitly modeling the user's potential psychological state through a structured reasoning chain, the model can make predictions from multiple perspectives such as long-term profile, historical behavior, current content matching, and decision-making patterns, which is significantly better than zero-sample prediction and free thought chain fine-tuning.

[0066] In the above embodiments, in the user interaction behavior prediction scenario, the target user's interaction data to be predicted is obtained. The interaction data to be predicted includes: historical interaction data, content to be interacted with, and predicted interaction behavior for the content to be interacted with. The interaction data to be predicted is input into the interaction prediction model. Based on the historical interaction data and the content to be interacted with, structured interaction psychological reasoning is performed on the target user for the predicted interaction behavior to obtain the target user's interaction psychological thought chain. The interaction psychological thought chain is used to prompt the target user's potential interaction decision logic for the predicted interaction behavior. Based on the interaction psychological thought chain, the predicted interaction behavior is executed to obtain the interaction behavior prediction result. By designing a structured interaction psychological reasoning format, the user behavior prediction process is transformed from the traditional "direct mapping of historical behavior and current content to the answer" to "historical behavior and current content are first mapped to a structured potential psychological state reasoning chain, and then the reasoning chain is mapped to the final behavior". The structured reasoning chain, as a natural language agent for the unobservable potential psychological state, enables the model to no longer rely solely on surface behavior similarity for prediction, but to first restore the user's potential psychological state and then deduce the user behavior, which can effectively improve the prediction performance of the interaction prediction model, thereby improving the accuracy of user interaction behavior prediction.

[0067] In an optional embodiment, such as Figure 5 As shown, the above interactive prediction model is trained in the following way: S501, Obtain sample interaction data of sample users. The sample interaction data includes: sample historical interaction data, sample content to be interacted with, sample predicted interaction behavior for sample content to be interacted with, and the execution results of the labeled behavior corresponding to the sample predicted interaction behavior.

[0068] Specifically, the results of the labeled behavior can be preset reference result labels based on sample user interaction data.

[0069] S502, input the sample interaction data into the teacher's language model, and based on the sample's historical interaction data and the sample's content to be interacted with, perform structured interactive psychological reasoning on the sample users to predict their interactive behavior, and obtain the sample users' labeled interactive psychological thought chain.

[0070] In an optional embodiment, the sample interaction data may include: historical sample interaction data, sample content to be interacted with, and sample predicted interaction behavior for the sample content to be interacted with. Accordingly, the detailed steps of "based on historical sample interaction data and sample content to be interacted with, performing structured interaction psychological reasoning on the sample user for the sample predicted interaction behavior to obtain the sample user's labeled interaction psychological thought chain" are the same as those described above. Figure 3 The specific detailed steps are similar and will not be repeated here.

[0071] In an optional embodiment, the sample interaction data may further include: sample user profile information of the sample users. Accordingly, the detailed steps of "based on historical sample interaction data and sample content to be interacted with, performing structured interactive psychological reasoning on the sample users to predict their interactive behavior and obtaining the sample users' labeled interactive psychological thought chain" are the same as those described above. Figure 4 The specific detailed steps are similar and will not be repeated here.

[0072] Specifically, for each prediction sample, the system inputs the sample user profile, sample historical behavior sequence, sample interactive content, and sample predicted interactive behavior into the teacher's language model, and requires the teacher's language model to strictly generate the interactive psychological thought chain (i.e., the reasoning content within the preset separator labels) and the final answer in accordance with the structured interactive psychological reasoning format.

[0073] S503, based on sample interaction data, labeled interaction mental thought chain and labeled behavior execution results, trains the student language model for user interaction prediction to obtain the interaction prediction model.

[0074] In one specific embodiment, the model size of the student language model is smaller than that of the teacher language model. Optionally, the model size can be measured by indicators such as the number of model parameters and memory size.

[0075] In the above embodiments, by distilling the structured thought chain of the teacher's language model and using the structured reasoning chain output by the teacher's language model and the final answer output by the model as the supervision target of the smaller student's language model, the smaller language model can approach or exceed the prediction performance of the larger language model in the user interaction prediction task, thereby reducing the calling cost and latency of user interaction prediction.

[0076] In an optional embodiment, such as Figure 6 As shown, the above method also includes: S601, based on the prediction task type corresponding to the sample prediction interaction behavior, construct the interactive psychological reasoning instruction corresponding to the prediction task type; the interactive psychological reasoning instruction is used to indicate the structured reasoning format for the potential interactive psychology of the sample users.

[0077] Specifically, the prediction task type can be the user interaction prediction task corresponding to the sample prediction interaction behavior. Optionally, the prediction task type can include, but is not limited to: category prediction type, numerical prediction type, text generation type, etc.

[0078] In an optional embodiment, the above-mentioned construction of interactive psychological reasoning instructions corresponding to the prediction task type based on the prediction task type of the sample prediction interaction behavior includes: If the prediction task type is a category prediction type, the preset mental reasoning instruction that does not carry the labeled behavior execution result will be used as the interactive mental reasoning instruction.

[0079] In one specific embodiment, the preset mental reasoning instruction can be a pre-set instruction text for indicating a structured reasoning format for potential interactive psychology of sample users. For example, the preset mental reasoning instruction could be: "In..." <think>The tags infer the user's potential state according to fixed dimensions. This structured thinking chain includes at least the following five dimensions: First, profile interpretation; second, current psychological state inference; third, current content matching; fourth, decision type judgment; and fifth, final decision statement.

[0080] Specifically, the prediction tasks corresponding to the category prediction type can include multi-class tasks. For example, for binary classification tasks, such as whether a video was completed, whether a click was made, whether a purchase was made, whether an ad was followed, or whether an ad was responded to, a pure forward inference strategy is adopted. That is, when the teacher language model generates the interactive mental reasoning chain, it is not injected with the results of the labeled behavior execution. Instead, it is required to make user interaction predictions based solely on the sample's historical interaction data, the sample's content to be interacted with, and the sample's predicted interaction behavior. This avoids the teacher language model from making post-hoc rationalizations after knowing the answer, reducing hindsight bias. For binary classification samples, the interactive mental reasoning chain of the sample is considered to have training value only when the final answer predicted by the teacher language model itself is consistent with the results of the labeled behavior execution.

[0081] In an optional embodiment, the above-mentioned construction of interactive psychological reasoning instructions corresponding to the prediction task type based on the prediction task type of the sample prediction interaction behavior includes: If the prediction task type is numerical prediction, set the numerical range prompt information corresponding to the execution result of the annotation behavior; Add numerical range prompts to the preset mental reasoning instructions to obtain interactive mental reasoning instructions.

[0082] Specifically, the prediction task corresponding to the numerical prediction type can include: numerical prediction task, where the numerical interval prompt information corresponding to the execution result of the labeled behavior is used to prompt the teacher's language model that the execution result of the sample user's sample prediction interaction behavior for the sample interactive content is within the target interval. The target interval can be the preset interval in which the execution result of the labeled behavior is located among multiple preset intervals. These multiple preset intervals can be obtained by dividing the possible value range of the execution result of the sample prediction interaction behavior.

[0083] For example, taking numerical prediction tasks as continuous value prediction tasks, such as viewing duration prediction, dwell time prediction, and spending amount prediction, due to the large output space and numerical sensitivity, pure forward inference is prone to generating semantically reasonable but numerically scaled results. Therefore, instead of directly providing precise answers, coarse-grained interval hints derived from the results of labeled behavior can be provided to the teacher language model. For example, when the actual viewing duration is 5 seconds, the numerical interval hint could be "the user's viewing duration is in the 'very short dwell time' or '0~10' second range"; when the actual viewing duration is 120 seconds, the numerical interval hint could be "the user's viewing duration is in the 'medium viewing range'". This approach improves the scale stability of the numerical values ​​generated by the teacher language model while maintaining the causal inference process.

[0084] In an optional embodiment, the above-mentioned construction of interactive psychological reasoning instructions corresponding to the prediction task type based on the prediction task type of the sample prediction interaction behavior includes: If the prediction task type is text generation, add the execution result of the labeled behavior to the preset mental reasoning instruction to obtain the interactive mental reasoning instruction. The interactive mental reasoning instruction is used to instruct the generation of a structured interactive mental thought chain that conforms to the execution result of the labeled behavior.

[0085] Specifically, for text generation tasks, such as predicting a user's next statement in a customer service dialogue scenario, the output space is open, and the same semantics can be expressed in multiple ways. Relying entirely on multiple free samplings would lead to high costs and difficulty in selection. Therefore, the annotation behavior execution results (i.e., annotated speech text) can be directly injected into the teacher's language model, and the teacher's language model is required to generate a structured interactive mental reasoning chain that can explain why the annotated speech text conforms to the sample user profile, sample historical interaction data, sample content to be interacted with, and sample predicted interactive behavior for the sample content to be interacted with. At this time, the role of the standard interactive mental reasoning chain is not to search for answers again, but to align the real speech with the potential user state, so that the student's language model can learn the generation mechanism of user language style, intent, and emotional expression.

[0086] The above-mentioned input of sample interaction data into the teacher's language model, based on the sample's historical interaction data and the sample's content to be interacted with, performs structured interactive psychological reasoning on the sample users to predict their interactive behavior, and obtains the sample users' labeled interactive psychological thought chain, which includes: S602, input the sample interaction data and interaction psychological reasoning instructions into the teacher's language model, and based on the sample's historical interaction data and the sample's content to be interacted with, perform structured interaction psychological reasoning on the sample user to predict the interaction behavior of the sample, and obtain the sample user's labeled interaction psychological thought chain.

[0087] In the above embodiments, teacher reasoning strategies are designed for three types of tasks: category prediction, numerical prediction, and text generation. Category prediction tasks do not inject real labels to avoid hindsight bias; numerical prediction tasks inject coarse-grained real ranges to improve numerical calibration capabilities; and text generation tasks inject real speech to guide the teacher's language model in interpreting the causal relationship between the generated text and the user's state when reasoning about interactive psychology. The resulting distilled data not only contains the final answer but also a structured reasoning process consistent with real user behavior. By designing different psychological reasoning instruction strategies for different task types, hindsight bias is reduced and data quality is improved.

[0088] In a specific embodiment, such as Figure 7 As shown above, the sample interaction data is input into the teacher's language model. Based on the sample's historical interaction data and the content to be interacted with, structured interactive psychological reasoning is performed on the sample users to predict their interactive behavior. The resulting labeled interactive psychological thought chain of the sample users includes: S701, Determine the target temperature parameter, which is used to adjust the probability distribution of the interactive psychological thought chain output by the teacher's language model.

[0089] S702: Input the sample interaction data into the teacher's language model corresponding to the target temperature parameter. Based on the sample's historical interaction data and the sample's content to be interacted with, perform multiple interactive psychological inferences on the sample user's predicted interactive behavior to obtain multiple candidate interactive psychological thought chains of the sample user.

[0090] Specifically, to obtain diverse explanations for potential psychological states, the system performs multiple parallel samplings on the same prediction sample. A higher temperature parameter can be set during sampling, enabling the teacher's language model to generate multiple candidate interaction reasoning chains. Each candidate interaction reasoning chain should contain complete structured interaction reasoning content and a final answer. This allows the teacher's language model to explore multiple potential explanations behind the same user behavior, and then, through subsequent filtering, retain high-quality reasoning chains that are consistent with the actual results and logically complete.

[0091] In an alternative embodiment, in addition to high-temperature sampling, sampling methods such as kernel sampling, beam search, multi-teacher model ensemble, self-consistent sampling, or candidate ranking based on reward models can be used to generate diverse interpretations of potential mental states.

[0092] S703, based on the prediction task type corresponding to the sample prediction interaction behavior, performs quality screening on multiple candidate interaction mental thought chains to obtain labeled interaction mental thought chains.

[0093] In an optional embodiment, when the prediction task type is a category prediction type, the above-mentioned quality screening of multiple candidate interaction mental thought chains based on the prediction task type corresponding to the sample prediction interaction behavior to obtain the labeled interaction mental thought chain may include: taking the candidate interaction mental thought chain among the multiple candidate interaction mental thought chains whose corresponding execution prediction summary text and the execution result of the labeled behavior are consistent as the labeled interaction mental thought chain.

[0094] Specifically, for category prediction tasks, the system parses the final answer in the candidate interaction mental reasoning chain and compares it with the labeled action execution result. If they match, the reasoning chain is retained; otherwise, it is discarded. If none of the candidate interaction mental reasoning chains for a sample match the labeled action execution result for that sample, the sample is not included in the model training sample set but is recorded separately for subsequent error analysis, hard sample analysis, or additional sampling.

[0095] In an optional embodiment, when the prediction task type is a numerical prediction type, the above-mentioned prediction task type based on the sample prediction interaction behavior, and the quality screening of multiple candidate interaction mental thought chains to obtain the labeled interaction mental thought chain may include: if the error between the interaction behavior prediction result output by the teacher's language model based on a certain candidate interaction mental thought chain and the labeled behavior execution result meets a preset error threshold, the candidate interaction mental thought chain is used as the labeled interaction mental thought chain.

[0096] In an optional embodiment, when the prediction task type is text generation, the above-mentioned quality screening of multiple candidate interaction mental thought chains based on the prediction task type corresponding to the sample prediction interaction behavior to obtain the labeled interaction mental thought chain may include: The teacher's language model performs semantic similarity analysis on the predicted interactive behavior output by each candidate interactive mental thought chain and the labeled behavior execution results to obtain the semantic similarity index corresponding to each candidate interactive mental thought chain; the selected interactive mental thought chain whose corresponding semantic similarity index meets the preset semantic similarity conditions is taken as the labeled interactive mental thought chain.

[0097] Optionally, the semantic similarity analysis here can be performed using methods such as keyword comparison and semantic vector similarity calculation. The preset semantic similarity conditions can be pre-set based on the semantic similarity accuracy requirements in actual applications. Taking semantic similarity analysis using semantic vector similarity calculation as an example, the preset semantic similarity condition can be that the semantic vector similarity is less than the preset similarity threshold.

[0098] As an illustration, for text generation tasks, the system can directly retain samples that are correctly formatted, have complete reasoning, and whose final text is semantically consistent with the target text, based on the injection strategy injected from the annotation execution results. If necessary, a semantic similarity model or a large model reviewer can also be introduced to evaluate the semantic similarity between the generated text and the annotation execution results. Figure 1 The consistency criteria include consistency in user profiles, consistency in knowledge boundaries, and consistency in semantics.

[0099] In an alternative embodiment, quality screening can also be carried out in other ways. For example, rule matching can be used to determine whether the final answers are consistent, or an evaluation model can be used to determine the quality of the inference chain, answer consistency, causal rationality, and format integrity. For numerical prediction tasks, error thresholds, binning consistency, or ranking consistency can be used as screening criteria. For text generation tasks, semantic similarity, manual annotation, or large model reviewers can be used for screening.

[0100] In the above embodiments, for each prediction sample, the teacher's language model generates multiple structured inference chains in parallel, and by rejecting sampling, only high-quality samples that are consistent with the final prediction and the actual result are retained for the student small model's supervised fine-tuning. This mechanism can filter out erroneous inferences and noisy samples, enabling the small model to learn the implicit decision-making logic behind real user behavior, thereby improving prediction accuracy while reducing inference costs, and alleviating overestimation bias, utopian bias and user homogenization bias.

[0101] In an optional embodiment, such as Figure 8 As shown, the above-mentioned user interaction prediction model, trained on the student language model based on sample interaction data, labeled interaction mental thought chain, and labeled behavior execution results, can include: S801: Input the sample interaction data into the student language model. Based on the sample's historical interaction data and the sample's content to be interacted with, perform structured interactive psychological reasoning on the sample user for the sample's predicted interactive behavior, obtain the sample user's predicted interactive psychological thought chain, and based on the predicted interactive psychological thought chain, perform execution prediction on the sample's predicted interactive behavior to obtain the predicted behavior execution result.

[0102] Specifically, the training objective of the interactive prediction model is to enable the student language model to learn conditional probabilities: given the user's historical behavior and the current context, to simultaneously generate a structured reasoning chain and a final predicted answer. Unlike traditional methods that only supervise the final answer, this application also includes the structured interactive mental reasoning chain as a supervision objective, enabling the student language model to not only learn "what to predict," but also "how to infer potential mental states from the user's history and derive behavior from those states."

[0103] S802, based on the labeled interactive mental thought chain, the labeled behavioral execution result, the predicted interactive mental thought chain, and the predicted behavioral execution result, determine the interactive prediction loss information.

[0104] Specifically, the interactive prediction loss information is used to characterize the difference between the teacher's language model prediction results and the student's language model prediction results.

[0105] In a specific embodiment, the above-mentioned determination of interaction prediction loss information based on the labeled interaction mental thought chain, the labeled behavior execution result, the predicted interaction mental thought chain, and the predicted behavior execution result may include: S8021, based on labeled interactive mental thought chains and predicted interactive mental thought chains, determines the loss information of thought chains.

[0106] Specifically, the thought chain loss information is used to characterize the difference between the labeled interactive mental thought chain and the predicted interactive mental thought chain. In a specific embodiment, the thought chain loss information between the labeled interactive mental thought chain and the predicted interactive mental thought chain can be determined based on a preset loss function. Optionally, the preset loss function may include: standard autoregressive negative log-likelihood loss.

[0107] S8022, based on the results of the labeled behavior execution and the results of the predicted behavior execution, determines the result loss information.

[0108] Specifically, the result loss information is used to characterize the difference between the execution results of the annotation behavior and the execution results of the prediction behavior. In a specific embodiment, the result loss information between the execution results of the annotation behavior and the execution results of the prediction behavior can be determined based on a preset loss function. Optionally, the preset loss function may include: standard autoregressive negative log-likelihood loss.

[0109] S8023, based on the fusion processing of thought chain loss information and outcome loss information, obtains interactive prediction loss information.

[0110] Specifically, the loss information of the thought chain and the loss information of the outcome are weighted and fused to obtain the interactive prediction loss information.

[0111] S803 updates the student language model based on the interactive prediction loss information to obtain the interactive prediction model.

[0112] In a specific embodiment, updating the student language model based on the interaction prediction loss information to obtain the interaction prediction model may include: Based on the interactive prediction loss information, update the model parameters of the student language model; perform the next round of iterative training based on the updated student language model until the preset convergence condition is met, and use the student language model that meets the preset convergence condition as the interactive prediction model.

[0113] In one specific embodiment, the student language model can be any small language model with interactive prediction capabilities in the prior art, and this application does not impose any particular limitation on it.

[0114] In a specific embodiment, the preset convergence conditions can be set according to the actual application, such as the number of iterations of training reaching a preset number, the interactive prediction loss information being less than a specified threshold, etc. Specifically, they can be set according to the training speed and model accuracy requirements.

[0115] In the above embodiments, the structured reasoning chain and the final answer output by the model are used as supervision targets at the same time, so that the student language model not only learns to predict what, but also learns how to infer potential psychological states from users' historical interaction data, and deduce behavior from these states, thereby effectively improving the training performance of the student language model.

[0116] As can be seen from the technical solutions provided in the embodiments of this specification above, in a user interaction behavior prediction scenario, the target user's interaction data to be predicted is obtained. This data includes historical interaction data, content to be interacted with, and predicted interaction behaviors for the content. The interaction data to be predicted is input into an interaction prediction model. Based on the historical interaction data and the content to be interacted with, structured interaction psychological reasoning is performed on the target user regarding the predicted interaction behaviors, resulting in the target user's interaction psychological thought chain. This interaction psychological thought chain serves to prompt the target user regarding the potential interaction decision logic for the predicted interaction behaviors. Based on the interaction psychological thought chain, the predicted interaction behaviors are then executed. The test obtains interactive behavior prediction results. By designing a structured interactive psychological reasoning format, the model first performs thought chain reasoning on the user's potential interactive psychology before outputting the prediction result. This mechanism makes the previously unobservable user's potential psychological state explicit, so that the model no longer relies solely on surface behavioral similarity for prediction, but first restores the user's potential psychological state and then deduces the user's behavior, which can effectively improve the predictive performance of the interactive prediction model, thereby improving the accuracy of user interactive behavior prediction. Secondly, by distilling the structured thought chain of the teacher's language model, and using the structured reasoning chain output by the teacher's language model and the final answer output by the model simultaneously as a more accurate prediction, the model can further enhance the prediction accuracy of the interactive behavior prediction model. The supervision objective of small student language models is to enable them to approach or surpass the prediction performance of large language models on user interaction prediction tasks, thereby reducing the invocation cost and latency of user interaction prediction. Furthermore, teacher inference strategies are designed for three types of tasks: category prediction, numerical prediction, and text generation. For category prediction tasks, no real labels are injected to avoid hindsight bias; for numerical prediction tasks, coarse-grained real ranges are injected to improve numerical calibration capabilities; and for text generation tasks, real speech is injected to guide the teacher's language model in interpreting the causal relationship between the generated text and the user's state when inferring the interaction psychology. The distilled data constructed in this way not only contains… The final answer also includes a structured reasoning process consistent with real user behavior. By designing different mental reasoning instruction strategies for different task types, it reduces hindsight bias and improves data quality. In addition, for each prediction sample, the teacher's language model generates multiple structured reasoning chains in parallel, and by rejecting sampling, only high-quality samples that are consistent with the final prediction are retained for supervised fine-tuning of the student's small model. This mechanism can filter out erroneous reasoning and noisy samples, enabling the small model to learn the implicit decision-making logic behind real user behavior. This reduces reasoning costs while improving prediction accuracy and mitigating overestimation bias, utopian bias, and user homogenization bias.

[0117] Figure 9 This is a block diagram illustrating an interactive prediction device according to an exemplary embodiment. (Refer to...) Figure 9 The device includes: The predictable data acquisition module 910 is used to acquire the predictable interaction data of the target user. The predictable interaction data includes: historical interaction data, content to be interacted with, and predicted interaction behavior for the content to be interacted with. The interaction prediction module 920 is used to input the interaction data to be predicted into the interaction prediction model, and based on historical interaction data and the content to be interacted with, to perform structured interaction psychological reasoning on the target user for the predicted interaction behavior, to obtain the target user's interaction psychological thought chain, and based on the interaction psychological thought chain, to perform execution prediction on the predicted interaction behavior, and to obtain the interaction behavior prediction result. Among them, the interactive psychological thought chain is used to prompt target users about the potential interactive decision-making logic for predicting interactive behavior.

[0118] In an optional embodiment, the interaction prediction module 920 includes: The first reasoning unit is used to reason about the target user's mental state based on historical interaction data, and obtain the target user's current mental state data. The second reasoning unit is used to perform interactive matching based on the current psychological state data and the content to be interacted with, and obtain the interactive matching result. The third reasoning unit is used to identify decision patterns of target users based on historical interaction data and content to be interacted with, and to obtain decision pattern data. The reasoning and summarizing unit is used to combine current psychological state data, interaction matching results, and decision-making pattern data into an interactive psychological thought chain.

[0119] In an optional embodiment, the interaction data to be predicted further includes: user profile information, and the device further includes: The fourth inference unit is used to extract interaction features of the target user based on user profile information, and obtain the interaction feature data of the target user. The second reasoning unit is also used to perform interaction matching analysis on the content to be interacted with based on interaction feature data and current psychological state data, and to obtain interaction matching results. The device also includes: The fifth reasoning unit is used to perform execution prediction on the predicted interactive behavior based on interaction feature data, current psychological state data, interaction matching results and decision pattern data, and obtain the execution prediction summary text. The reasoning and summarizing unit is also used to combine interactive feature data, current psychological state data, interactive matching results, decision-making pattern data, and execution prediction summary text as an interactive psychological thought chain.

[0120] In an optional embodiment, the interactive prediction model is trained using the following method: The sample data acquisition module is used to acquire sample user interaction data. The sample interaction data includes: sample historical interaction data, sample content to be interacted with, sample predicted interaction behavior for sample content to be interacted with, and the execution results of the labeled behavior corresponding to the sample predicted interaction behavior. The interactive psychological reasoning module is used to input sample interaction data into the teacher's language model. Based on the sample's historical interaction data and the sample's content to be interacted with, it performs structured interactive psychological reasoning on the sample user to predict the sample's interactive behavior, and obtains the sample user's labeled interactive psychological thought chain. The labeled interactive psychological thought chain is used to prompt the sample user's potential interactive decision logic for the sample's predicted interactive behavior. The model training module is used to train the student language model for user interaction prediction based on sample interaction data, labeled interaction mental thought chains, and labeled behavior execution results, so as to obtain an interaction prediction model.

[0121] In an optional embodiment, the device further includes: The instruction construction module is used to construct interactive psychological reasoning instructions corresponding to the prediction task type based on the sample prediction interaction behavior; the interactive psychological reasoning instructions are used to indicate the structured reasoning format for the potential interactive psychology of the sample users. The interactive psychological reasoning module is also used to: input sample interaction data and interactive psychological reasoning instructions into the teacher's language model; based on the sample's historical interaction data and the sample's content to be interacted with, perform structured interactive psychological reasoning on the sample user to predict the sample's interactive behavior, and obtain the sample user's labeled interactive psychological thought chain.

[0122] In an optional embodiment, the instruction building module includes: The first instruction construction unit is used to, if the prediction task type is a category prediction type, use a preset mental reasoning instruction that does not carry the execution result of the labeled behavior as an interactive mental reasoning instruction.

[0123] In an optional embodiment, the instruction building module includes: The second instruction construction unit is used to set numerical range prompt information corresponding to the execution result of the labeled behavior if the prediction task type is numerical prediction type; and to add numerical range prompt information to the preset mental reasoning instructions to obtain interactive mental reasoning instructions.

[0124] In an optional embodiment, the instruction building module includes: The third instruction construction unit is used to add the execution result of the labeled behavior to the preset mental reasoning instruction if the prediction task type is text generation type, so as to obtain the interactive mental reasoning instruction. The interactive mental reasoning instruction is used to instruct the generation of a structured interactive mental thought chain that conforms to the execution result of the labeled behavior.

[0125] In an optional embodiment, the interactive mental reasoning module includes: The temperature parameter determination unit is used to determine the target temperature parameter, which is used to adjust the probability distribution of the interactive psychological thought chain output by the teacher's language model. The candidate thought chain sampling unit is used to input sample interaction data into the teacher's language model corresponding to the target temperature parameter. Based on the sample's historical interaction data and the sample's content to be interacted with, it performs multiple interaction psychological inferences on the sample user for the sample's predicted interaction behavior, and obtains multiple candidate interaction psychological thought chains of the sample user. The thought chain screening unit is used to perform quality screening on multiple candidate interactive psychological thought chains based on the prediction task type corresponding to the sample prediction interaction behavior, and to obtain labeled interactive psychological thought chains.

[0126] In an optional embodiment, the model training module includes: The thought chain prediction unit is used to input sample interaction data into the student language model. Based on the sample's historical interaction data and the sample's content to be interacted with, it performs structured interactive psychological reasoning on the sample user for the sample's predicted interactive behavior, obtains the sample user's predicted interactive psychological thought chain, and performs execution prediction on the sample's predicted interactive behavior based on the predicted interactive psychological thought chain to obtain the predicted behavior execution result. The loss information determination unit is used to determine the interactive prediction loss information based on the labeled interactive mental thought chain, the labeled behavior execution result, the predicted interactive mental thought chain, and the predicted behavior execution result. The model update unit is used to update the student language model based on the interactive prediction loss information to obtain the interactive prediction model.

[0127] Regarding the apparatus in the above embodiments, the specific manner in which each module performs its operation has been described in detail in the embodiments related to the method, and will not be elaborated upon here.

[0128] Figure 10 This is a block diagram illustrating an electronic device for interactive prediction according to an exemplary embodiment. The electronic device may be a terminal, and its internal structure diagram may be as follows: Figure 10 As shown, it may include RF (Radio Frequency) circuitry 1010, a memory 1020 including one or more computer-readable storage media, an input unit 1030, a display unit 1040, a sensor 1050, an audio circuit 1060, a WiFi (Wireless Fidelity) module 1070, a processor 1080 including one or more processing cores, and a power supply 1090, among other components. Those skilled in the art will understand that... Figure 10 The terminal structure shown does not constitute a limitation on the terminal and may include more or fewer components than shown, or combine certain components, or have different component arrangements. Wherein: The RF circuit 1010 can be used for receiving and transmitting signals during information transmission or calls. Specifically, it receives downlink information from the base station and hands it over to one or more processors 1080 for processing; additionally, it transmits uplink data to the base station. Typically, the RF circuit 1010 includes, but is not limited to, an antenna, at least one amplifier, a tuner, one or more oscillators, a Subscriber Identity Module (SIM) card, a transceiver, a coupler, an LNA (Low Noise Amplifier), a duplexer, etc. Furthermore, the RF circuit 1010 can also communicate wirelessly with networks and other terminals. Wireless communication can use any communication standard or protocol, including but not limited to GSM (Global System for Mobile communication), GPRS (General Packet Radio Service), CDMA (Code Division Multiple Access), WCDMA (Wideband Code Division Multiple Access), LTE (Long Term Evolution), email, SMS (Short Messaging Service), etc.

[0129] The memory 1020 can be used to store software programs and modules. The processor 1080 executes various functional applications and data processing by running the software programs and modules stored in the memory 1020. The memory 1020 may mainly include a program storage area and a data storage area. The program storage area may store the operating system, application programs required for the functions, etc.; the data storage area may store data created according to the use of the terminal, etc. In addition, the memory 1020 may include high-speed random access memory, and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other volatile solid-state storage device. Accordingly, the memory 1020 may also include a memory controller to provide access to the memory 1020 for the processor 1080 and the input unit 1030.

[0130] The input unit 1030 can be used to receive input numerical or character information, and to generate keyboard, mouse, joystick, optical, or trackball signal inputs related to user settings and function control. Specifically, the input unit 1030 may include a touch-sensitive surface 1031 and other input devices 1032. The touch-sensitive surface 1031, also known as a touch display screen or touchpad, can collect touch operations performed by the user on or near it (such as operations performed by the user using a finger, stylus, or any suitable object or accessory on or near the touch-sensitive surface 1031), and drive the corresponding connection device according to a pre-set program. Optionally, the touch-sensitive surface 1031 may include two parts: a touch detection device and a touch controller. The touch detection device detects the user's touch position and the signal generated by the touch operation, and transmits the signal to the touch controller; the touch controller receives touch information from the touch detection device, converts it into touch point coordinates, sends it to the processor 1080, and can receive and execute commands from the processor 1080. In addition, the touch-sensitive surface 1031 can be implemented using various types such as resistive, capacitive, infrared, and surface acoustic wave. Besides the touch-sensitive surface 1031, the input unit 1030 may also include other input devices 1032. Specifically, other input devices 1032 may include, but are not limited to, one or more of the following: a physical keyboard, function keys (such as volume control buttons, power buttons, etc.), a trackball, a mouse, and a joystick. The display unit 1040 can be used to display information input by the user or information provided to the user, as well as various graphical user interfaces of the terminal. These graphical user interfaces can be composed of graphics, text, icons, video, and any combination thereof. The display unit 1040 may include a display panel 1041, which may optionally be configured as an LCD (Liquid Crystal Display), OLED (Organic Light-Emitting Diode), or similar display panel 1041. Further, a touch-sensitive surface 1031 may cover the display panel 1041. When the touch-sensitive surface 1031 detects a touch operation on or near it, it transmits the information to the processor 1080 to determine the type of touch event. Subsequently, the processor 1080 provides corresponding visual output on the display panel 1041 according to the type of touch event. The touch-sensitive surface 1031 and the display panel 1041 can be two independent components to implement input and output functions. However, in some embodiments, the touch-sensitive surface 1031 and the display panel 1041 can be integrated to achieve both input and output functions.

[0131] The terminal may also include at least one sensor 1050, such as a light sensor, a motion sensor, and other sensors. Specifically, the light sensor may include an ambient light sensor and a proximity sensor. The ambient light sensor can adjust the brightness of the display panel 1041 according to the ambient light level, and the proximity sensor can turn off the display panel 1041 and / or the backlight when the terminal is moved to the ear. As a type of motion sensor, a gravity acceleration sensor can detect the magnitude of acceleration in various directions (generally three axes). When stationary, it can detect the magnitude and direction of gravity and can be used for applications that identify the terminal's posture (such as landscape / portrait switching, related games, magnetometer posture calibration), vibration recognition-related functions (such as pedometer, tapping), etc. Other sensors that may be configured on the terminal, such as gyroscopes, barometers, hygrometers, thermometers, and infrared sensors, will not be described in detail here.

[0132] Audio circuitry 1060, speaker 1061, and microphone 1062 provide an audio interface between the user and the terminal. Audio circuitry 1060 converts received audio data into electrical signals, which are then transmitted to speaker 1061, where they are converted into sound signals for output. Conversely, microphone 1062 converts collected sound signals into electrical signals, which are received by audio circuitry 1060, converted back into audio data, and then processed by processor 1080 before being transmitted via RF circuitry 1010 to, for example, another terminal, or output to memory 1020 for further processing. Audio circuitry 1060 may also include an earphone jack to facilitate communication between a peripheral headset and the terminal.

[0133] WiFi is a short-range wireless transmission technology. This terminal, through the WiFi module 1070, can help users send and receive emails, browse web pages, and access streaming media, providing users with wireless broadband internet access. Although Figure 10 The WiFi module 1070 is shown, but it is understood that it is not a necessary component of the terminal and can be omitted as needed without changing the nature of the invention.

[0134] The processor 1080 is the control center of the terminal, connecting various parts of the terminal through various interfaces and lines. It executes software programs and / or modules stored in the memory 1020 and calls data stored in the memory 1020 to perform various functions and process data, thereby enabling overall monitoring of the terminal. Optionally, the processor 1080 may include one or more processing cores; preferably, the processor 1080 may integrate an application processor and a modem processor, wherein the application processor mainly handles the operating system, user interface, and applications, while the modem processor mainly handles wireless communication. It is understood that the modem processor may not be integrated into the processor 1080.

[0135] The terminal also includes a power supply 1090 (such as a battery) to power various components. Preferably, the power supply can be logically connected to the processor 1080 through a power management system, thereby enabling functions such as charging, discharging, and power consumption management through the power management system. The power supply 1090 may also include one or more DC or AC power supplies, recharging systems, power fault detection circuits, power converters or inverters, power status indicators, and other arbitrary components.

[0136] Although not shown, the terminal may also include a camera, Bluetooth module, etc., which will not be described in detail here. Specifically, in this embodiment, the display unit of the terminal is a touch screen display, and the terminal also includes a memory and one or more programs, wherein one or more programs are stored in the memory and configured to be executed by one or more processors of the instructions in the method embodiment of the present invention.

[0137] In an exemplary embodiment, an electronic device is also provided, including: a processor; and a memory for storing processor-executable instructions; wherein the processor is configured to execute the instructions to implement the method as described in any one of the interaction prediction methods of the embodiments of this disclosure.

[0138] In an exemplary embodiment, a computer-readable storage medium is also provided, wherein when the instructions in the storage medium are executed by a processor of an electronic device, the electronic device is enabled to perform any of the methods described in the interaction prediction methods of the embodiments of this disclosure. In an exemplary embodiment, a computer program product including instructions is also provided, which, when run on a computer, causes the computer to perform any of the methods described in the interactive prediction methods of the embodiments of this disclosure.

[0139] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. This computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and RAMbus dynamic RAM (RDRAM), etc.

[0140] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this disclosure are indicated by the following claims.

[0141] It should be understood that this disclosure is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this disclosure is limited only by the appended claims.< / think>

Claims

1. An interactive prediction method, characterized in that, The method includes: Obtain the target user's interaction data to be predicted, which includes: historical interaction data, content to be interacted with, and predicted interaction behavior for the content to be interacted with; The interaction data to be predicted is input into the interaction prediction model. Based on the historical interaction data and the content to be interacted with, the target user is subjected to structured interaction psychological reasoning for the predicted interaction behavior to obtain the interaction psychological thought chain of the target user. Based on the interaction psychological thought chain, the predicted interaction behavior is executed to obtain the interaction behavior prediction result. The interactive psychological thought chain is used to prompt the target user regarding the potential interactive decision-making logic of the predicted interactive behavior.

2. The interactive prediction method according to claim 1, characterized in that, The step of performing structured interactive psychological reasoning on the target user based on the historical interaction data and the content to be interacted with, to obtain the target user's interactive psychological thought chain, includes: Based on the historical interaction data, the psychological state of the target user is inferred to obtain the current psychological state data of the target user; Based on the current psychological state data and the content to be interacted with, an interaction matching result is obtained; Based on the historical interaction data and the content to be interacted with, decision pattern recognition is performed on the target user for the predicted interaction behavior to obtain decision pattern data. The current psychological state data, the interaction matching results, and the decision-making pattern data are used as the interactive psychological thought chain.

3. The interactive prediction method according to claim 2, characterized in that, The interaction data to be predicted further includes: user profile information. Before performing interaction matching based on the current psychological state data and the content to be interacted with to obtain the interaction matching result, the method further includes: Based on the user profile information, interaction features are extracted from the target user to obtain the interaction feature data of the target user; The interaction matching based on the current psychological state data and the content to be interacted with, to obtain the interaction matching result, includes: Based on the interaction feature data and the current psychological state data, the interaction matching analysis is performed on the content to be interacted with to obtain the interaction matching result; After identifying decision pattern data for the target user's predicted interaction behavior based on the historical interaction data and the content to be interacted with, the method further includes: Based on the interaction feature data, the current psychological state data, the interaction matching results, and the decision pattern data, the predicted interaction behavior is executed to obtain an execution prediction summary text. The step of using the current psychological state data, the interaction matching result, and the decision-making pattern data as the interactive psychological thought chain includes: The interaction feature data, the current psychological state data, the interaction matching result, the decision-making pattern data, and the execution prediction summary text are used as the interaction psychological thought chain.

4. The interactive prediction method according to any one of claims 1 to 3, characterized in that, The interactive prediction model was trained in the following manner: Obtain sample interaction data of sample users, the sample interaction data including: sample historical interaction data, sample content to be interacted with, sample predicted interaction behavior for the sample content to be interacted with, and the execution result of the labeled behavior corresponding to the sample predicted interaction behavior; The sample interaction data is input into the teacher's language model. Based on the sample's historical interaction data and the sample's content to be interacted with, the teacher performs structured interactive psychological reasoning on the sample user to predict the interaction behavior of the sample, and obtains the sample user's labeled interactive psychological thought chain. Based on the sample interaction data, the labeled interaction mental thought chain, and the labeled behavior execution results, the student language model is trained for user interaction prediction to obtain the interaction prediction model.

5. The interactive prediction method according to claim 4, characterized in that, The method further includes: Based on the prediction task type corresponding to the predicted interaction behavior of the sample, an interactive psychological reasoning instruction corresponding to the prediction task type is constructed; the interactive psychological reasoning instruction is used to indicate the structured reasoning format for the potential interactive psychology of the sample user. The step of inputting the sample interaction data into the teacher's language model, and performing structured interactive psychological reasoning on the sample user based on the sample's historical interaction data and the sample's content to be interacted with, to obtain the sample user's labeled interactive psychological thought chain includes: The sample interaction data and the interaction psychological reasoning instructions are input into the teacher language model. Based on the sample historical interaction data and the sample content to be interacted with, structured interaction psychological reasoning is performed on the sample user to predict the interaction behavior of the sample, and the labeled interaction psychological thought chain of the sample user is obtained.

6. The interactive prediction method according to claim 5, characterized in that, The step of constructing interactive psychological reasoning instructions corresponding to the prediction task type based on the sample prediction interaction behavior includes: If the prediction task type is a category prediction type, the preset mental reasoning instruction that does not carry the execution result of the labeled behavior will be used as the interactive mental reasoning instruction.

7. The interactive prediction method according to claim 5, characterized in that, The step of constructing interactive psychological reasoning instructions corresponding to the prediction task type based on the sample prediction interaction behavior includes: If the prediction task type is a numerical prediction type, set the numerical range prompt information corresponding to the execution result of the annotation behavior; The numerical range prompt information is added to the preset mental reasoning instructions to obtain the interactive mental reasoning instructions.

8. The interactive prediction method according to claim 5, characterized in that, The step of constructing interactive psychological reasoning instructions corresponding to the prediction task type based on the sample prediction interaction behavior includes: If the prediction task type is text generation, the execution result of the annotation behavior is added to the preset mental reasoning instruction to obtain the interactive mental reasoning instruction. The interactive mental reasoning instruction is used to instruct the generation of a structured interactive mental thought chain that conforms to the execution result of the annotation behavior.

9. The interactive prediction method according to claim 4, characterized in that, The step of inputting the sample interaction data into the teacher's language model, and performing structured interactive psychological reasoning on the sample user based on the sample's historical interaction data and the sample's content to be interacted with, to obtain the sample user's labeled interactive psychological thought chain includes: A target temperature parameter is determined, which is used to adjust the probability distribution of the interactive psychological thought chain output by the teacher's language model; The sample interaction data is input into the teacher's language model corresponding to the target temperature parameter. Based on the sample's historical interaction data and the sample's content to be interacted with, multiple interactive psychological inferences are performed on the sample user to predict the interaction behavior of the sample, thereby obtaining multiple candidate interactive psychological thought chains of the sample user. Based on the prediction task type corresponding to the predicted interaction behavior of the sample, the quality of the various candidate interaction mental thought chains is screened to obtain the labeled interaction mental thought chain.

10. The interactive prediction method according to claim 4, characterized in that, The step of training the student language model for user interaction prediction based on the sample interaction data, the labeled interaction mental chain, and the labeled behavior execution results, to obtain the interaction prediction model, includes: The sample interaction data is input into the student language model. Based on the sample historical interaction data and the sample content to be interacted with, the sample user is subjected to structured interaction psychological reasoning for the predicted interaction behavior of the sample user. The predicted interaction psychological thought chain of the sample user is obtained. Based on the predicted interaction psychological thought chain, the predicted interaction behavior of the sample is executed to obtain the predicted behavior execution result. Based on the labeled interactive mental thought chain, the labeled behavior execution result, the predicted interactive mental thought chain, and the predicted behavior execution result, the interactive prediction loss information is determined. Based on the interaction prediction loss information, the student language model is updated to obtain the interaction prediction model.

11. An interactive prediction device, characterized in that, The device includes: The predictable data acquisition module is used to acquire the predictable interaction data of the target user. The predictable interaction data includes: historical interaction data, content to be interacted with, and predicted interaction behavior for the content to be interacted with. The interaction prediction module is used to input the interaction data to be predicted into the interaction prediction model, and based on the historical interaction data and the content to be interacted with, to perform structured interaction psychological reasoning on the target user for the predicted interaction behavior, to obtain the interaction psychological thought chain of the target user, and to perform execution prediction on the predicted interaction behavior based on the interaction psychological thought chain, to obtain the interaction behavior prediction result. The interactive psychological thought chain is used to prompt the target user regarding the potential interactive decision-making logic of the predicted interactive behavior.

12. An electronic device, characterized in that, include: processor; Memory used to store the processor's executable instructions; The processor is configured to execute the instructions to implement the interactive prediction method as described in any one of claims 1 to 10.

13. A computer-readable storage medium, characterized in that, When the instructions in the storage medium are executed by the processor of the electronic device, the electronic device is able to perform the interactive prediction method as described in any one of claims 1 to 10.

14. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the interactive prediction method as described in any one of claims 1 to 10.

Images