A session information processing method, device and electronic equipment

Abstract

This application discloses a conversation information processing method, apparatus, and electronic device, relating to the field of artificial intelligence technology. The method includes: responding to a user inputting a question on a conversational interaction interface, determining a target virtual object associated with the interface; determining first historical interaction data between the target virtual object and the user, the first historical interaction data including: at least one interaction summary information and its corresponding summary category, the interaction summary information being a summary generated based on at least one historical interaction information between the target virtual object and the user; obtaining identity feature information of the target virtual object; determining at least one target interaction summary information from the at least one interaction summary information based on the identity feature information and the summary category of the interaction summary information; and determining the target virtual object's response information to the question based on the target interaction summary information. This application can improve reasoning accuracy while maintaining the reasoning efficiency of intelligent dialogue interaction.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a method, apparatus and electronic device for processing conversational information. Background Technology

[0002] The application scenarios of intelligent dialogue based on artificial intelligence technology are increasing. For example, intelligent customer service can engage in conversations or answer questions with users; or artificial intelligence technology can be used to generate virtual characters that simulate characters in movies and TV shows, allowing users to interact with these virtual characters through a conversational interface.

[0003] After a user inputs a question into the conversational interface of a conversational application, in order to reasonably determine the response, it is usually necessary to obtain historical conversation information about the user's interactions with the virtual object in the conversational interface, and then infer the response information based on this historical information. However, due to limitations in the data processing capabilities of conversational applications, to ensure the efficiency of inferring the response information, current methods rely on a small amount of recent historical conversation information between the user and the virtual object. Since recent historical conversation information contains relatively little relevant information, it cannot accurately reflect the interaction characteristics between the user and the virtual object, resulting in inaccurate inference of the response information. Conversely, while combining a large amount of historical conversation data from long-term interactions between the user and the virtual object can improve the accuracy of the response information, it leads to lower inference efficiency. Therefore, how to balance the efficiency and accuracy of inferring the response information in intelligent dialogue is a technical problem that needs to be solved by those skilled in the art. Summary of the Invention

[0004] In view of the above problems, this application provides a conversation information processing method, apparatus and electronic device, which can more accurately infer the response information based on the user's input question information while taking into account reasoning efficiency.

[0005] Firstly, this application provides a session information processing method, including:

[0006] In response to a user entering a question in the conversational interaction interface, a target virtual object associated with the conversational interaction interface is determined. The target virtual object is an object in the conversational interaction interface used to simulate a conversational interaction between the conversational participant and the user.

[0007] Determine the first historical interaction data between the target virtual object and the user. The first historical interaction data includes: at least one interaction summary information and a summary category corresponding to the interaction summary information. The interaction summary information is a summary information generated based on at least one historical interaction information between the target virtual object and the user.

[0008] Obtain the identity feature information pre-configured for the target virtual object;

[0009] Based on the identity feature information and the summary category of the interaction summary information, at least one target interaction summary information is determined from the at least one interaction summary information;

[0010] Based on the at least one target interaction summary information, determine the response information of the target virtual object to the question information.

[0011] In one possible implementation of the first aspect, before determining at least one target interaction summary message, the method further includes:

[0012] From the at least one interactive summary information, determine the top target number of candidate interactive summary information that has the highest semantic similarity to the question information;

[0013] The step of determining at least one target interaction summary from the at least one interaction summary based on the summary category of the identity feature information and the interaction summary information includes:

[0014] Based on the identity feature information and the summary category of the interaction summary information, at least one target interaction summary information is determined from the previous target number of candidate interaction summary information.

[0015] In another possible implementation of the first aspect, the first historical interaction data further includes: the time period for the generation of the at least one historical interaction information corresponding to the interaction summary information;

[0016] The step of determining at least one target interaction summary from the at least one interaction summary based on the summary category of the identity feature information and the interaction summary information includes:

[0017] The importance of the interactive summary information is determined based on the time elapsed between the time period corresponding to the generation of the interactive summary information and the current time.

[0018] Based on the identity feature information and the summary category and importance of the interaction summary information, at least one target interaction summary information is determined from the at least one interaction summary information.

[0019] In another possible implementation of the first aspect, the identity feature information includes: the interest coefficients of the target virtual object for different summary categories;

[0020] The step of determining at least one target interaction summary from the at least one interaction summary based on the summary category of the identity feature information and the interaction summary information includes:

[0021] Based on the interest coefficients of the target virtual object for different summary categories and the summary category of the interactive summary information, the interest score of the target virtual object for the interactive summary information is determined;

[0022] Based on the interest scores corresponding to the interactive summary information, sorted from high to low, determine at least one target interactive summary information that ranks first among the at least one interactive summary information.

[0023] In another possible implementation of the first aspect, the interaction summary information is generated based on at least one historical interaction message between the target virtual object and the user during a first historical time period;

[0024] Before determining the response information of the target virtual object to the question, the process also includes:

[0025] Determine the second historical interaction data between the target virtual object and the user. The second historical interaction data includes at least one topic feature information. The topic feature information is the content feature of the topic described by the historical interaction information extracted from the historical interaction information between the target virtual object and the user in a second historical time period. The second historical time period is later than the first historical time period.

[0026] Based on the topic category corresponding to the identity feature information and the topic feature information, at least one target topic feature information is determined from the at least one topic feature information;

[0027] The step of determining the response information of the target virtual object to the question information based on the at least one target interaction summary information includes:

[0028] Based on the at least one target interaction summary information and the at least one target topic feature information, determine the response information of the target virtual object to the question information.

[0029] In another possible implementation of the first aspect, the interaction summary information is a summary information generated based on topic feature information corresponding to at least one historical interaction between the target virtual object and the user;

[0030] The topic feature information corresponding to the historical interaction information is the content feature of the topic described by the historical interaction information extracted from the historical interaction information.

[0031] Among them, the semantic similarity between the topic feature information of at least one historical interaction information corresponding to the interaction summary information exceeds a set threshold.

[0032] In another possible implementation of the first aspect, after obtaining the question information, the method further includes: storing the question information as historical interaction information between the target virtual object and the user;

[0033] After determining the response information, the process also includes:

[0034] The response information is stored as historical interaction information between the target virtual object and the user;

[0035] In response to the current satisfaction of the set topic extraction conditions, based on at least one set topic category, topic feature information belonging to the topic category is extracted from the historical interaction information to obtain topic feature information under at least one topic category corresponding to the historical interaction information.

[0036] The topic feature information corresponding to the historical interaction information is stored in the second historical interaction data between the target virtual object and the user;

[0037] In response to the current satisfaction of the set topic classification conditions, based on the semantic similarity between topic feature information, the topic feature information in the second historical interaction data is clustered to obtain at least one topic feature group, and the topic feature group includes at least one topic feature information.

[0038] For each topic feature group, the summary information corresponding to each topic feature information in the topic feature group is determined by the summary generation model to obtain the interactive summary information of the topic feature group.

[0039] The interaction summary information is stored in the first historical interaction data between the target virtual object and the user.

[0040] In another possible implementation of the first aspect, before determining the response information of the target virtual object to the query information, the method further includes:

[0041] Identify at least one reference virtual object that is associated with the target virtual object, wherein the reference virtual object and the target virtual object are used to simulate different objects in the same media scene;

[0042] Determine the third historical interaction data between the reference virtual object and the user, the third historical interaction data including: at least one reference summary information and the summary category corresponding to the reference summary information, the reference summary information being summary information generated based on at least one historical interaction information between the reference virtual object and the user;

[0043] Based on the identity feature information and the summary category of the reference summary information, at least one target reference summary information is determined from the at least one reference summary information;

[0044] The step of determining the response information of the target virtual object to the question information based on the at least one target interaction summary information includes:

[0045] Based on the at least one target interaction summary information and the at least one reference summary information, the response information of the target virtual object to the question information is determined.

[0046] Secondly, this application also provides a session information processing apparatus, comprising:

[0047] The dialogue determination unit is used to determine the target virtual object associated with the dialogue interaction interface in response to the user inputting a question on the dialogue interaction interface. The target virtual object is the object in the dialogue interaction interface used to simulate the dialogue interaction between the dialogue participant and the user.

[0048] A first data determining unit is used to determine the first historical interaction data between the target virtual object and the user. The first historical interaction data includes: at least one interaction summary information and a summary category corresponding to the interaction summary information. The interaction summary information is a summary information generated based on at least one historical interaction information between the target virtual object and the user.

[0049] An identity determination unit is used to obtain identity feature information pre-configured for the target virtual object;

[0050] A summary selection unit is used to determine at least one target interactive summary from the at least one interactive summary based on the identity feature information and the summary category of the interactive summary information;

[0051] The response determination unit is used to determine the response information of the target virtual object to the question information based on the at least one target interaction summary information.

[0052] Thirdly, this application also provides an electronic device, including at least one processor and a memory connected to the processor, wherein:

[0053] The memory is used to store computer programs;

[0054] The processor is used to execute the computer program to enable the electronic device to implement the session information processing method as described above.

[0055] In this application, after obtaining the question information input by the user in the conversational interaction interface, at least one interaction summary information between the user and the target virtual object associated with the conversational interaction interface is determined. Since the interaction summary information is generated based on at least one historical interaction information between the user and the target virtual object, the data volume of at least one interaction summary information is relatively small compared to a large amount of historical interaction information. Moreover, the at least one interaction summary information can also characterize the key information of the historical interaction between the target virtual object and the user. Based on this, this application can further filter target interaction summary information that matches the identity of the target virtual object to infer the response information based on the summary category of each interaction summary information and the identity feature information of the target virtual object. This not only further reduces the data volume of the interaction data on which the response information is based, but also selects target interaction summary information that is more consistent with the identity of the target virtual object. This allows for a more reasonable inference of the response information that the target virtual object should output in response to the question information based on the selected target interaction summary information, thereby improving the accuracy of the inferred response information. Attached Figure Description

[0056] The above and other features, advantages, and aspects of the embodiments of this disclosure will become more apparent from the accompanying drawings and the following detailed description. Throughout the drawings, the same or similar reference numerals denote the same or similar elements. It should be understood that the drawings are schematic, and the originals and elements are not necessarily drawn to scale.

[0057] Figure 1 A flowchart illustrating the session information processing method provided in this application;

[0058] Figure 2 Another flowchart illustrating the session information processing method provided in this application;

[0059] Figure 3 This is a schematic diagram illustrating one implementation process for generating interactive summary information in this application;

[0060] Figure 4 Another flowchart illustrating the session information processing method provided in this application;

[0061] Figure 5 A schematic diagram illustrating the implementation principle framework of the session information processing method provided in this application;

[0062] Figure 6 Another flowchart illustrating the session information processing method provided in this application;

[0063] Figure 7 Another flowchart illustrating the session information processing method provided in this application;

[0064] Figure 8 A schematic diagram of the architecture of the session information processing device provided in this application;

[0065] Figure 9 A schematic diagram of the component architecture of the electronic device provided in this application. Detailed Implementation

[0066] This application solution is applicable to scenarios of intelligent dialogue based on artificial intelligence (AI) technology, such as conversational interaction scenarios such as chat dialogue, question answering, or information query between real users and intelligent agents, models, or virtual objects, without any specific restrictions.

[0067] The solution in this embodiment can balance reasoning efficiency and the accuracy of reasoning response information during intelligent dialogue interaction.

[0068] The embodiments of this application are described below with reference to the accompanying drawings. The terminology used in the implementation section of this application is only for explaining specific embodiments and is not intended to limit the application. Those skilled in the art will recognize that, with technological advancements and the emergence of new scenarios, the technical solutions provided in the embodiments of this application are also applicable to similar technical problems.

[0069] The terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such terms are interchangeable where appropriate; this is merely a way of distinguishing objects with the same attributes in the embodiments of this application. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion, so that a process, method, system, product, or apparatus that comprises a series of elements is not necessarily limited to those elements, but may include other elements not explicitly listed or inherent to those processes, methods, products, or apparatuses.

[0070] like Figure 1 This illustration shows a schematic diagram of an implementation flow of the session information processing method provided in this application. The method of this embodiment can be applied to electronic devices, such as servers, hosts, or device nodes in cloud systems that provide session interaction services, without any specific limitations.

[0071] The method in this embodiment may include the following steps:

[0072] S101, in response to the user entering a question in the conversational interaction interface, determine the target virtual object associated with the conversational interaction interface.

[0073] The conversational interface can be an interactive interface displayed on a terminal device by an electronic device for information exchange. For example, in this application, the conversational interface can be used to obtain user-inputted questions and output responses from the electronic device in response to those questions. The terminal device can be a mobile phone, laptop, or desktop computer, etc., and can establish a communication connection with the electronic device and display the conversational interface. For example, the terminal device can have a conversational application installed to connect to the electronic device, and based on this application, a connection can be established with the electronic device and the conversational interface can be displayed.

[0074] The question information is the information entered by the user in the conversation interaction interface. For example, the question information can be a question, chat statement, or task instruction, etc., and there are no restrictions on it.

[0075] In this application, a virtual object is an object used in a conversational interface to simulate conversational interaction between a conversational participant and the user. For example, the conversational interface may display the user's avatar or username, and it may also display the virtual object's avatar or nickname. Therefore, the virtual object represented by this object identifier is the virtual object associated with the conversational interface. Different users can choose different virtual objects for chat or question-and-answer sessions as needed; therefore, the conversational interface will differ depending on the virtual object selected by the user. For ease of distinction, this application refers to the virtual object associated with the conversational interface corresponding to the user as the target virtual object.

[0076] For example, the target virtual object is a conversational role object simulated by an intelligent agent or model. For instance, the conversational role object can be a virtual character, a virtual cartoon object, or an AI role object used to simulate a character in a movie or TV series.

[0077] For example, the target virtual object can also be an intelligent agent or the model itself. When the conversational interaction interface is different, it may be necessary to use different intelligent agents or models to interact with the user. Therefore, the intelligent agents or models associated with different conversational interaction interfaces can be different.

[0078] S102, determine the first historical interaction data between the target virtual object and the user.

[0079] The first historical interaction data includes: at least one interaction summary and the summary category corresponding to the interaction summary.

[0080] The interaction summary information is generated based on at least one historical interaction between the target virtual object and the user. The historical interaction information refers to the interaction information between the target virtual object and the user before the current moment (i.e., in history). Therefore, the historical interaction information may include at least one of the following: a historical question input by the user and a historical response from the target virtual object to the historical question.

[0081] In one alternative approach, in order to more accurately reflect the summary content of a dialogue interaction between a user and a target virtual object, the historical interaction information may include a historical question and the corresponding historical response.

[0082] In this application, each interaction summary is generated based on at least one historical interaction. This at least one historical interaction can be a pre-determined historical interaction with a specific relationship, such as semantic similarity exceeding a set threshold, or belonging to the same topic, etc., without any specific limitations.

[0083] The interaction summary information can be a summary statement or a description of the theme content of at least one historical interaction message. For example, if at least one historical interaction message is an interaction message in which the user and the target virtual object discussed the time, cause, condition of the leg injury, and emotional distress after the injury, then the interaction summary information could be "Leg injury, emotional distress".

[0084] In this application, there are several possible implementations for determining interaction summary information based on at least one historical interaction message. For example, a summary generation model can be used to determine the summary information corresponding to the at least one historical interaction message, thus obtaining the interaction summary information. This summary generation model can be a large language model, or other trained machine learning models, etc., without limitation. Of course, interaction summary information can also be determined through topic extraction or other methods, without specific limitations.

[0085] The summary category of the interactive summary information refers to the category of the summary information described by the interactive summary information. For example, each summary category can be a topic category, and the summary category to which the interactive summary information belongs can be classified based on its content. For example, if the interactive summary information is: "Recently feeling down, experiencing the death of a pet and feelings of loneliness," then the corresponding summary category can be "feeling lonely."

[0086] It is understandable that, since the interaction summary information determined based on at least one historical interaction information is a summary of at least one historical interaction information, the interaction summary information can accurately express the key information such as the summary topic of the at least one summary information, and the amount of data in the interaction summary information is less than that of the at least one historical interaction information.

[0087] S103, obtain the identity feature information pre-configured for the target virtual object.

[0088] The identity feature information is used to characterize the identity features of the target virtual object.

[0089] For example, taking the target virtual object as a simulated conversational role object, the identity characteristics of the conversational role object can reflect one or more of the following: the role it represents, the role category, the role personality, the role occupation, the role age, and the role gender.

[0090] For example, taking the target virtual object as an intelligent agent or model, the identity feature information of the target virtual object is used to characterize the types of tasks that the intelligent agent or model is good at performing and the task processing style, etc.

[0091] In one example, identity feature information can exist in the form of a vector, where different dimensions of the vector correspond to the feature information of the target virtual object in one identity dimension.

[0092] S104, Based on the summary category of identity feature information and interaction summary information, determine at least one target interaction summary information from at least one interaction summary information.

[0093] For example, from at least one interactive summary message, determine a set number of target interactive summary messages that have the highest matching degree with the identity feature information, or at least one target interactive summary message whose matching degree exceeds a set threshold, etc.

[0094] It is understandable that when the identity characteristics of the target virtual object are different, the target interaction summary information determined by this application from at least one interaction summary information will also be different. This will enable the target virtual object to determine the response information based on different interaction summary information during the interaction between the user and different target virtual objects, allowing the user to experience different interactions with objects with different identity characteristics.

[0095] S105, based on at least one target interaction summary information, determine the response information of the target virtual object to the question information.

[0096] Depending on the type of question, the response can have several possibilities. For example, if the question is a chat message entered by the user, the response can be a chat reply; if the question is a problem inquiry, the response can be an answer to the problem.

[0097] In this application, there are no restrictions on the specific implementation of generating response information based on the at least one target interaction summary information and the question information. For example, a model such as a large language model can be used to generate response information for the question information based on the at least one target interaction summary information and the question information.

[0098] Of course, in practical applications, the interaction information of the user within the most recent set time in the conversation interface before the user entered the question can also be used as context information. Combined with the target interaction summary information and context information, the corresponding response information can be determined.

[0099] It is understandable that after generating the response information, this application can output the response information to the session interaction interface, and the specifics will not be elaborated further.

[0100] As can be seen from the above, in this application, after obtaining the question information input by the user in the conversation interaction interface, at least one interaction summary information between the user and the target virtual object associated with the conversation interaction interface is determined. Since the interaction summary information is generated based on at least one historical interaction information between the user and the target virtual object, the data volume of at least one interaction summary information is relatively small compared to a large amount of historical interaction information. Moreover, the at least one interaction summary information can also characterize the key information of the historical interaction between the target virtual object and the user. Based on this, this application can further filter target interaction summary information that matches the identity of the target virtual object to infer the response information based on the summary category of each interaction summary information and the identity feature information of the target virtual object. This not only further reduces the data volume of the interaction data on which the response information is based, but also selects target interaction summary information that is more consistent with the identity of the target virtual object. This allows for a more reasonable inference of the response information that the target virtual object should output in response to the question information based on the selected target interaction summary information, thereby improving the accuracy of the inferred response information.

[0101] In this application, there are several possibilities for determining the specific implementation of the target interaction summary information from at least one interaction summary information based on the identity feature information of the target virtual object.

[0102] In one possible implementation, to more intuitively represent the identity characteristics of the target virtual object, this application uses the identity characteristic information to characterize the target virtual object's degree of interest in different topic categories. Accordingly, this application can determine at least one target interaction summary that the target virtual object is interested in based on the identity characteristic information and the summary categories of each interaction summary.

[0103] In this implementation, since the identity feature information of the target virtual object can characterize the target virtual object's degree of interest in different topic categories, the target interaction summary information that the target virtual object is interested in can be matched from at least one interaction summary information based on the identity feature information. The response information can be combined with the interaction summary information corresponding to the historical interaction information that the target virtual object is interested in, so that the generated response information can better reflect the identity feature of the target virtual object.

[0104] For example, let's take a simulated virtual character object as the target virtual object:

[0105] If the target virtual object is a professor in a movie or TV series, then the identity characteristics of the target virtual object can indicate that the target virtual object is more interested in topics related to academics, teaching, and various disciplines. Therefore, based on the identity characteristics of the target virtual object, the matched target interaction summary information can be some target interaction summary information related to academics, teaching, and disciplines that the user has interacted with the target virtual object used to simulate a professor in the past.

[0106] If the target virtual object is a palace maid in a period drama, then the palace maid may pay more attention to some chat details and historical information. Based on the identity characteristics of the target virtual object, the matched target interaction summary information can be some event information and historical interaction summary information that the user has talked about with the target virtual object in the past.

[0107] The following example uses a specific form of identity feature information, combined with... Figure 2 Please provide an explanation. For example... Figure 2 This illustrates another flowchart of the session information processing method provided in this application. The method in this embodiment may include:

[0108] S201, in response to the user entering a question in the conversational interaction interface, determine the target virtual object associated with the conversational interaction interface.

[0109] The target virtual object is the object used in the conversational interface to simulate the conversational interaction between the conversational participant and the user.

[0110] S202, determine the first historical interaction data between the target virtual object and the user.

[0111] The first historical interaction data includes at least one interaction summary and the corresponding summary category. The interaction summary is generated based on at least one historical interaction between the target virtual object and the user.

[0112] S203, obtain the identity feature information pre-configured for the target virtual object.

[0113] The identity feature information includes: the interest coefficients of the target virtual object for different summary categories. The interest coefficient for each summary category represents the degree of interest the target virtual object has in that particular summary category.

[0114] For example, suppose that the interaction summary information generated based on the user's historical interaction information with the target virtual object can be divided into 5 summary categories, namely: emotion, work, stress, studies and economy.

[0115] Therefore, the identity feature information of the target virtual object can be represented as {1, 0.5, 0.4, 0.2, 0.9}, where the interest coefficient corresponding to the summary category of emotion is 1, which is the largest interest coefficient. Therefore, the target virtual object has the greatest interest in the interactive summary information of the emotion category.

[0116] S204, for each summary category corresponding to each interactive summary information, based on the interest coefficient of the target virtual object for different summary categories and the summary category of the interactive summary information, determine the interest score of the target virtual object for the interactive summary information.

[0117] Specifically, for each type of interactive summary information, the interest score of the interactive summary information is positively correlated with the interest coefficient of the target virtual object for the summary category to which the interactive summary information belongs.

[0118] For example, the interest score of interactive summary information can be the product of the base score of the summary category corresponding to the interactive summary information and the interest coefficient of the target virtual object. The base score corresponding to the summary category can be preset; different summary categories can have the same or different base scores. For instance, the base score for some commonly used summary categories that are of interest to most users can be set relatively high, while the base score for other summary categories with relatively lower importance can be set relatively low.

[0119] S205, based on the interest scores corresponding to the interactive summary information in descending order, determine the top-ranked target interactive summary information from the at least one interactive summary information.

[0120] For example, the maximum number of targets from which target interaction summary information can be selected can be preset, and correspondingly, the number of target interaction summary information items with the highest interest scores can be determined.

[0121] S206, Based on the at least one target interaction summary information, determine the response information of the target virtual object to the question information.

[0122] This step S206 can be referred to in the relevant description of the previous embodiments, and will not be repeated here.

[0123] In this embodiment, the identity feature information of the target virtual object may include the target virtual object's interest coefficients for different summary categories. Based on the target virtual object's interest coefficients for different summary categories and the summary category of each interactive summary, the target virtual object's interest score for each interactive summary can be determined. Based on this, this application can select at least one target interactive summary with the highest interest score based on the target virtual object's interest score for each interactive summary. This enables the reasonable selection of target interactive summary information for interaction between the target virtual object and the user, based on the target virtual object's identity features such as its role. This allows for a more reasonable and efficient selection of interactive summary information for generating response information by combining the identity features of different target virtual objects.

[0124] Understandably, historical interaction information between a user and a target virtual object accumulates over time. In practical applications, interaction summary information can be generated periodically or irregularly based on the stored historical interaction information. The earlier the generation time of at least one historical interaction corresponding to the interaction summary information, the lower the likelihood that the user might need to pay attention to that interaction summary information. Similarly, for the target virtual object simulating a conversational interaction, the longer the time period between the generation of at least one historical interaction corresponding to the interaction summary information and the current moment, the lower the likelihood that determining the target virtual object's response information requires relying on that interaction summary information.

[0125] Specifically, when the target virtual object is a dialogue character, such as a simulated human or a cartoon character, considering the different identities and memory characteristics of different dialogue characters, the required time period for generating interaction summary information varies for each dialogue character to generate a response. In other words, different dialogue characters have different "memory durations" for different interaction summary information. For example, if the dialogue character is a professor in a movie or TV show, professors generally have good memory; therefore, interaction summary information extracted from historical interactions within the last month can be used as their memory information to generate a response. If the dialogue character is a kindergarten student in the same movie or TV show, since kindergarten students have relatively short memory spans, only interaction summary information generated from historical interactions within the last week may be used as their memory information to generate a response.

[0126] Based on this, in any of the above embodiments of this application, the first historical interaction data may further include: the generation time period of at least one historical interaction information corresponding to each interaction summary information. The generation time period may cover the generation time (i.e., the time of generation) of each of the at least one piece of historical interaction data.

[0127] For example, assuming that the historical interaction information that generated the interaction summary information was generated between the user and the target virtual object at 19:00 on May 1, 2025, 10:30 on May 2, 2025, and 12:30 on May 2, 2025, the time period for the generation of the historical interaction information corresponding to the interaction summary information is: "19:00 on May 1, 2025 - 12:30 on May 2, 2025".

[0128] Accordingly, in this application, determining at least one target interaction summary based on the identity feature information of the target virtual object and the summary category of each interaction summary can be as follows: for each type of interaction summary, the importance of the interaction summary is determined based on the time elapsed between the generation time period corresponding to the interaction summary and the current time; then, based on the identity feature information and the summary category and importance of the interaction summary, at least one target interaction summary is determined from the at least one interaction summary.

[0129] The importance of the interactive summary information is negatively correlated with the time elapsed between the time period in which it was generated and the current time. Of course, the importance of the interactive summary information is also related to its summary category; interactive summaries within different summary categories may be assigned different initial levels of importance.

[0130] Of course, in practical applications, considering that different virtual objects have different identity characteristics, their "memory duration" also varies. Therefore, in this application, the identity characteristics of the target virtual object can be used to characterize its memory decay characteristics. These characteristics reflect the target virtual object's level of attention to interactive summary information generated within a week. For example, the target virtual object may have the same level of attention to interactive summary information generated a week ago; however, for interactive summary information generated a week ago, the longer the time elapsed between the current date and the current date, the lower the target virtual object's level of attention to that interactive summary information.

[0131] Among them, the memory decay characteristics represented by the identity feature information of different virtual objects can be different.

[0132] Based on this, this application can also combine the identity feature information of the target virtual object with the time period corresponding to the generation of the interaction summary information to determine the importance of the interaction summary information.

[0133] Among them, the specific implementation of selecting the target interaction summary information can be implemented in a variety of ways, based on the identity feature information and the summary category and importance of the interaction summary information, without any specific restrictions.

[0134] For example, firstly, select at least one candidate interactive summary from at least one interactive summary that has an importance level exceeding a set threshold. Then, based on identity feature information and the summary category of the interactive summary information, determine the target interactive summary from the at least one candidate interactive summary. For example, such as Figure 2 As shown in the example, the interest score of each candidate interactive summary is calculated, and then at least one interactive summary with the highest interest score is selected as the target interactive summary.

[0135] For example, a weight value can be determined for each interactive summary based on its importance. Then, based on this weight value, the identity feature information, and the summary category of the interactive summary, an importance score can be determined. For instance, based on the identity feature information and the summary category, an interest score can be determined for each interactive summary. Multiplying this interest score by the weight value yields the importance score for each interactive summary. Based on this, a predetermined number of interactive summaries with high importance scores can be selected as target interactive summaries.

[0136] Understandably, the importance of interactive summaries is determined based on the time elapsed between their generation and the current time. This approach takes into account that interactive summaries generated a longer time elapsed from the current time are less important for determining the current response. This helps to filter out target interactive summaries that are generated a shorter time elapsed from the current time and match the identity characteristics of the target virtual object. As a result, the filtered target interactive summaries are more closely related to the current question, which naturally improves the accuracy and reliability of the generated response.

[0137] In this application, there are multiple possible ways to generate interactive summary information, without any specific restrictions.

[0138] For example, in the first possible implementation, the interaction summary information can be obtained by extracting the summary information of at least one historical interaction information that is semantically similar between the target virtual object and the user using a summary generation model.

[0139] In the second possible implementation, in order to further reduce the amount of data in the interaction summary information and enable the interaction summary information to more accurately reflect the summary information corresponding to at least one historical interaction information, the interaction summary information can be a summary information generated based on the topic feature information corresponding to each of the target virtual object and the user's at least one historical interaction information.

[0140] The topic feature information corresponding to the historical interaction information is the content feature of the topic described by the historical interaction information extracted from it. In this application, the topic category can be pre-configured; for example, topic categories can be divided into multiple categories such as: events, emotions, character profiles, and intention expressions, without specific limitations. Each piece of historical interaction information can describe the topic content under one or more topic categories; therefore, each piece of historical interaction information can have topic feature information corresponding to at least one topic category.

[0141] Specifically, the semantic similarity between the topic feature information of at least one historical interaction message corresponding to the interaction summary information exceeds a set threshold. For example, in practical applications, this application can cluster the topic feature information of each historical interaction message based on its semantics to obtain at least one topic feature group. Each topic feature group includes at least one semantically similar topic feature information. Based on this, an interaction summary message can be generated for each topic feature information in a cluster.

[0142] There are various possible implementations for extracting topic feature information from historical interaction information, and no specific restrictions are imposed.

[0143] For example, a topic extraction model can be used to extract the content features of the topics described by historical interaction information, and obtain the topic feature information corresponding to the historical interaction information. The topic extraction model can be a large language model or a trained machine learning model, etc., without any restrictions.

[0144] For example, this application can pre-configure topic extraction templates corresponding to different topic categories. Based on this, and using the topic extraction templates corresponding to various topic categories, the content features of different topics in historical interaction information can be extracted to obtain at least one topic feature information corresponding to the historical interaction information.

[0145] One type of topic extraction template can define extraction rules for content features related to that topic category. For example, the template can define at least one topic keyword contained in the historical interaction information belonging to that topic category, as well as the types of information to be extracted under that topic. For example, the topic extraction template for the topic category of "event" includes, but is not limited to: event occurrence time: [specific time, accurate to day / hour]; event type: [preset category: such as consumption behavior, information acquisition, health-related, etc.]; event description: [concise summary of the core process of the event, without subjective expression]; event intensity: [0-1 value, representing the importance to the user].

[0146] In practical applications, the two methods mentioned above can be combined to extract topic feature information corresponding to historical interaction information. That is, based on the topic extraction template corresponding to each topic, the topic extraction model is used to extract the topic feature information corresponding to historical interaction information. In this case, this application can first construct prompt words based on the prompt word templates corresponding to each topic category to indicate the extraction of topic feature information in combination with the topic extraction template. Then, the prompt words and the historical interaction information are input into the topic extraction model to obtain the extracted topic feature information.

[0147] For example:

[0148] For the topic category "Event", the prompt template can be: "You are an event extraction tool. Please refer to the 'Event Extraction Template' based on the 'Interaction Information' below to check if any events have occurred. If so, please extract the content features of the topic 'Event' in no more than 30 words. If not, please ignore this."

[0149] For the topic category of "emotion", the prompt template can be: "You are an emotion extraction tool. Please refer to the 'emotion extraction template' based on the 'interaction information' below to check if there is any emotion-related content. If so, please extract the emotion category, the reason for the emotion, and the emotion intensity (between 0 and 1) for the topic of 'emotion', etc., with a word count not exceeding 10 words. If not, please ignore it."

[0150] For the topic category "Intent", the prompt template is: "You are an intent extraction tool. Based on the 'interaction information' below and referring to the 'intent extraction template,' infer whether there is relevant intent information. If so, extract the content features of the intent in no more than 20 characters. If not, please ignore it."

[0151] Suppose the historical interaction information includes: a historical question: "Yesterday I had a latte at Cafe A and I was very happy. Please recommend another drink from Cafe A, preferably with less sweetness," and a historical reply to this question: "Café A's iced Americano is their signature drink; I recommend you try it." Using a large model to extract topic features from this historical interaction information, we can obtain the following topic feature information:

[0152] The topic characteristics of this event category: I had a latte at Cafe A.

[0153] The topic characteristic information for the emotion category is: happy, with an intensity of 0.8.

[0154] The topic characteristics of this topic category are: A coffee shop's product recommendation request, with a preference for non-sweet drinks.

[0155] In particular, since the data output by the large language model is structured data, the feature information for each topic can also be structured data. For example, taking the topic category of emotion as an example, the structured data of the topic feature types for emotion output by the large language model can be as follows:

[0156] {"time": "May 1, 2025" / / Time when the emotion occurred,}

[0157] "Type": Positive emotion / / Emotion type

[0158] "emotion": "happy" / / Emotional description

[0159] "score": "0.8" / / Emotional intensity.

[0160] Of course, the above example illustrates that a single historical interaction message can extract topic feature information corresponding to multiple topics. In practical applications, a single historical interaction message may belong to only one topic. Therefore, a single historical interaction message may only extract one topic feature information.

[0161] Understandably, in this second possible implementation, this application first determines the topic feature information of historical interaction information. Since the topic feature information of historical interaction information is a content feature describing a topic category extracted from historical interaction information, the amount of data in this topic feature information is relatively small, and it can more accurately describe the content features of a topic described by the historical interaction information. Based on this, this application generates interaction summary information based on topic feature information with semantic similarity exceeding a set threshold. This not only further reduces the information redundancy of the generated interaction summary information and reduces the amount of data, but also enables the interaction summary information to accurately express the summary content of the user's history and the target virtual object's history of interaction on similar topics.

[0162] In the above embodiments of this application, the specific method for generating the interaction summary information corresponding to the target virtual object and the user is not limited. Considering that the topic content described by semantically similar topic feature information in the topic feature information corresponding to each historical interaction information is related, which is beneficial to accurately extracting the summary information of the interaction between the user and the target virtual object on that topic category, this application can also first cluster each topic feature information, and then generate interaction summary information for each clustered topic group. The following is combined with Figure 3 Please provide an explanation. For example... Figure 3 This diagram illustrates one implementation process for generating interactive summary information according to this application. The process may include:

[0163] S301, obtain at least one historical interaction information between the stored user and the target virtual object.

[0164] It is understandable that the user's historical input questions and the target virtual object's historical responses to those questions can be stored as historical interaction information during each interaction. Therefore, this historical interaction information can be the interaction information generated between the user and the target virtual object on the current session interface before the current moment, or it can be the interaction information generated during conversations between the user and the target virtual object on historical session interfaces; there are no specific restrictions.

[0165] As mentioned above, each piece of historical interaction information may include one or both of the following: a historical question and the corresponding historical response.

[0166] In one possible implementation, each time historical interaction information between the user and the target virtual object is obtained, it can be stored in a database or a designated storage area. Accordingly, at least one piece of historical interaction information stored in the database or the designated storage area can be obtained.

[0167] Of course, after obtaining at least one piece of historical interaction information, this application can also preprocess the at least one piece of historical interaction information to remove historical interaction information with format errors or duplicate historical interaction data.

[0168] S302, in response to the current fulfillment of the set topic extraction conditions, based on at least one set topic category, extract topic feature information belonging to the topic category from the historical interaction information, and obtain topic feature information under at least one topic category corresponding to the historical interaction information.

[0169] The topic extraction condition represents the current condition for extracting topic feature information from historical interaction information.

[0170] The extraction criteria for this topic can be set according to actual needs, without any specific restrictions.

[0171] For example, a topic extraction period can be set to extract topic features from historical interaction information obtained within the most recent period that has not yet had its topic feature information extracted. Correspondingly, if the current time for topic extraction is determined based on the set topic extraction period, then the topic extraction condition is considered met. The duration of this topic extraction period can be 1 day, 1 week, or 1 month, etc., without specific restrictions. For example, if the topic extraction period is 7 days, then the topic extraction condition will be determined to be met once every 7 days, and step S302 will be executed for each piece of historical interaction information obtained within the most recent 7 days.

[0172] For example, the topic extraction condition can be that the number of historical interaction information between the user and the target virtual object reaches a first set number.

[0173] For example, the topic extraction condition can also be that at least one historical interaction message contains historical interaction information related to a specified topic category. For instance, if historical interaction information related to the topic category of "event" is detected, then the topic extraction condition is determined to be met.

[0174] Of course, there are other possibilities for topic extraction conditions, which can be set according to actual needs without restriction. In practical applications, it is also possible to combine the above-mentioned topic extraction conditions or satisfy several of them simultaneously, which will not be elaborated here.

[0175] There are multiple ways to extract topic feature information belonging to a certain topic category from a historical interaction message. For specific implementations, please refer to the previous introductions. For example, you can combine the topic extraction template corresponding to the topic category, or use a large language model to extract the topic feature information of the historical interaction message under the topic type, etc., which will not be elaborated here.

[0176] S303, store the topic feature information corresponding to the historical interaction information into the second historical interaction data between the target virtual object and the user.

[0177] In this application, the second historical interaction data may include at least one topic feature information corresponding to the user and the target virtual object.

[0178] For example, the topic feature information in the second historical interaction data can be stored in the form of a knowledge graph, where the topic feature information of each historical interaction information serves as a branch node of the historical interaction information.

[0179] Understandably, each historical interaction can correspond to at least one topic feature, indicating a relatively large amount of data related to topic features in historical interactions. However, because topic feature information describes the content characteristics of the topics described in historical interactions, it represents detailed information within a specific topic category within the historical interactions, thus more accurately reflecting the detailed content of the topics described in the most recently collected historical interactions. Therefore, the topic feature information in the second set of historical interaction data can reflect the content characteristics of the topics described in the most recently collected historical interactions.

[0180] It is understandable that after extracting topic feature information corresponding to at least one topic category from historical interaction information, the topic category corresponding to each topic feature information is determined. In order to ensure that the second historical interaction data can comprehensively include the relevant information of each topic feature information, this application can also store the topic category to which the topic feature information belongs in the second historical interaction data.

[0181] S304, in response to the current satisfaction of the set topic classification conditions, based on the semantic similarity between topic feature information, cluster the topic feature information in the second historical interaction data to obtain at least one topic feature group.

[0182] The topic feature group includes at least one topic feature. Since this application clusters topic feature information based on the semantic similarity between them, at least one topic feature in the topic feature group is a topic feature with a semantic similarity exceeding a set threshold.

[0183] The specific implementation method for clustering the feature information of each topic can be unrestricted.

[0184] In one example, a specified clustering algorithm can be used to cluster the topic feature information in the second historical interaction data based on the semantic similarity between topic feature information. There can be a variety of clustering algorithms, and this application does not limit them.

[0185] In another example, models such as large language models can be used to combine the semantic similarity of topic feature information to cluster the topic feature information in the second historical interaction data, and at least one topic feature group can be obtained from the clustering.

[0186] The topic classification conditions are pre-set conditions used to cluster topic feature information in the second historical interaction data. They can be set as needed and are not restricted.

[0187] For example, meeting the topic categorization criteria can include: determining the current time when topic categorization is required according to a set topic categorization cycle. This cycle can be two weeks, one month, or two months, etc., with no specific restrictions. Typically, the duration of this topic categorization cycle can be longer than the topic extraction cycle.

[0188] For example, meeting the topic classification criteria may include: the number of topic feature information in the second historical interaction data reaches a second set number.

[0189] For example, meeting the topic classification conditions may include: the topic category corresponding to the topic feature information in the second historical interaction data is an emotion topic, and the emotion intensity or the degree of fluctuation of emotion intensity in the topic feature information of the emotion topic exceeds a set intensity threshold.

[0190] For example, meeting the topic categorization criteria could include: the existence of topic feature information with specified keywords within a defined topic category in the second set of historical interaction data. For instance, the defined topic category could be an event topic, and the specified keywords under this event topic could include, but are not limited to, event keywords representing important moments such as graduation, work, marriage, and childbirth.

[0191] Of course, there are other possibilities for this topic classification criteria. Furthermore, topic classification criteria can also include combinations of the above, which will not be elaborated further.

[0192] S305, For each topic feature group, use the summary generation model to determine the summary information corresponding to each topic feature information in the topic feature group, and obtain the interactive summary information of the topic feature group.

[0193] As mentioned earlier, the summary generation model can be a large language model or other machine learning models, as described above, and will not be repeated here.

[0194] For example, let's take a simple example and assume that the topic feature group includes the following simple topic feature information:

[0195] Topic Feature Information 1: Event Topic: On April 3, 2025, a cat I had raised for 5 years died;

[0196] Topic Feature Information 2: Emotional Topic, April 3, 2025, Very sad, Emotional Intensity 0.9;

[0197] Topic Feature Information 3: Event Topic, April 12, 2025, I've been feeling very lonely lately;

[0198] Topic Feature Information 4: Emotional Topic: April 12, 2025, Loneliness, Emotional Intensity 0.8;

[0199] Based on the above topic features, the generated interactive summary information could be: From April 3, 2025 to April 12, 2025, the individual experienced the loss of a pet, felt depressed, and experienced loneliness.

[0200] In this embodiment, the interactive summary information is the summary information of at least one topic feature information with semantic similarity. Since each summary topic feature is the content feature of a historical interactive information under a certain topic category, the interactive summary information can reflect the summary information of at least one historical interactive information under similar topic features. This makes the interactive summary information relatively less data than at least one historical interactive information, thus enabling the summary content of at least one historical interactive information under similar topics to be reflected with less data.

[0201] Understandably, after generating the interaction summary information, in order to efficiently locate the main content of the user's interactions with the target virtual object over a relatively long period of time, this application can also store the interaction summary information in the first historical interaction data between the target virtual object and the user. For example, each interaction summary information can be used as a node in a knowledge graph, and the interaction summary information in the first historical interaction data can be stored in the form of a knowledge graph. Of course, other forms can also be used to store the interaction summary information, and there are no restrictions on this.

[0202] As can be seen from the process of generating interaction summary information, after obtaining the historical interaction information between the user and the target virtual object, this application first extracts topic feature information from the recently obtained historical interaction information and stores the extracted topic feature information in the second historical interaction data. On this basis, only when the topic classification conditions are met will the topic feature information in the second historical interaction data be clustered. Then, an interaction summary information is generated for each clustered topic feature group and stored in the first historical interaction data. It can be seen that, relative to the second historical interaction data, the summary interaction data stored in the first historical interaction data is equivalent to long-term memory data corresponding to historical interaction information that is much further away from the present.

[0203] Furthermore, the topic feature information in the second historical interaction data can be used as short-term memory data corresponding to the recently collected historical feature information.

[0204] It is understandable that for conversation feature information that has been clustered into topic feature groups, since interaction summary information has already been generated based on each conversation feature information in the conversation feature group and stored in the first historical interaction data, that is, the interaction summary has already been used as long-term memory data stored in the first historical interaction data, in order to avoid duplicate storage of information between the first historical interaction data and the first historical interaction data (i.e., long-term memory data and short-term memory data), this application can delete each conversation feature information in the conversation feature group from the second historical interaction data after generating the interaction summary information.

[0205] Similarly, after storing the session feature information corresponding to the historical interaction information into the first historical interaction data, this application can delete the stored historical interaction information so that the session feature information can be extracted only from the newly added interaction information in the future.

[0206] It is understood that, simultaneously with or after generating interactive summary information, this application can also determine the summary category corresponding to the interactive summary information. For example, while generating interactive summary information using a summary generation model, the model can output the summary category corresponding to the interactive summary information. Alternatively, the model can analyze the topic category described by the interactive summary information and use that topic category as the summary category. Another example is that, based on at least one keyword corresponding to each of the different defined summary categories, the summary category matching the keywords in the interactive summary information can be determined, and that summary category can be used as the summary category of the interactive summary information. Of course, there can be other ways to determine the summary category, and there are no limitations on this.

[0207] Based on this, this application may also store the summary category corresponding to the interaction summary information in the first historical interaction data.

[0208] Furthermore, this application can also determine at least one historical interaction information to which each topic feature information in the topic feature group of the interaction summary information pair belongs, and determine the generation time period corresponding to the interaction summary information based on the generation time of each of the at least one historical interaction information, as described above, and will not be repeated here. Accordingly, this application can also store the generation time period corresponding to the interaction summary information in the first historical interaction data.

[0209] Understandably, as the number of questions and responses between the user and the target virtual object increases, this application can continuously update historical interaction information and update the first and second historical interaction data based on the updated historical interaction information. This ensures that the session feature information in the second historical interaction data accurately reflects the content characteristics of the user's most recent interaction with the target virtual object, and that the second historical interaction data also accurately and comprehensively reflects the content characteristics of the interaction between the user and the target virtual object over a longer period. Therefore, after obtaining the question information, this application can also store the question information as historical interaction information between the target virtual object and the user. Similarly, after determining the response information, this application can also store the response information as historical interaction information between the target virtual object and the user.

[0210] For example, the question can be stored as a historical interaction record, and once a response is determined, that response can also be stored in the historical interaction record corresponding to the question. Alternatively, the question and response can be stored separately in the historical interaction records; there are no specific restrictions.

[0211] It is understandable that after storing the question and reply information as historical interaction information, if the set topic extraction conditions are met later, this application can still perform the operations of steps S302 to S305 and store the generated interaction summary information in the first historical interaction data between the target virtual object and the user, as described above, and will not be repeated here.

[0212] In the above embodiments of this application, since the interaction summary information in the first historical interaction data can be generated based on at least one historical interaction information between the target virtual object and the user in the first historical time period, the interaction summary information of the first historical interaction data needs to be generated after a period of time after obtaining the historical interaction information. That is, each interaction summary information in the first historical interaction data reflects the summary information of historical interaction information that is a certain time away from the current time.

[0213] As mentioned above, in practical applications, recently collected historical interaction information is also important for determining the response information corresponding to the prompt information. Based on this, in this application, at least one target historical interaction information that matches the identity feature information can be selected from the recently collected historical interaction information based on the identity feature information of the target virtual object, and the response information can be determined based on the target historical interaction information and the target interaction summary information.

[0214] Considering the potentially large volume of recently collected historical interaction information, and given the limitations of AI models in supporting input sequences in intelligent dialogue scenarios, this application addresses the need to reduce the amount of data processed to determine responses while maintaining accuracy. Specifically, it first converts historical interaction information into session feature information, then identifies only the target session feature information matching the identity feature information of the target virtual object for generating response information. The following section discusses this approach in conjunction with... Figure 4 Please provide an explanation.

[0215] like Figure 4 This illustrates another flowchart of the session information processing method provided in this application. The method in this embodiment may include:

[0216] S401, in response to the user entering a question in the interactive session interface, determine the target virtual object associated with the interactive session interface.

[0217] The target virtual object is the object used in the conversational interface to simulate conversational interaction between the conversational participant and the user.

[0218] S402, determine the first historical interaction data between the target virtual object and the user.

[0219] The first historical interaction data includes at least one interaction summary and its corresponding summary category. The interaction summary is generated based on at least one historical interaction between the target virtual object and the user during the first historical time period. The specific implementation for generating the interaction summary can be found in the previous embodiments and will not be repeated here.

[0220] S403, obtain the identity feature information pre-configured for the target virtual object.

[0221] S404, based on the identity feature information and the summary category of the interaction summary information, determine at least one target interaction summary information from at least one interaction summary information in the first historical interaction data.

[0222] The above steps S401 to S404 can be found in the description of any embodiment in this application, and will not be repeated here.

[0223] S405, determine the second historical interaction data between the target virtual object and the user.

[0224] The second historical interaction data includes at least one topic feature information. This topic feature information is the content feature of the topic described by the historical interaction information extracted from the historical interaction information between the target virtual object and the user in the second historical time period.

[0225] Each topic feature corresponds to a historical interaction record. For details on generating topic feature information based on historical interaction information and the corresponding topic category, please refer to the previous sections. Figure 3 The relevant descriptions of the embodiments will not be repeated here.

[0226] In this embodiment, the second historical time period is later than the first historical time period. For example, the first historical time period could be one month ago, while the second historical time period could be the most recent month. That is, the time when the historical interaction information used to generate the interaction summary information in the first historical interaction data was generated is earlier than the time when the historical interaction information used to generate the session feature information in the second historical interaction data was generated.

[0227] Based on this, the interaction summary information in the first historical interaction data can serve as long-term memory data of the interaction between the user and the target virtual object, while the topic feature information in the second historical interaction data can serve as short-term memory data of the interaction between the user and the target virtual object. However, unlike current applications that directly use historical interaction information for both long-term and short-term memory data, this application uses the session feature information extracted from the historical interaction information as short-term memory data, and further clusters and extracts summaries from the session feature information in the short-term memory data to obtain the interaction summary information for long-term memory data. This results in relatively less data in both long-term and short-term memory data, and more accurately and intuitively reflects the interaction content characteristics between the user and the target virtual object.

[0228] It should be noted that the order of steps S405 and S402 is not limited to... Figure 4 As shown, in practical applications, the order of these two steps can be interchanged or performed simultaneously.

[0229] S406, Based on the identity feature information and the topic category corresponding to the topic feature information, determine at least one target topic feature information from the at least one topic feature information.

[0230] In this embodiment, there are multiple possible implementations for determining the target topic information, and no restrictions are imposed on them.

[0231] In one example, the similarity between the topic category and the identity feature information of each topic feature information is calculated, and at least one topic feature information with the highest similarity ranking is selected as at least one target topic feature information.

[0232] In another example, the identity feature information also includes: the target virtual object's preference coefficient for different topic categories, which represents the degree to which the target virtual object likes the topic category. Accordingly, based on the target virtual object's preference coefficient for different topic categories and the topic category of the topic feature information, the target virtual object's preference score for that topic feature information can be determined. Then, according to the preference scores corresponding to the conversation feature information, ranked from high to low, at least one top-ranked target conversation feature information is determined from the second historical interaction data.

[0233] The process of determining the preference score is similar to that of determining the interest score, and there are no specific restrictions. For example, the preference score can be determined by multiplying the base score of the topic feature information by the preference coefficient of the target virtual object for that topic category. The base score of the topic feature information can be set according to actual needs.

[0234] In one alternative approach, if the topic category of the topic feature information is a topic category other than emotional topics, then the base score corresponding to the topic feature information is 1; if the topic category is an emotional topic, the emotional intensity in the topic feature information can be used as the base score corresponding to the topic feature information.

[0235] In another example, based on the identity feature information, the topic category corresponding to the topic feature information, and the generation time corresponding to the topic feature information, at least one target topic feature information is determined from the at least one topic feature information. The generation time corresponding to the topic feature information can be the generation time of the historical interaction information corresponding to the topic feature information.

[0236] Of course, there are other ways to determine the characteristics of a target topic, without any restrictions.

[0237] S407, Based on the at least one target interaction summary information and the at least one target topic feature information, determine the response information of the target virtual object to the question information.

[0238] For example, the target interaction summary information, the target topic feature information, and the prompt information can be input into AI models such as large language models to obtain the response information output by the AI ​​models.

[0239] This embodiment does not impose any restrictions on the specific implementation of generating this response information.

[0240] In this embodiment, not only is at least one target interaction summary information obtained from the first historical interaction data based on the identity feature information of the target virtual object, but at least one target session feature information is also obtained from the second historical interaction data. Since the first and second historical interaction data are generated from historical interaction data obtained in different time periods, they can respectively reflect the information characteristics of the long-term and recent interactions between the target virtual object and the user. Moreover, the interaction summary information in the first historical interaction data and the session feature information in the second historical interaction data have less data volume and are more concise and refined than the historical interaction data. Therefore, determining the response information of the target virtual object based on at least one target interaction summary information and at least one target topic feature information not only reduces the amount of data processing and improves data inference consumption, but also makes the data referenced in generating the response information more comprehensive, further improving the accuracy of the inferred response information.

[0241] To facilitate understanding of this embodiment, the following is combined with... Figure 5 A brief introduction will be given. For example... Figure 5 A schematic diagram illustrating the implementation principle framework of the session information processing method provided in this application is shown.

[0242] Depend on Figure 5 It can be seen that after obtaining and storing the interaction information between the user and the target virtual object, if the topic extraction conditions are met (e.g., the topic extraction time is determined according to the set topic extraction cycle), the topic feature information of the historical interaction information belonging to the topic category will be extracted based on the set conversation categories such as emotion, event and user profile, and the topic feature information of at least one topic category corresponding to the historical interaction information will be stored as the second historical interaction data (i.e. short-term memory data) between the user and the target virtual object.

[0243] If the second historical interaction data meets the topic classification conditions, the topic feature information is clustered based on the semantic similarity between the topic feature information in the second historical interaction data. For each topic feature group that is clustered, interaction summary information is generated based on the topic feature information in that topic feature group. The interaction summary information is stored in the second historical interaction data (i.e., long-term memory data) between the user and the target virtual object.

[0244] Based on this, by Figure 5As can be seen, after detecting that a user has entered a question into the conversational interface, this application can, based on the identity feature information of the target virtual object associated with the conversational interface, match at least one target conversation feature from the second historical interaction data (short-term memory data) and at least one target interaction summary from the first historical interaction data (long-term memory data). Then, by inputting the question, the matched target conversation feature information, and the target interaction summary information into the model, a response can be generated for the question.

[0245] Since the target session feature information and the target interaction summary information represent the characteristics of the user's short-term memory data and long-term memory data with the target virtual object, respectively, and the target session feature information and the target interaction summary information have less data volume than the complete historical interaction information, while ensuring the integrity of key information, the accuracy of the reasoning response information can be further improved while taking into account reasoning efficiency.

[0246] In any of the above embodiments of this application, considering that the historical interaction information required to generate the corresponding response information will also differ depending on the question information, in order to generate the response information more accurately, before determining at least one target interaction summary information, this application can also determine the top target number of candidate interaction summary information with the highest semantic similarity to the question information from at least one interaction summary information of the first historical interaction data. The target number can be set according to actual needs, such as 15 or 20. Correspondingly, this application can determine at least one target interaction summary information from the top target number of candidate interaction summary information based on identity feature information and the summary category of the interaction summary information.

[0247] To facilitate understanding of this method of determining target interactive summary information, the following will combine... Figure 6 Let's take one specific implementation method as an example for illustration. For example... Figure 6 This illustrates another flowchart of the session information processing method provided in this application. The method in this embodiment may include:

[0248] S601, in response to the user entering a question in the interactive session interface, determines the target virtual object associated with the interactive session interface.

[0249] The target virtual object is the object used in the conversational interface to simulate conversational interaction between the conversational participant and the user.

[0250] S602, determine the first historical interaction data between the target virtual object and the user.

[0251] The first historical interaction data includes: at least one interaction summary, the summary category corresponding to the interaction summary, and the time period corresponding to the generation of the interaction summary.

[0252] The generation of interactive summary information can be found in the description of any of the preceding embodiments, and will not be repeated here.

[0253] The interaction summary information is the time period corresponding to at least one historical interaction information (such as historical interaction information corresponding to at least one session feature information) used to generate the interaction summary.

[0254] S603, obtain the identity feature information pre-configured for the target virtual object.

[0255] For example, this identity feature information is used to characterize the degree of interest of the target virtual object in different summary categories and different topic categories.

[0256] The above steps can be found in the descriptions of other embodiments of this application, and will not be repeated here.

[0257] S604, from at least one interactive summary information, determine the number of candidate interactive summary information that have the highest semantic similarity to the question information.

[0258] The semantic similarity between the question information and the candidate interactive summary information can be obtained by calculating the vector feature similarity between the vector features corresponding to the question information and the vector features corresponding to the candidate interactive summary information, or it can be calculated in other ways without restriction.

[0259] S605, for each candidate interaction summary information, the importance of the candidate interaction summary information is determined based on the time elapsed between the time period corresponding to the generation of the interaction summary information and the current time.

[0260] In this embodiment, one implementation method is used as an example for illustration. If the interactive summary information is not considered, that is, if step S605 is not executed, it is also applicable to this embodiment.

[0261] S606, based on the identity feature information and the summary category and importance of the candidate interaction summary information, determine at least one target interaction summary information from the target number of candidate interaction summary information.

[0262] For example, based on the target virtual object's interest coefficient for different summary categories, the summary category of the interactive summary information, and its importance, an interest score for the target virtual object regarding the interactive summary information can be determined. For instance, since importance can be represented by an importance score, the interest score can be the product of the target virtual object's interest coefficient for the summary category of the interactive summary information, the basic score of the interactive summary information, and the importance score. Correspondingly, at least one target interactive summary information can be identified from the at least one interactive summary information by ranking the interactive summary information from highest to lowest interest score.

[0263] For example, a predetermined number of candidate summary messages with higher importance can be identified from the target number of candidate summary messages, where the predetermined number is less than the target number. Then, based on the identity feature information and the summary category of the candidate interaction summary messages, at least one target interaction summary message can be identified from the predetermined number of candidate interaction summary messages.

[0264] Of course, there are other ways to determine the target interaction summary information, and there are no restrictions on this.

[0265] S607, Based on the at least one target interaction summary information, determine the response information of the target virtual object to the question information.

[0266] It is understood that, in this embodiment, before step S607, at least one target topic feature information can be determined from at least one topic feature information in the second historical interaction data based on the topic category corresponding to the identity feature information and topic feature information. Accordingly, step S607 can be based on the at least one target interaction summary information and the at least one topic feature information to determine the response information of the target virtual object to the question information.

[0267] Of course, in this embodiment and any of the above embodiments of this application, after the response information is determined, the response information corresponding to the target virtual object can also be output to the conversation interaction interface.

[0268] In any of the above embodiments of this application, considering that if two or more virtual objects are different objects in the same simulated media scene (such as the same film or television drama, the same game, or characters from the same historical period), then in order to improve the flexibility of interaction, the historical interaction information between these virtual objects and the user can also be shared. The following is in conjunction with... Figure 7 Please provide an explanation. For example... Figure 7 This illustration shows another flowchart of the session information processing method provided in this application. The method in this embodiment may include:

[0269] S701, in response to the user entering a question in the interactive session interface, determines the target virtual object associated with the interactive session interface.

[0270] The target virtual object is the object used in the conversational interface to simulate conversational interaction between the conversational participant and the user.

[0271] S702, determine the first historical interaction data between the target virtual object and the user.

[0272] The first historical interaction data includes at least one interaction summary and the summary category corresponding to each interaction summary. It also includes the time period corresponding to each interaction summary.

[0273] S703, obtain the identity feature information pre-configured for the target virtual object.

[0274] S704, based on the identity feature information and the summary category of the interaction summary information, determine at least one target interaction summary information from the at least one interaction summary information.

[0275] The above steps can be found in the description of any of the preceding embodiments, and will not be repeated here.

[0276] S705, determine at least one reference virtual object that is associated with the target virtual object.

[0277] The reference virtual object and the target virtual object are used to simulate different objects in the same media scene. These different objects in the same media scene can be different characters from the same film or television series, game characters from the same game, different characters from the same historical period, or different characters from the same book, etc.

[0278] For example, if the target virtual object is a character A in a certain film or television drama M, then the reference virtual object can be a character B in the same film or television drama M that has a direct or indirect relationship with character A. For example, if character A is a princess, character B can be the princess's father, the princess's maid, or the princess's siblings, etc.

[0279] In practical applications, reference virtual objects that are associated with the target virtual object can be pre-configured.

[0280] S706, determine the third historical interaction data between the reference virtual object and the user.

[0281] The third historical interaction data includes at least one reference summary and the corresponding summary category. The reference summary is generated based on at least one historical interaction between a reference virtual object and the user.

[0282] In this application, the reference summary information is actually the interaction summary information between the reference virtual object and the user. For ease of distinction, the interaction summary information between the reference virtual object and the user is referred to as reference summary information. Similarly, for ease of distinction, the historical interaction information between the reference virtual object and the user can also be referred to as reference historical interaction information.

[0283] The specific implementation of generating reference summary information based on each reference historical interaction information is similar to the process of generating interaction summary information based on at least one historical interaction information between the target virtual object and the user, and will not be repeated here.

[0284] S707, Based on the identity feature information and the summary category of the reference summary information, at least one target reference summary information is determined from the at least one reference summary information.

[0285] The method of selecting target reference summary information from at least one reference summary information based on identity feature information is similar to the previous method of selecting target interactive summary information. For example, at least one target reference summary information whose summary category matches the identity feature information can be selected. The specific details will not be elaborated here.

[0286] S708, based on at least one target interaction summary information and at least one reference summary information, determine the response information of the target virtual object to the question information.

[0287] It is understandable that, since the reference summary information is generated based on the historical interaction information between the reference virtual object and the user, and the reference virtual object and the target virtual object belong to the same media scene and have a relationship, the reference summary information between the reference virtual object and the user can also serve as the reference information that the target virtual object needs to refer to when answering the question. This allows the target virtual object to understand the historical interaction information with the user through intelligent dialogue by other virtual objects, thereby enabling artificial intelligence technology to more accurately determine the response information that the target virtual object needs to give to the question.

[0288] For example, if the target virtual object is a princess in a movie or TV series M, and the reference virtual object can be a palace maid of the princess in the same movie or TV series M, then by determining the target reference summary information that the princess is interested in from the reference summary information of the "palace maid" and the user, it is possible to achieve the purpose of the "princess" learning about the user's recent emotions or events from the "palace maid," thereby enabling the "princess" to give more accurate and flexible responses to the questions given by the user.

[0289] Furthermore, this application also provides a session information processing apparatus. For example... Figure 8 This diagram illustrates a possible structural composition of the session information processing apparatus provided in this application. The apparatus in this embodiment may include:

[0290] The dialogue determination unit 801 is used to determine the target virtual object associated with the dialogue interaction interface in response to the user inputting a question on the dialogue interaction interface. The target virtual object is the object in the dialogue interaction interface used to simulate the dialogue interaction between the dialogue participant and the user.

[0291] The first data determination unit 802 is used to determine the first historical interaction data between the target virtual object and the user. The first historical interaction data includes: at least one interaction summary information and the summary category corresponding to the interaction summary information. The interaction summary information is a summary information generated based on at least one historical interaction information between the target virtual object and the user.

[0292] The identity determination unit 803 is used to obtain the identity feature information pre-configured for the target virtual object;

[0293] The summary selection unit 804 is used to determine at least one target interactive summary information from the at least one interactive summary information based on the identity feature information and the summary category of the interactive summary information;

[0294] The response determination unit 805 is used to determine the response information of the target virtual object to the question information based on the at least one target interaction summary information.

[0295] In one possible implementation, the session information processing device further includes:

[0296] The candidate determination unit is used to determine the top number of candidate interactive summary information with the highest semantic similarity to the question information from at least one interactive summary information before the summary selection unit determines at least one target interactive summary information;

[0297] Accordingly, the summary selection unit is specifically used to determine at least one target interactive summary from the previous number of candidate interactive summary information based on the summary category of identity feature information and interactive summary information.

[0298] In another possible implementation, the first historical interaction data determined by the first data determining unit further includes: the generation time period of at least one historical interaction information corresponding to the interaction summary information;

[0299] The abstract selection unit includes:

[0300] The importance determination subunit is used to determine the importance of the interactive summary information based on the time elapsed between the time period corresponding to the generation of the interactive summary information and the current time.

[0301] The summary selection sub-unit is used to determine at least one target interactive summary from at least one interactive summary based on identity feature information and the summary category and importance of interactive summary information.

[0302] In another possible implementation, the identity feature information determined by the identity determination unit includes: the interest coefficients of the target virtual object for different summary categories;

[0303] The abstract selection unit includes:

[0304] The score determination subunit is used to determine the interest score of the target virtual object for the interactive summary information based on the interest coefficient of the target virtual object for different summary categories and the summary category of the interactive summary information;

[0305] The sorting selection sub-unit is used to sort the interactive summary information from high to low according to the interest score corresponding to the interactive summary information, and to determine the top-ranked target interactive summary information from at least one interactive summary information.

[0306] In another possible implementation, the interaction summary information is generated based on at least one historical interaction message between the target virtual object and the user during the first historical time period;

[0307] The session information processing device also includes:

[0308] The second data determining unit is used to determine the second historical interaction data between the target virtual object and the user before determining the response information of the target virtual object to the question information. The second historical interaction data includes: at least one topic feature information. The topic feature information is the content feature of the topic described by the historical interaction information extracted from the historical interaction information between the target virtual object and the user in the second historical time period. The second historical time period is later than the first historical time period.

[0309] The topic selection unit is used to determine at least one target topic feature from at least one topic feature based on the topic category corresponding to the identity feature information and the topic feature information;

[0310] The response identifies the unit, including:

[0311] The first determining subunit is used to determine the response information of the target virtual object to the question information based on at least one target interaction summary information and at least one target topic feature information.

[0312] In another possible implementation, the interaction summary information is a summary information generated based on the topic feature information corresponding to at least one historical interaction between the target virtual object and the user.

[0313] The topic feature information corresponding to historical interaction information is the content feature of the topic described by the historical interaction information extracted from the historical interaction information.

[0314] Among them, the semantic similarity between the topic feature information of at least one historical interaction information corresponding to the interaction summary information exceeds a set threshold.

[0315] In yet another possible implementation, the session information processing apparatus further includes:

[0316] The first storage unit is used to store the question information as historical interaction information between the target virtual object and the user after the dialogue determination unit obtains the question information;

[0317] The second storage unit is used to store the response information as historical interaction information between the target virtual object and the user after the response determination unit determines the response information;

[0318] The topic extraction unit is used to extract topic feature information belonging to the topic category from historical interaction information based on at least one set topic category in response to the current meeting of the set topic extraction conditions, so as to obtain the topic feature information under at least one topic category corresponding to the historical interaction information.

[0319] The topic storage unit is used to store the topic feature information corresponding to the historical interaction information into the second historical interaction data between the target virtual object and the user;

[0320] The topic classification unit is used to cluster the topic feature information in the second historical interaction data based on the semantic similarity between topic feature information in response to the current meeting of the set topic classification conditions, to obtain at least one topic feature group, and the topic feature group includes at least one topic feature information.

[0321] The summary generation unit is used to determine the summary information corresponding to each topic feature information in each topic feature group using the summary generation model, and to obtain the interactive summary information of the topic feature group.

[0322] The summary storage unit is used to store the interaction summary information into the first historical interaction data between the target virtual object and the user.

[0323] In yet another possible implementation, the session information processing apparatus further includes:

[0324] The associated object determination unit is used to determine at least one reference virtual object that is associated with the target virtual object before the response determination unit determines the response information of the target virtual object to the question information. The reference virtual object and the target virtual object are used to simulate different objects in the same media scene.

[0325] The third data determination unit is used to determine the third historical interaction data between the reference virtual object and the user. The third historical interaction data includes: at least one reference summary information and the summary category corresponding to the reference summary information. The reference summary information is a summary information generated based on at least one historical interaction information between the reference virtual object and the user.

[0326] A reference determination unit is used to determine at least one target reference summary from at least one reference summary based on the summary category of identity feature information and reference summary information;

[0327] The response determination unit includes: a second determination subunit, used to determine the response information of the target virtual object to the question information based on at least one target interaction summary information and at least one reference summary information.

[0328] This application also provides an electronic device in its embodiments. For example... Figure 9 As shown, a schematic diagram of the composition structure of the electronic device is presented. The electronic device includes at least one processor 901 and a memory 902 connected to the processor.

[0329] This memory 902 is used to store computer programs;

[0330] The processor 901 is used to execute the computer program so that the electronic device can implement the session information processing method as described in any of the above embodiments.

[0331] Understandably, the electronic device may also include: a communication unit 903 for obtaining user-inputted question information in the interactive interface.

[0332] In addition, the electronic device may also have an input unit 904 such as a keyboard or mouse.

[0333] Of course, the electronic device can also have more than Figure 9 There are no restrictions on the number of more or fewer components.

[0334] This application also provides a computer program product, including computer-readable instructions, which, when executed on an electronic device, cause the electronic device to implement any of the session information processing methods provided in this application.

[0335] This application also provides a computer-readable storage medium that carries one or more computer programs. When the one or more computer programs are executed by an electronic device, the electronic device can implement any of the session information processing methods provided in this application.

[0336] It should also be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. In addition, in the device embodiment drawings provided in this application, the connection relationship between modules indicates that they have a communication connection, which can be implemented as one or more communication buses or signal lines.

[0337] Through the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented by means of software plus necessary general-purpose hardware, or it can be implemented by special-purpose hardware including application-specific integrated circuits, special-purpose CPUs, special-purpose memory, special-purpose components, etc. Generally, any function performed by a computer program can be easily implemented by corresponding hardware, and the specific hardware structure used to implement the same function can also be diverse, such as analog circuits, digital circuits, or special-purpose circuits. However, for this application, software program implementation is more often the preferred implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a readable storage medium, such as a computer floppy disk, USB flash drive, mobile hard disk, ROM, RAM, magnetic disk, or optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, training equipment, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0338] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product.

[0339] The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions may be transmitted from one website, computer, training device, or data center to another website, computer, training device, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium may be any available medium that a computer can store or a data storage device such as a training device or data center that integrates one or more available media. The available media may be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media (e.g., solid-state drives (SSDs)).

Claims

1. A method for processing conversational information, characterized in that, include: In response to a user entering a question in the conversational interaction interface, a target virtual object associated with the conversational interaction interface is determined. The target virtual object is an object in the conversational interaction interface used to simulate a conversational interaction between the conversational participant and the user. Determine the first historical interaction data between the target virtual object and the user. The first historical interaction data includes: at least one interaction summary information and a summary category corresponding to the interaction summary information. The interaction summary information is a summary information generated based on at least one historical interaction information between the target virtual object and the user. Obtain the identity feature information pre-configured for the target virtual object; Based on the identity feature information and the summary category of the interaction summary information, at least one target interaction summary information is determined from the at least one interaction summary information; Based on the at least one target interaction summary information, determine the response information of the target virtual object to the question information.

2. The session information processing method according to claim 1, characterized in that, Before determining at least one target interaction summary, the following is also included: From the at least one interactive summary information, determine the top target number of candidate interactive summary information that has the highest semantic similarity to the question information; The step of determining at least one target interaction summary from the at least one interaction summary based on the summary category of the identity feature information and the interaction summary information includes: Based on the identity feature information and the summary category of the interaction summary information, at least one target interaction summary information is determined from the previous target number of candidate interaction summary information.

3. The session information processing method according to claim 1, characterized in that, The first historical interaction data further includes: the time period during which the at least one historical interaction information was generated, corresponding to the interaction summary information; The step of determining at least one target interaction summary from the at least one interaction summary based on the summary category of the identity feature information and the interaction summary information includes: The importance of the interactive summary information is determined based on the time elapsed between the time period corresponding to the generation of the interactive summary information and the current time. Based on the identity feature information and the summary category and importance of the interaction summary information, at least one target interaction summary information is determined from the at least one interaction summary information.

4. The session information processing method according to claim 1, characterized in that, The identity feature information includes: the interest coefficient of the target virtual object for different summary categories; The step of determining at least one target interaction summary from the at least one interaction summary based on the summary category of the identity feature information and the interaction summary information includes: Based on the interest coefficients of the target virtual object for different summary categories and the summary category of the interactive summary information, the interest score of the target virtual object for the interactive summary information is determined; Based on the interest scores corresponding to the interactive summary information, sorted from high to low, determine at least one target interactive summary information that ranks first among the at least one interactive summary information.

5. The session information processing method according to claim 1, characterized in that, The interaction summary information is generated based on at least one historical interaction message between the target virtual object and the user during a first historical time period; Before determining the response information of the target virtual object to the question, the process also includes: Determine the second historical interaction data between the target virtual object and the user. The second historical interaction data includes at least one topic feature information. The topic feature information is the content feature of the topic described by the historical interaction information extracted from the historical interaction information between the target virtual object and the user in a second historical time period. The second historical time period is later than the first historical time period. Based on the topic category corresponding to the identity feature information and the topic feature information, at least one target topic feature information is determined from the at least one topic feature information; The step of determining the response information of the target virtual object to the question information based on the at least one target interaction summary information includes: Based on the at least one target interaction summary information and the at least one target topic feature information, determine the response information of the target virtual object to the question information.

6. The session information processing method according to claim 1, characterized in that, The interaction summary information is a summary information generated based on the topic feature information corresponding to at least one historical interaction between the target virtual object and the user; The topic feature information corresponding to the historical interaction information is the content feature of the topic described by the historical interaction information extracted from the historical interaction information. Among them, the semantic similarity between the topic feature information of at least one historical interaction information corresponding to the interaction summary information exceeds a set threshold.

7. The session information processing method according to claim 1 or 6, characterized in that, After obtaining the question information, the method further includes: storing the question information as historical interaction information between the target virtual object and the user; After determining the response information, the process also includes: The response information is stored as historical interaction information between the target virtual object and the user; In response to the current satisfaction of the set topic extraction conditions, based on at least one set topic category, topic feature information belonging to the topic category is extracted from the historical interaction information to obtain topic feature information under at least one topic category corresponding to the historical interaction information. The topic feature information corresponding to the historical interaction information is stored in the second historical interaction data between the target virtual object and the user; In response to the current satisfaction of the set topic classification conditions, based on the semantic similarity between topic feature information, the topic feature information in the second historical interaction data is clustered to obtain at least one topic feature group, and the topic feature group includes at least one topic feature information. For each topic feature group, the summary information corresponding to each topic feature information in the topic feature group is determined by the summary generation model to obtain the interactive summary information of the topic feature group. The interaction summary information is stored in the first historical interaction data between the target virtual object and the user.

8. The session information processing method according to claim 1, characterized in that, Before determining the response information of the target virtual object to the question, the process also includes: Identify at least one reference virtual object that is associated with the target virtual object, wherein the reference virtual object and the target virtual object are used to simulate different objects in the same media scene; Determine the third historical interaction data between the reference virtual object and the user, the third historical interaction data including: at least one reference summary information and the summary category corresponding to the reference summary information, the reference summary information being summary information generated based on at least one historical interaction information between the reference virtual object and the user; Based on the identity feature information and the summary category of the reference summary information, at least one target reference summary information is determined from the at least one reference summary information; The step of determining the response information of the target virtual object to the question information based on the at least one target interaction summary information includes: Based on the at least one target interaction summary information and the at least one reference summary information, the response information of the target virtual object to the question information is determined.

9. A conversation information processing device, characterized in that, include: The dialogue determination unit is used to determine the target virtual object associated with the dialogue interaction interface in response to the user inputting a question on the dialogue interaction interface. The target virtual object is the object in the dialogue interaction interface used to simulate the dialogue interaction between the dialogue participant and the user. A first data determining unit is used to determine the first historical interaction data between the target virtual object and the user. The first historical interaction data includes: at least one interaction summary information and a summary category corresponding to the interaction summary information. The interaction summary information is a summary information generated based on at least one historical interaction information between the target virtual object and the user. An identity determination unit is used to obtain identity feature information pre-configured for the target virtual object; A summary selection unit is used to determine at least one target interactive summary from the at least one interactive summary based on the identity feature information and the summary category of the interactive summary information; The response determination unit is used to determine the response information of the target virtual object to the question information based on the at least one target interaction summary information.

10. An electronic device, characterized in that, It includes at least one processor and a memory connected to the processor, wherein: The memory is used to store computer programs; The processor is used to execute the computer program to enable the electronic device to implement the session information processing method as described in any one of claims 1 to 8.

Images