Automatic chat reply method and device in game scene, equipment and storage medium

Abstract

The embodiment of the invention discloses an automatic chat reply method and device in a game scene, equipment and a storage medium. The chat automatic reply method in the game scene provided by the embodiment of the invention comprises the following steps: acquiring chat content in a game chat channel, and carrying out contextual information integration processing based on the chat content to obtain similar chat aggregation data; performing semantic analysis on the same kind of chat aggregated data to determine a target problem; retrieving the target question based on a set knowledge vector library to obtain corresponding matched knowledge entries, and generating corresponding reply contents according to the knowledge entries; and the reply content is published to the game chat channel through the set virtual user role. According to the scheme, the technical problem that the question answering efficiency of the user in the game chat channel is low can be solved, and the question answering efficiency of the user in the game chat channel is remarkably improved through automatic identification and automatic reply answering of the target question in the chat content, so that the overall use experience of the user is remarkably improved.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a method, apparatus, device, and storage medium for automatic chat replies in a game scenario. Background Technology

[0002] To provide a healthy gaming ecosystem, gaming platforms typically offer chat channels within game scenarios for users to interact and communicate. This serves two purposes: firstly, it increases users' willingness to form social teams, thereby enhancing user engagement; secondly, it allows for a better understanding of user feedback, facilitating subsequent game optimization. In actual user chat, users frequently raise questions about gameplay, events, and system functions in the chat channel, which are usually answered by experienced players or internal staff. This method of answering questions is difficult to comprehensively and quickly address all user concerns, resulting in relatively low efficiency in resolving user issues in the game chat channel, thus impacting the user experience. Summary of the Invention This application provides a method, apparatus, device, and storage medium for automatic chat replies in game scenarios, which can solve the technical problem of low efficiency in answering user questions in game chat channels. By automatically identifying and automatically answering target questions in chat content, the efficiency of answering user questions in game chat channels is significantly improved, thereby significantly improving the overall user experience.

[0003] In a first aspect, embodiments of this application provide a method for automatic chat reply in a game scenario, including: Get the chat content from the game chat channel, and integrate and process the context information based on the chat content to obtain aggregated chat data of the same type; Semantic analysis of similar chat aggregation data is used to identify the target problem; The target question is retrieved based on the set knowledge vector base, corresponding matching knowledge entries are obtained, and corresponding response content is generated based on the knowledge entries; The reply content is published to the game chat channel by setting up a virtual user role.

[0004] Furthermore, based on the contextual information integration of chat content, similar chat aggregation data is obtained, including: Determine a preset number of target chat messages based on the chat content and the settings of the sliding window; The first target data is obtained by merging the messages of the same target user based on the target chat content; The second target data is obtained by performing cross-user interaction recognition processing on the target chat content to identify the corresponding target user. Integrate the primary target data and the secondary target data to obtain similar chat aggregation data.

[0005] Furthermore, based on the target chat content, cross-user interaction identification processing is performed on the corresponding target user to obtain the second target data, including: Identify target users based on the target chat content, and determine interactive information based on the target users' messages; Based on the interaction information, cross-user interaction identification processing is performed to determine the content of the interactive users' statements, and the content of the interactive users' statements is integrated to obtain the second target data.

[0006] Furthermore, semantic analysis is performed on similar chat aggregation data to identify the target problem, including: Semantic vectors are obtained by extracting features from aggregated chat data of the same type, and text clusters are obtained by clustering based on semantic vectors; Keyword and entity extraction is performed on text clusters to obtain core entities, keywords, and game elements, and text classification is performed on text clusters to obtain core intent tags. Text clusters, core entities, keywords, and core intent labels are input into a pre-defined large language model to generate the corresponding target question.

[0007] Furthermore, obtain the chat content in the game chat channel, including: Get the initial chat content from the game chat channel; The initial chat content is processed for risk control identification to identify sensitive and invalid messages; Sensitive content is processed according to its sensitivity level, and invalid content is filtered to obtain the target chat content.

[0008] Furthermore, by setting up virtual user characters, replies can be posted to the game chat channel, including: Acquire in-game behavior data and chat feature data of target users in the game chat channel; Construct user attribute feature tags and user behavior status tags based on in-game behavior data and chat feature data; The target response strategy is determined by matching response strategies based on user attribute feature tags and user behavior status tags. The target response content is obtained by modifying the response content according to the target response strategy; The target's reply is published to the game chat channel by setting up a virtual user role.

[0009] Furthermore, by setting up virtual user characters, replies can be posted to the game chat channel, including: Identify the target game character in the current discussion based on aggregated chat data from similar chat sources; Personalized responses are obtained by rewriting the responses based on the target game character; Retrieve the target virtual user character based on the target game character and the corresponding virtual game personality; Personalized replies are posted to the game chat channel using the target virtual user avatar.

[0010] In a second aspect, embodiments of this application provide an automatic chat reply device for a game scenario, comprising: The chat content acquisition module is used to acquire chat content from the game's chat channel; The chat data aggregation module is used to integrate and process contextual information based on chat content to obtain aggregated chat data of the same type. The target question identification module is used to perform semantic analysis on similar chat aggregation data to determine the target question. The knowledge entry retrieval module is used to retrieve the target question based on the set knowledge vector library and obtain the corresponding matching knowledge entries; The response content determination module is used to generate corresponding response content based on the knowledge entries; The reply posting module is used to post reply content to the game chat channel through the set virtual user role.

[0011] In a third aspect, embodiments of this application provide an automatic chat reply device for game scenarios, comprising: Memory and one or more processors; Memory, used to store one or more programs; When one or more programs are executed by one or more processors, the one or more processors implement the automatic chat reply method as in the game scenario of the first aspect.

[0012] In a fourth aspect, embodiments of this application provide a storage medium for storing computer-executable instructions, which, when executed by a computer processor, are used to perform an automatic chat reply method in a game scenario as described in the first aspect.

[0013] This application embodiment acquires chat content from a game chat channel during gameplay, integrates contextual information based on the chat content to obtain aggregated chat data of similar types, performs semantic analysis on the aggregated chat data to determine the target question, retrieves matching knowledge entries based on a set knowledge vector library, generates corresponding response content based on the knowledge entries, and publishes the response content to the game chat channel through a set virtual user role. By employing the above technical means, the automatic identification and automatic response to the target question in the chat content can avoid the technical problem of low user question-answering efficiency in game chat channels, significantly improving the automation and intelligence of user question answering in game chat channels, thereby significantly improving the efficiency of user question answering in game chat channels and ultimately significantly improving the overall user experience.

[0014] The beneficial effects of the chat auto-reply device, chat auto-reply equipment, and storage medium in the game scenario described above can be referenced from the beneficial effects of the chat auto-reply method in the game scenario. Attached Figure Description

[0015] Figure 1 This is a flowchart illustrating an automatic chat reply method in a game scenario provided in an embodiment of this application; Figure 2 This is a flowchart of another automatic chat reply method in a game scenario provided by an embodiment of this application; Figure 3 This is a flowchart of another automatic chat reply method in a game scenario provided by an embodiment of this application; Figure 4 This is a flowchart of another automatic chat reply method in a game scenario provided by an embodiment of this application; Figure 5 This is a flowchart of another automatic chat reply method in a game scenario provided by an embodiment of this application; Figure 6 This is a schematic diagram of the structure of an automatic chat reply device in a game scene provided in an embodiment of this application; Figure 7 This is a schematic diagram of the structure of an automatic chat reply device in a game scenario provided in an embodiment of this application. Detailed Implementation

[0016] To make the objectives, technical solutions, and advantages of this application clearer, specific embodiments of this application will be described in further detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are merely for explaining this application and not for limiting it. It should also be noted that, for ease of description, only the parts relevant to this application are shown in the drawings, not all of them. Before discussing exemplary embodiments in more detail, it should be mentioned that some exemplary embodiments are described as processes or methods depicted as flowcharts. Although the flowcharts describe operations (or steps) as sequential processes, many of these operations can be performed in parallel, concurrently, or simultaneously. Furthermore, the order of the operations can be rearranged. The process can be terminated when its operation is completed, but may also have additional steps not included in the drawings. The process can correspond to a method, function, procedure, subroutine, subprogram, etc.

[0017] In actual user chat, users frequently raise questions about gameplay, events, and system functions in the chat channel, which are usually answered by experienced players or internal staff. This method of answering questions is difficult to comprehensively and quickly address all user concerns, resulting in relatively low efficiency in resolving user issues in the game chat channel, thus impacting the user experience.

[0018] Based on this, this application provides a method, apparatus, device, and storage medium for automatic chat replies in a game scenario. The aim is to acquire chat content from a game chat channel during chat, integrate contextual information based on the chat content to obtain aggregated chat data of similar types, perform semantic analysis on the aggregated chat data to determine the target question, retrieve matching knowledge entries based on a set knowledge vector library, generate corresponding reply content based on the knowledge entries, and publish the reply content to the game chat channel through a set virtual user character. By employing the above technical means, automatic identification and automatic reply to the target question in the chat content significantly improves the automation and intelligence of answering user questions in the game chat channel compared to existing methods that rely on experienced players or internal staff for answers. This significantly improves the efficiency of answering user questions in the game chat channel, thereby significantly enhancing the overall user experience. Furthermore, by publishing the reply content to the game chat channel through a set virtual user character to answer user questions, personalized interaction is achieved, reducing the sense of distance and mechanical feeling, strengthening social belonging and immersion, and further enhancing the overall user experience.

[0019] Figure 1A flowchart of an automatic chat reply method in a game scene provided in this application embodiment is given. The automatic chat reply method in the game scene provided in this embodiment can be executed by an automatic chat reply device in the game scene. The automatic chat reply device in the game scene can be implemented by software and / or hardware. The automatic chat reply device in the game scene can be composed of two or more physical entities, or it can be composed of a single physical entity. Generally speaking, the automatic chat reply device in the game scene can be a computer device.

[0020] The following description uses a computer device as the primary example to illustrate the automatic chat reply method in a game scenario. (Refer to...) Figure 1 The automatic chat reply methods in this game scenario specifically include: S11. Obtain the chat content in the game chat channel, and perform contextual information integration processing based on the chat content to obtain similar chat aggregate data.

[0021] In gaming scenarios, users frequently ask questions about gameplay, events, and system functions in game chat channels. To understand these questions, the chat content can be retrieved from these channels. Based on the retrieved chat content, the target user's question and other user interactions can be identified. For example, game chat channels can be all public chat channels within the game, such as world channels, guild channels, and team channels, or private chat channels. A data channel can be established through the game server's real-time communication interface to collect chat content from these game chat channels. The retrieved chat content is then processed to integrate contextual information, resulting in aggregated chat data of similar types. For example, consecutive messages from the same target user within a short period can be timestamped and pieced together to create a complete dialogue, generating a user-level dialogue chain. For other users, multi-user interaction chains can be constructed based on reply or mention relationships (such as "@" mentions). Integrating the user-level dialogue chain and the multi-user interaction chain yields the corresponding aggregated chat data of similar types.

[0022] In one embodiment, the acquired chat content may include not only text but also images (such as game screenshots) and voice messages. Therefore, after acquiring the initial chat content from the game chat channel, the images in the initial chat content undergo interface element recognition and optical character recognition processing to obtain the first text content. The voice messages in the initial chat content undergo speech-to-text processing to obtain the second text content. The initial text content, the aforementioned first text content, and the second text content are integrated to obtain the final first target chat content. Subsequent question identification and response processing can be based on the first target chat content. By parsing and processing multimodal questions in the chat content, cross-modal semantic fusion is achieved, allowing users to input questions in multiple ways, significantly improving the user experience.

[0023] In one embodiment, core chat information can be determined based on the chat content, such as core chat information including interrogative words like "how" or "where." The context window of each core chat message is extracted, and the chat content within the context window is aggregated to obtain similar chat aggregation data. For example, semantic vectors can be obtained by extracting semantic features from the chat content in the context window. Clustering is then performed on these semantic vectors to group semantic vectors with a cosine similarity greater than a preset threshold into a unified cluster. High-frequency keywords are extracted from each cluster to generate cluster topic tags. Similar chat aggregation data is generated based on cluster identifiers, high-frequency keywords, and cluster topic tags for subsequent use.

[0024] The above-mentioned method integrates and processes the contextual information of chat content in the game chat channel to obtain similar chat aggregation data. This transforms fragmented chats into structured similar chat aggregation data, providing a precise data foundation for subsequent user question identification and improving the accuracy of subsequent user question identification. Furthermore, by using similar chat aggregation data, the workload of repetitive analysis can be reduced, the speed of subsequent user question identification can be increased, and the overall efficiency of automatic chat response can be improved.

[0025] S12. Perform semantic analysis on similar chat aggregation data to determine the target problem.

[0026] After obtaining the aggregated chat data of similar types in S11, semantic analysis is performed on the aggregated chat data to determine the target problem. For example, feature extraction can be performed on the aggregated chat data to obtain semantic vectors, and clustering can be performed based on these semantic vectors to obtain text clusters. For instance, a pre-trained language model can be used to convert each text content in the aggregated chat data into a fixed-dimensional semantic vector (e.g., 384-dimensional). Contextual relationships need to be preserved during semantic vector generation to ensure semantic integrity. After obtaining the semantic vectors, density clustering or hierarchical clustering algorithms can be used to cluster the texts according to the similarity of their semantic vectors to obtain text clusters. For example, the texts "How to claim the rewards for the new activity?" and "Where to claim the rewards for the Spring Festival activity?" are considered semantically similar after semantic vector conversion, and if the cosine similarity of their semantic vectors is ≥0.85 (preset threshold), they are considered semantically similar and are grouped into a single text cluster. Keyword extraction and entity extraction are then performed on the obtained text clusters to obtain core entities, keywords, and game elements, and text classification is performed on the text clusters to obtain core intent labels. For example, a corresponding keyword extraction algorithm can be used to extract high-frequency words with a pre-defined weight (e.g., the top 10) from the text within a text cluster as keywords. These keywords are then identified as the core theme of the text cluster, for example, "activity reward extraction problem". After determining the core theme, the topic consistency score of the text within the text cluster can be calculated using an LDA topic model. If the score is lower than a preset threshold, the text cluster is split, meaning that the original text cluster may contain multiple sub-themes, requiring further splitting. If the score exceeds the preset threshold, the core theme is confirmed to be accurate. During entity extraction, key entities can be extracted from the text of the text cluster using an entity recognition model. Key entities include core entities and game elements (entities). For example, extracted game elements (entities) include activity names (e.g., "Spring Festival activity"), functional modules (e.g., "prize platform" or "NPC"), and items (e.g., "gold ingots" or "skins"); core entities include actions (e.g., "receive", "feedback", or "query") and statuses (e.g., "not received" or "overdue"). When performing text classification, the core intent of text within a text cluster can be determined based on a pre-defined multi-classification model to identify the corresponding core intent label. For example, the preset intent labels for the multi-classification model include inquiry-based (e.g., "inquiring about the claim path" or "querying rules"), feedback-based (e.g., "reward not received" or "claim failed"), and suggestion-based (e.g., "hoping to extend the claim time"). If a text cluster contains multiple intents (e.g., 30% of the text's intent is "inquiring about the path," and 70% of the intent is "feedback not received"), the intent with the highest percentage is taken as the core intent (e.g., "feedback not received"), and the corresponding core intent label is assigned to the text cluster.After identifying the text clusters, core entities, keywords, and core intent labels, these elements are input into a pre-defined large language model for question generation to obtain the corresponding target question. For example, the large language model can use the input text clusters, core entities, keywords, and core intent labels to call the prompt word project to generate a general target question. For instance, the prompt word template is: Task: Generate a concise and accurate core question based on the following features: - Topic: {cluster topic, such as "Spring Festival Activity Rewards"} - Core Entities: {Entity list, such as "Spring Festival Activities, Rewards, Claimed, Not Yet Received"} -Core Intent: {Intent Tag, such as "Feedback Reward Not Received"} - Typical text: {Three representative texts within the cluster, such as "Did not receive the Spring Festival event rewards" and "Received the event rewards but they are not in my inventory"} Requirements: The question must contain the core entity, clearly state the intent, be ≤25 characters long, and cover more than 80% of the text within the cluster.

[0027] Therefore, based on the above prompt template, the target question output is "What should I do if the Spring Festival event rewards have not been credited to my account?".

[0028] As described above, by performing semantic analysis on the aggregated chat data of similar types in the game chat channel to determine the target question, the automatic identification of user questions in the game chat channel is achieved without manual screening and analysis, which greatly improves the automation level of user question identification, thereby increasing the speed of user question identification and the speed of subsequent automatic replies based on user questions, and thus improving the overall efficiency of automatic chat replies and significantly enhancing the overall user experience.

[0029] S13. Based on the set knowledge vector base, retrieve the target question, obtain the corresponding matching knowledge entries, and generate the corresponding response content according to the knowledge entries.

[0030] After determining the target question in S12, the corresponding matching knowledge entries can be retrieved based on the set knowledge vector library. For example, the target question can be vectorized to obtain a question semantic vector. The set knowledge vector library is then searched based on the question semantic vector. The cosine similarity between the question semantic vector and the knowledge vector corresponding to each knowledge entry in the knowledge vector library is calculated, and initial knowledge entries with a cosine similarity greater than a preset threshold are selected. Each knowledge entry includes an associated entity field and an applicable scenario field. The question semantic vector is matched with the associated entity field and applicable scenario field of the initial knowledge entries to select target knowledge entries that match the target question. For example, when matching associated entity fields, the overlap between the associated entity fields of the initial knowledge entries and the core entities of the target question (such as "Spring Festival activities" and "rewards not received") is checked, and target knowledge entries with an overlap ≥ 80% (preset threshold) are prioritized. When matching applicable scenario fields, if the target question is associated with a specific user scenario (such as "new player question"), target knowledge entries with the applicable scenario field "new player" are prioritized. After identifying the target knowledge items, the content of these items is integrated and refined to obtain the response content. For example, essential information (unmodifiable core content) and adjustable information (content with optimizable wording) are extracted from the matched target knowledge items. Essential information includes: "Reward distribution time = within 24 hours," "Feedback path = Settings - Customer Service Center," and "Processing time = 1 business day." Adjustable information includes: connecting words (such as "If not received, please proceed") and modifiers (such as "please" or "suggest"). After determining the essential and adjustable information, the style of the response content can be adapted based on these elements. For example, user characteristic adaptation or game worldview adaptation can be performed to generate response content with a corresponding style. When performing user characteristic adaptation, the wording can be adjusted by combining target user attributes (such as level, VIP level, and historical interaction style). For example, for new players, basic guidance details need to be added (such as "The 'Settings' button is the gear icon in the upper right corner of the game's main interface"); for advanced players, the descriptions need to be simplified and the core steps highlighted (such as "Rewards will arrive within 24 hours; if not, contact customer service"). When adapting to the game's world view, role-playing style responses can be used, transforming standardized answers into language appropriate to the character's identity. For example, using the "Jianghu Messenger" character from the martial arts world view, the response could be: "Don't worry, young hero! The Spring Festival event rewards will arrive in your inventory within 24 hours. If you haven't received them, please go to 'Settings' - 'Customer Service Station' and submit your character name and proof of participation. The manager will check within one day."

[0031] As described above, by retrieving the target question from the established knowledge vector database, corresponding matching knowledge entries are obtained. Compared to traditional keyword retrieval, this embodiment captures deep semantic relationships in the text through cosine similarity calculation of semantic vectors, greatly improving the accuracy and efficiency of knowledge entry matching. Furthermore, corresponding response content is generated based on the matched knowledge entries, and the response content can be traced back to specific knowledge entries, improving the reliability of knowledge tracing. In addition, response content can be automatically generated based on the target question without manual determination of the response content, greatly improving the automation and intelligence of chat responses, and thus significantly improving the efficiency of response content generation.

[0032] S14. Post the reply to the game chat channel using the set virtual user role.

[0033] After generating the reply content in S13, the reply content can be published to the game chat channel using a set virtual user role to address the target user's specific question. This virtual user role can be a virtual customer service role, such as a "game customer service assistant," a "dungeon strategy expert," or an "event benefits officer," or it can be a corresponding virtual game personality. The language style is determined based on the set virtual user role; for example, a "friendly and professional" style for a "game customer service assistant" and a "concise and efficient" style for a "dungeon strategy expert." Based on the determined language style, the aforementioned reply content is rewritten using the corresponding virtual user role's exclusive script template to obtain the target reply content that matches the virtual user role's tone and style. The target reply content is then precisely pushed to the target scenario, i.e., the corresponding game chat channel, through the game chat channel interface. A precise reply can be given by "@" the target user or by referencing chat content related to the target question.

[0034] As mentioned above, replying through virtual user avatars can create a visual focus in the information-saturated game chat channels, increasing the view rate of replies by target users. Replies using virtual user avatars also allow for personalized expression, reducing the mechanical feel and enhancing user immersion and emotional connection, thereby improving the overall user experience in the game chat channels. Based on the official attributes of the virtual user avatars, users can identify the official source of the replies, increasing user trust in the content compared to replies from ordinary players.

[0035] The above-described method involves acquiring chat content from the game chat channel, integrating contextual information based on the chat content to obtain aggregated chat data of similar types, performing semantic analysis on this aggregated data to determine the target question, retrieving matching knowledge entries based on a set knowledge vector library, generating corresponding response content based on the knowledge entries, and publishing the response content to the game chat channel through a set virtual user character. By employing the above technical means, the automatic identification and automatic response to the target question in the chat content significantly improves the automation and intelligence of answering user questions in the game chat channel compared to existing methods that rely on experienced players or internal server personnel. This significantly improves the efficiency of answering user questions in the game chat channel, thereby significantly enhancing the overall user experience. Furthermore, the use of a set virtual user character to publish the response content to the game chat channel to answer user questions enables personalized interaction, reduces the sense of distance and mechanicalness, strengthens social belonging and immersion, and further enhances the overall user experience.

[0036] Figure 2 This is a flowchart of another chat auto-reply method in a game scenario provided in this application embodiment, see below. Figure 2 The automatic chat reply methods in this game scenario specifically include: S111, Get the initial chat content in the game chat channel.

[0037] When acquiring chat content from the game chat channel in S11, the initial chat content can be obtained first. This initial chat content includes initial text, images, and voice messages. After acquiring the initial chat content, the images in the initial chat content undergo interface element recognition and optical character recognition (OCR) processing to obtain the first text content. The voice messages in the initial chat content undergo speech-to-text processing to obtain the second text content. Integrating the initial text content, the first text content, and the second text content yields the final target chat content. Subsequent actions, such as risk control identification, problem determination, and automatic reply processing, can be performed based on this target chat content.

[0038] S112. Perform risk control identification processing on the initial chat content to identify sensitive and invalid messages.

[0039] After obtaining the initial chat content or the first target chat content in step S111, risk control identification processing is performed on the initial chat content or the first target chat content to identify sensitive and invalid statements. During risk control identification processing, the initial chat content or the first target chat content can be matched with a set sensitive word database to determine the corresponding sensitive statements. Alternatively, the initial chat content or the first target chat content can be input into a sensitive semantic classification model for semantic recognition processing to identify the corresponding sensitive statements. During risk control identification processing, corresponding invalid statements can be identified using preset rules. For example, preset rules include statements whose plain text length is ≤ a preset number and does not contain game terms; in this case, the corresponding statements are marked as invalid. For example, plain text length ≤ 2 characters (such as "oh" or "um") and does not contain game terms (such as "6" being invalid, "6 stars" being valid). The preset rules also include calculating the text hash value using the SimHash algorithm (similarity hashing algorithm), comparing it with a preset number (e.g., 5) of historical messages from the same user, and marking messages with a repetition rate ≥ a preset threshold (e.g., 90%) as "invalid for spamming". The preset rules also include calculating text information entropy; if the entropy value is < a preset threshold (e.g., 1.5) (e.g., "ahhhhhh" or "hehehehe"), the corresponding message is marked as "low-information invalid". The preset rules also include marking messages with an emoticon or special symbol (e.g., "★, ■") percentage ≥ a preset threshold (e.g., 80%) as invalid.

[0040] S113. Sensitive speech content is processed according to its corresponding sensitivity level, and invalid speech content is filtered to obtain the target chat content.

[0041] After identifying sensitive and invalid chat content in step S112, sensitive chat content is processed according to its sensitivity level. For example, the sensitivity level of a chat message is determined: if the sensitivity level is mild, the corresponding sensitive message is replaced with a special symbol; if the sensitivity level is moderate, the corresponding user is temporarily muted; if the sensitivity level is severe, the corresponding user is permanently muted and their account is frozen. Invalid chat content is filtered out. The process of processing sensitive chat content according to its sensitivity level and filtering invalid chat content yields the target chat content.

[0042] The above-mentioned risk control and identification process, which identifies sensitive content and applies corresponding risk control measures to sensitive content obtained from game chat channels, can effectively block sensitive and illegal content, purify the game chat environment, and reduce user complaints about inappropriate content. Furthermore, by identifying and filtering invalid content, spam and meaningless text are eliminated, thereby increasing the information density of the target chat content. This helps in the subsequent identification of corresponding target issues based on the target chat content, improving the accuracy and efficiency of target issue identification.

[0043] Figure 3 This is a flowchart illustrating another method for automatic chat reply in a game scenario provided in this application embodiment, see below. Figure 3 The automatic chat reply methods in this game scenario specifically include: S114. Determine a preset number of target chat contents based on the chat content and the set sliding window.

[0044] After obtaining the chat content from the game chat channel in S11, a preset number of target chat messages can be determined based on the chat content and the set sliding window. For example, a sliding window mechanism can be used to filter out valid and relevant target chat messages from real-time or historical chat streams (i.e., chat content). The sliding window's parameter configuration can be preset, including the window type and a preset numerical threshold. The window type supports both time window and message count window modes, and the time parameter for the time window and the message count parameter for the message count window can be configured as needed. For example, the time parameter for the time window can be configured as "chat content within the last 5 minutes," and the message count parameter for the message count window can be configured as "the last 20 messages." The preset numerical threshold can be dynamically adjusted based on the activity level of the game chat channel. For example, the preset number is 30 messages for the world channel (high activity) and 20 messages for the guild channel (medium activity), ensuring sufficient context coverage and avoiding data overload.

[0045] When acquiring target chat content, only the valid target chat content that has undergone risk control and filtering processes can be retained from the chat content in the window, i.e., valid target chat content without sensitive or invalid statements. For specific risk control and filtering processes, refer to S111-S113 above.

[0046] S115. Merge the messages of the same target user based on the target chat content to obtain the first target data.

[0047] After determining the target chat content in S114, the first target data can be obtained by focusing on the continuous statements of a single target user within the target chat content and merging the derivative statements of the same target user. For example, the target chat content within the sliding window can be traversed, and all independent users can be identified as target users based on their user identifiers. When merging the first target data, time-related merging can be used. Specifically, for continuous statements from the same target user within a short time interval (e.g., 2 minutes), the statements are stitched together in timestamp order to form a complete dialogue, thus forming the first target data. For example: User A states 1: "How do I enter the level 70 dungeon?" (t=00:01); User A states 2: "Can't find the entrance location" (t=00:02); therefore, the merged first target data is: "How do I enter the level 70 dungeon? Can't find the entrance location." When merging the first target data, semantic-related merging can also be used. Specifically, if the interval between user statements exceeds a preset time (e.g., 2 minutes), but the content is semantically related, the cosine similarity of the semantic vectors can be calculated. If the cosine similarity of the semantic vectors of the two users is greater than a preset threshold (e.g., 0.7), then the content is considered semantically related. Semantically related statements are merged into a single statement, which is then combined into the primary target data for that user. For example: User A's statement 1: "What classes are needed for the dungeon?" (t=00:05); User A's statement 2: "Are there enough tanks? Or do we need a healer?" (t=00:08, semantic similarity 0.85); The merged primary target data is: "What classes are needed for the dungeon? Are there enough tanks? Or do we need a healer?".

[0048] Redundancy filtering is performed on the merged primary target data to remove duplicate expressions. For example, the core of "The dungeon is so hard" is retained, replacing "The dungeon is really hard." Key entities are also retained, such as "Level 70 dungeon," "class," and "tank." The resulting primary target data is then stored in a structured format, with each data entry associated with a user ID, recording the complete merged statement, core keywords, and time range. This structured storage of the primary target data facilitates direct retrieval later.

[0049] S116. Based on the target chat content, perform cross-user interaction recognition processing to obtain the second target data.

[0050] After determining the target chat content in S114, the target user can be identified based on the target chat content, and the interaction information can be determined based on the target user's speech content. Cross-user interaction identification processing is performed based on the interaction information to determine the speech content of the interacting users, and the speech content of the interacting users is integrated to obtain the second target data. When determining the interaction information, interaction signals can be identified in the target chat content. Interaction signals include explicit interaction signals and implicit interaction signals. Explicit interaction signals include reply interaction signals and mention interaction signals; for example, the reply interaction signal is "quote," and the mention interaction signal is "@." Implicit interaction signals can be identified through semantic similarity. For example, the cosine similarity (threshold ≥ 0.65) between user B's speech and target user A's historical speech is calculated. If a match is found, it is marked as an "implicit reply" (e.g., user A asks "dungeon strategy," and user B replies "the dungeon requires clearing mobs first"). The speech content of the corresponding interacting users is determined based on the identified explicit and implicit interaction signals. The speech content of the interacting users is integrated to obtain the second target data. The second target data is stored in a structured manner, recording the interaction entity, type, content, and associated user (i.e., the associated target user). By storing the second target data in a structured manner, it is easier to call it directly later.

[0051] S117. Integrate the first target data and the second target data to obtain similar chat aggregation data.

[0052] After determining the first target data in S115 and the second target data in S116, the first target data and the corresponding second target data are integrated to obtain similar chat aggregation data. For example, individual user statements (i.e., the first target data) and cross-user interaction data (i.e., the second target data) are clustered by topic to form a similar content dataset (i.e., similar chat aggregation data). During topic clustering, the core topics of the statements in the first and second target data can be extracted using an LDA topic model, such as "Level 70 Dungeon Guide" and "Team Requirements." Clustering algorithms can also be used to aggregate content with a topic similarity greater than a preset threshold (e.g., 0.75) based on semantic vectors to obtain clusters, generating cluster labels. Each cluster includes core user statements, related interactions, and statistical features. Core user statements are the combined statements of target users belonging to that cluster in the first target data; related interactions are the statements of cross-user interactions related to the topic of that cluster in the second target data; and statistical features include the number of users within the cluster, the number of interaction rounds, and high-frequency keywords. All clusters are integrated to obtain a dataset of similar content (i.e., aggregated chat data of the same type). The obtained aggregated chat data of the same type is then stored in a structured manner for easy access later.

[0053] As described above, acquiring target chat content through a sliding window ensures that related content is not fragmented, avoiding semantic misjudgments caused by fragmentation. This preserves the integrity of the context, improving the accuracy of semantic understanding and consequently enhancing the accuracy of subsequent aggregation of similar chat data. Cross-user interaction recognition avoids information gaps caused by isolated analysis of single user statements; and aggregating similar chat data reduces repetitive analysis workload, speeds up the identification of subsequent user questions, and ultimately improves the overall efficiency of automatic chat responses.

[0054] Figure 4 This is a flowchart of another chat auto-reply method in a game scenario provided in this application embodiment, see below. Figure 4 The automatic chat reply methods in this game scenario specifically include: S141. Obtain the target user's in-game behavior data and chat feature data in the game chat channel.

[0055] After generating the reply content in S13, and before publishing it, the target user's in-game behavior data and chat characteristic data in the game chat channel can be obtained. This facilitates determining the style of the reply content based on the in-game behavior data and chat characteristic data. For example, the target user's in-game behavior data can be obtained in real time through the game server log synchronization interface and updated incrementally on an hourly basis. The in-game behavioral data includes basic attributes and core behaviors. Basic attributes include user ID, level, VIP level, registration duration, and average daily online time, which can be extracted from the user center database on the game server. Core behaviors include combat data, quest data, item data, and social data. Combat data includes the number of times dungeons have been entered, the completion rate, the number of failures (e.g., "failed 3 times in a level 70 dungeon"), and the most frequently used class or lineup in the past few days (e.g., 7 days). Quest data includes the progress of main or side quests (completion rate and key points) and the frequency of participation in daily quests (e.g., "failed to complete daily quests for 3 consecutive days"). Item data includes the type of items in the inventory (offensive / defensive ratio) and the frequency of consumption (e.g., "average daily consumption of healing potions ≥ 5"). Social data includes the number of times the player has teamed up, guild activity (frequency of speaking / donating), and the number of interactions with friends. For example, chat logs of target users in game chat channels can be monitored through the chat system API to extract chat characteristic data. This chat characteristic data includes content feature data and interaction feature data. Content feature data includes chat question texts from the past few days (e.g., "How to clear the dungeon" and "Where to claim the rewards?"), question frequency (average number of questions per day), and topic preferences (Top 3 topics extracted using LDA topic model, such as "Dungeon strategy" and "Event rewards"). Interaction feature data includes response rate (the proportion of responses to other people's questions), number of times actively helping others (e.g., answering other players' questions ≥ 3 times / day), and tone style (identifying "humble" or "direct" through sentiment analysis model).

[0056] S142. Construct user attribute feature tags and user behavior status tags based on in-game behavior data and chat feature data.

[0057] After obtaining the target user's in-game behavior data and chat feature data in S141, user attribute feature tags and user behavior status are constructed based on the in-game behavior data and chat feature data. This facilitates the subsequent determination of the style of the reply content based on the user attribute feature tags and user behavior status. For example, the in-game behavior data and chat feature data can be processed by a rule engine to obtain the corresponding user attribute tags. The user attribute tags include game level tags, user level tags, and activity level tags. For example, for the game level tag, the game level tag for a game level less than 30 is "new player", the game level tag for a game level between 30 and 70 is "intermediate player", and the game level tag for a game level greater than 70 is "high-level player". For the user level tag, the user level tag for a VIP level greater than 5 is "high-VIP user", otherwise it is "ordinary user". For the activity level tag, the activity level tag for an average daily online time greater than or equal to a preset time (e.g., 3 hours) is "highly active user", and the activity level tag for an average daily online time less than a preset time (e.g., 1 hour) is "low-active user". Machine learning can be used to process in-game behavior data and chat feature data to obtain corresponding user attribute tags. For example, the K-Means clustering model can be used to cluster user spending habits and gameplay preferences to generate user attribute feature tags such as "casual player," "hardcore player," and "social player." For instance, a rule engine can be used to extract and process in-game behavior data and chat feature data to obtain corresponding user behavior status tags, including combat status tags, quest status tags, and resource status tags. For example, regarding combat status tags, a combat status tag for failing the same dungeon more than or equal to a preset number of times (e.g., 3 times) and not clearing it within a preset time (e.g., 24 hours) is "Dungeon Stuck," while a combat status tag for not entering a dungeon within a preset time (e.g., 7 days) is "Dungeon Neglected." Regarding quest status tags, a quest status tag for main quests that have stalled for more than or equal to a preset time (e.g., 48 hours) is "Quest Stuck," while a quest status tag for daily quests with a completion rate less than a preset number (e.g., 50%) is "Quest Neglect." Regarding resource status tags, a resource status tag for having less than a preset amount of gold (e.g., 1000) and not participating in resource dungeons is "Resource Scarcity." Semantic analysis can also be performed on in-game behavioral data and chat feature data to obtain corresponding user behavior status tags. For example, by analyzing chat content using the BERT intent classification model, if "stuck" and "can't beat" account for more than 60% of the questions, the corresponding user behavior tag is determined to be "Help-Seeking Status"; if the topic focuses on "PVP" and the win rate is 70%, the corresponding user behavior tag is determined to be "PVP Core Status."

[0058] S143. Determine the target response strategy by matching response strategies based on user attribute feature tags and user behavior status tags.

[0059] After determining the user attribute feature tags and user behavior status tags in S142, the target response strategy is determined by matching response strategies based on these tags. For example, the optimal response strategy can be matched from a strategy library based on a combination of tags (i.e., a combination of user attribute feature tags and user behavior status tags) to ensure the response accurately matches the user's needs. For instance, a corresponding strategy library can be set up according to content depth, presentation format, recommendation direction, and tone style. The design of the corresponding strategy library is shown in the table below.

[0060] When matching response strategies, for each candidate strategy, a matching score for the user attribute tag is calculated. For example, the matching score between "new player + stuck in dungeon" and "new player stuck strategy" is 0.3 × 0.9 + 0.4 × 0.95 = 0.64. The response strategy with the highest score is selected as the target response strategy. It should be noted that if the scores are the same, the strategy corresponding to the user behavior status tag is matched first, because dynamic status is closer to real-time needs.

[0061] S144. Modify the response content according to the target response strategy to obtain the target response content.

[0062] After determining the target response strategy in S143, the determined response content is modified according to the target response strategy to obtain the target response content. For example, modifications can be made in terms of content depth adjustment, presentation optimization, and tone and interaction adjustment. For example, when adjusting the content depth, for "new players", the response content "Level 70 dungeon requires a 3-person team" is expanded to obtain the target response content "Level 70 dungeon requires a 3-person team~ Recommended 1 tank + 1 healer + 1 DPS. The tank is responsible for absorbing damage, the healer heals, and the DPS deals damage. The steps are very simple~"; for "high-VIP players", the response content is simplified to obtain the target response content "Optimal team for level 70 dungeon: tank (XX class) + DPS (XX build). The core mechanism is to focus fire after breaking the BOSS's shield. You can try the extreme output method to improve efficiency." To optimize the presentation, we added image and text links for new players, such as: "Click to view screenshot of the dungeon entrance → [link]"; and provided video guides for high-level players, such as: "Advanced technique video → [link], including BOSS skill cycle analysis." Regarding tone and interaction adjustments, we added encouraging phrases for "humble" users, such as: "Your previous actions were already great, try this method~"; and prioritized reassurance for users "stuck in a dungeon," such as: "I know this level is a bit difficult, don't worry, follow the steps and you'll definitely pass~." Through these adjustments to content depth, presentation, and tone, we achieved the final target response content.

[0063] S145. Publish the target's reply to the game chat channel using the set virtual user role.

[0064] After obtaining the target response content in S144, the content can be published to the game chat channel using a pre-defined virtual user role to address the target user's question. This virtual user role can be a virtual customer service role, such as a "Game Customer Service Assistant," a "Dungeon Strategy Expert," or an "Event Benefits Officer," or it can be a corresponding virtual game personality. The language style is determined based on the virtual user role; for example, a "Friendly and Professional" style for a "Game Customer Service Assistant," and a "Concise and Efficient" style for a "Dungeon Strategy Expert." The target response content is then precisely pushed to the target context, i.e., the corresponding game chat channel, via the game chat channel interface. Targeted replies can be sent by tagging the target user with "@" or by referencing relevant chat content related to the target question.

[0065] The above-mentioned approach determines response strategy matching through user attribute feature tags and user behavior status tags, improving the relevance of responses to users' actual needs, avoiding generic, mechanical replies, improving response accuracy and personalization, and thus enhancing user satisfaction. By using personalized virtual user avatars to publish responses, and adapting the virtual avatars' tone and style to user attribute feature tags and user behavior status tags, emotional connection and immersion are strengthened, thereby increasing users' willingness to interact. Automated tag and response strategy matching reduces the workload of manually judging user needs, lowers the cost of manual intervention, and improves the operational efficiency of automated chat responses in game scenarios.

[0066] Figure 5 This is a flowchart illustrating another method for automatic chat reply in a game scenario provided in this application embodiment, see below. Figure 5 The automatic chat reply methods in this game scenario specifically include: S146. Determine the target game character in the current discussion based on similar chat aggregation data.

[0067] After determining the reply content in S13, the target game character in the current discussion can be determined based on similar chat aggregation data. For example, high-frequency entities are extracted from the topic_label, keywords, and user_speeches of similar chat aggregation data. For instance, if the keywords in the similar chat aggregation data include "warrior," "block skill," and "tank positioning," and the topic label is "warrior class gameplay discussion," then "warrior" is initially identified as a candidate character. If user comments frequently mention "village chief Li Dashan" and "quest rewards," then the candidate character is the NPC "village chief Li Dashan." Based on the extracted high-frequency entities, character entity recognition and disambiguation are performed. For example, a game-specific named entity recognition (NER) model (training data includes all character names, nicknames, and attributes in the game) is called to accurately identify character types (class / NPC / BOSS / pet, etc.), such as distinguishing between "mage" (class) and "archmage Merlin" (NPC). Next, ambiguity regarding character entities is addressed. For example, if "dragon" refers to both the boss "Red Dragon Skasa" and the pet "baby dragon," disambiguation is performed based on the context (e.g., "can't beat the dragon" → targeting the boss). After identifying the character entities, the character with the highest frequency of appearance and user discussion share in similar chat aggregation data is selected as the target game character.

[0068] S147. Based on the target game character, the reply content is rewritten in a character-based manner to obtain personalized reply content.

[0069] After identifying the target game character in S146, the response content is rewritten to reflect the target game character, resulting in personalized response content. For example, data such as the corresponding character style, identity, and language style can be retrieved from the game character character database based on the target game character. The identified response content is then rewritten based on this data. For instance, regarding language style, assuming the target game character is "Ironclad Warrior," the response "When in a team, be sure to protect your teammates" could be rewritten as "When in a team, you have to push forward! Taunt the BOSS, protect the backline—that's a warrior's duty!"

[0070] S148. Retrieve the target virtual user character corresponding to the virtual game personality based on the target game character.

[0071] After obtaining the personalized response in S147, the target virtual user character corresponding to the target game character is retrieved. For example, a virtual user character library is set up, where each virtual user character is bound to a unique game character. Each virtual user character includes attributes such as visual appearance, identity identifier, and speaking permissions. The target game character is matched with the virtual user characters in the virtual user character library, and the matched personalized target virtual user character is invoked.

[0072] S149. Publish personalized reply content to the game chat channel through the target virtual user character.

[0073] After determining the target virtual user role in S148, the personalized reply content is published to the game chat channel through the target virtual user role to answer the target user's target question. For example, the personalized reply content can be precisely pushed to the target scenario, i.e., the corresponding game chat channel, through the game chat channel interface. A precise reply can be given by using "@" to indicate the target user, or by referencing chat content related to the target question.

[0074] As described above, by using character-based language in responses, the language style and visuals of the responses align with the game character's persona. This allows players to perceive a genuine dialogue with an in-game character, enhancing their sense of immersion in the game's world and thus strengthening their overall user experience. Replies containing gameplay suggestions through a virtual user avatar corresponding to the game character are more readily accepted by users than ordinary replies, increasing the adoption rate. Furthermore, the unique character language improves user recall of the responses, reducing the need for repeated questions.

[0075] Compared to the traditional method of relying on other users to proactively reply, this embodiment can automatically identify and respond to problems, greatly improving the efficiency of resolving in-game chat issues and providing more accurate and official responses. By organizing and analyzing chat data, it can help identify frequently reported problems within the game, better assisting in game strategy optimization. By automatically identifying and proactively replying to user issues, it invigorates the user atmosphere in the game chat channel, enhances user acceptance of the game and communication efficiency, and is conducive to long-term user engagement.

[0076] Based on the above embodiments, Figure 6 This is a schematic diagram of an automatic chat reply device in a game scene, provided as an embodiment of this application. (Reference) Figure 6The automatic chat reply device in the game scenario provided in this embodiment specifically includes: a chat content acquisition module 21, a chat data aggregation module 22, a target question determination module 23, a knowledge item retrieval module 24, a reply content determination module 25, and a reply publishing module 26.

[0077] Among them, the chat content acquisition module 21 is used to acquire chat content in the game chat channel; Chat data aggregation module 22 is used to integrate and process contextual information based on chat content to obtain similar chat aggregate data; The target problem determination module 23 is used to perform semantic analysis on similar chat aggregation data to determine the target problem; The knowledge entry retrieval module 24 is used to retrieve the target question based on the set knowledge vector library and obtain the corresponding matching knowledge entries; The response content determination module 25 is used to generate corresponding response content based on the knowledge entries; The reply posting module 26 is used to post reply content to the game chat channel through the set virtual user role.

[0078] In one embodiment, the chat data aggregation module 22 includes: a target chat content determination submodule, a first target data determination submodule, a second target data determination submodule, and a chat data aggregation submodule; The target chat content determination submodule is used to determine a preset number of target chat contents based on the chat content and the settings of the sliding window. The first target data determination submodule is used to merge the messages of the same target user based on the target chat content to obtain the first target data; The second target data determination submodule is used to obtain the second target data by performing cross-user interaction identification processing of the corresponding target user based on the target chat content. The chat data aggregation submodule is used to integrate the first target data and the second target data to obtain similar chat aggregate data.

[0079] In one embodiment, the second target data determination submodule includes: a target user determination unit, an interaction information determination unit, an interaction speech content determination unit, and a second target data determination unit; The target user identification unit is used to identify target users based on the target chat content. The interactive information determination unit is used to determine interactive information based on the content of the target user's speech. The interactive speech content determination unit is used to determine the speech content of interactive users based on cross-user interaction identification processing of interactive information. The second target data determination unit is used to integrate the content of interactive users' statements to obtain the second target data.

[0080] In one embodiment, the target question determination module 23 includes: a semantic vector extraction submodule, a text cluster determination submodule, a multi-attribute extraction submodule, an intent label determination submodule, and a target question generation submodule; The semantic vector extraction submodule is used to extract semantic vectors from aggregated chat data of the same type. The text cluster determination submodule is used to perform clustering based on semantic vectors to obtain text clusters; The multi-attribute extraction submodule is used to extract keywords and entities from text clusters to obtain core entities, keywords, and game elements. The intent label determination submodule is used to perform text classification processing on text clusters to obtain core intent labels; The target question generation submodule is used to input text clusters, core entities, keywords, and core intent labels into a preset large language model to generate the corresponding target question.

[0081] In one embodiment, the chat content acquisition module 21 includes: an initial content acquisition submodule, a risk control identification submodule, and a target chat content determination submodule; The initial content retrieval submodule is used to retrieve the initial chat content from the game chat channel; The risk control identification submodule is used to perform risk control identification processing on the initial chat content in order to identify sensitive and invalid messages. The target chat content determination submodule is used to perform risk control processing on sensitive messages according to their corresponding sensitivity levels and to filter invalid messages to obtain the target chat content.

[0082] In one embodiment, the reply publishing module 26 includes: a behavior data acquisition submodule, a feature data acquisition submodule, a tag construction submodule, a reply strategy determination submodule, a reply content determination submodule, and a first reply publishing submodule; The behavior data acquisition submodule is used to acquire in-game behavior data of the target user; The feature data acquisition submodule is used to acquire chat feature data of target users in the game chat channel; The tag building submodule is used to build user attribute feature tags and user behavior status tags based on in-game behavior data and chat feature data. The response strategy determination submodule is used to determine the target response strategy by matching response strategies based on user attribute feature tags and user behavior status tags. The response content determination submodule is used to modify the response content according to the target response strategy to obtain the target response content; The first reply posting submodule is used to post the target reply content to the game chat channel through the set virtual user role.

[0083] In one embodiment, the reply publishing module 26 includes: a target game character determination submodule, a personalized reply content determination submodule, a target virtual user character retrieval submodule, and a second reply publishing submodule; The target game character determination submodule is used to determine the target game character in the current discussion based on similar chat aggregation data. The personalized reply content determination submodule is used to rewrite the reply content in a character-based manner based on the target game character to obtain personalized reply content; The target virtual user role retrieval submodule is used to retrieve the target virtual user role corresponding to the virtual game personality based on the target game role. The second reply publishing submodule is used to publish personalized reply content to the game chat channel through the target virtual user character.

[0084] The automatic chat reply device in the game scene provided in this application embodiment can be used to execute the automatic chat reply method in the game scene provided in the above embodiment, and has corresponding functions and beneficial effects.

[0085] This application provides an automatic chat reply device in a game scenario, referring to... Figure 7 The chat auto-reply device in this game scene includes: a processor 31, a memory 32, a communication module 33, an input device 34, and an output device 35. The number of processors and the number of memories in the chat auto-reply device can both be one and more. The processor, memory, communication module, input device, and output device of the chat auto-reply device can be connected via a bus or other means.

[0086] The memory 32, as a computer-readable storage medium, can be used to store software programs, computer-executable programs, and modules, such as program instructions / modules corresponding to the chat auto-reply method in the game scene described in any embodiment of this application (e.g., chat content acquisition module, chat data aggregation module, target question determination module, knowledge item retrieval module, reply content determination module, and reply publishing module in the chat auto-reply device in the game scene). The memory may mainly include a program storage area and a data storage area, wherein the program storage area may store the operating system and at least one application required for a function; the data storage area may store data created according to the use of the device, etc. In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other non-volatile solid-state storage device. In some instances, the memory may further include memory remotely located relative to the processor, and these remote memories can be connected to the device via a network. Examples of the above-mentioned networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0087] The communication module 33 is used for data transmission.

[0088] The processor 31 executes various functional applications and data processing of the device by running software programs, instructions and modules stored in the memory, thereby realizing the automatic chat reply method in the game scenario described above.

[0089] Input device 34 can be used to receive input digital or character information, and to generate key signal inputs related to user settings and function control of the device. Output device 35 may include display devices such as a display screen.

[0090] The chat auto-reply device in the game scenario provided above can be used to execute the chat auto-reply method in the game scenario provided in the above embodiments, and has corresponding functions and beneficial effects.

[0091] This application embodiment also provides a storage medium for storing computer-executable instructions. When executed by a computer processor, the computer-executable instructions are used to execute an automatic chat reply method in a game scenario. The automatic chat reply method in the game scenario includes: acquiring chat content in a game chat channel; integrating and processing contextual information based on the chat content to obtain similar chat aggregation data; performing semantic analysis on the similar chat aggregation data to determine the target question; retrieving the target question based on a set knowledge vector library to obtain corresponding matching knowledge entries; generating corresponding reply content based on the knowledge entries; and publishing the reply content to the game chat channel through a set virtual user role.

[0092] Storage medium – any type of memory device or storage device. The term “storage medium” is intended to include: mounting media, such as CD-ROM, floppy disk, or magnetic tape devices; computer system memory or random access memory, such as DRAM, DDR RAM, SRAM, EDO RAM, Rambus RAM, etc.; non-volatile memory, such as flash memory, magnetic media (e.g., hard disk or optical storage); registers or other similar types of memory elements, etc. Storage medium may also include other types of memory or combinations thereof. Furthermore, storage medium may reside in a first computer system in which the program is executed, or it may reside in a different second computer system connected to the first computer system via a network (such as the Internet). The second computer system can provide program instructions to the first computer for execution. The term “storage medium” can include two or more storage media residing in different locations (e.g., in different computer systems connected via a network). Storage medium may store program instructions (e.g., specifically implemented as a computer program) executable by one or more processors.

[0093] Of course, the computer-executable instructions provided in the embodiments of this application are not limited to the chat auto-reply method in the game scene as described above, but can also execute related operations in the chat auto-reply method in the game scene provided in any embodiment of this application.

[0094] The chat auto-reply device, storage medium, and chat auto-reply equipment in the game scene provided in the above embodiments can execute the chat auto-reply method in the game scene provided in any embodiment of this application. For technical details not described in detail in the above embodiments, please refer to the chat auto-reply method in the game scene provided in any embodiment of this application.

[0095] The above description is merely a preferred embodiment and the technical principles employed in this application. This application is not limited to the specific embodiments described herein, and various obvious changes, readjustments, and substitutions that can be made by those skilled in the art will not depart from the scope of protection of this application. Therefore, although this application has been described in detail through the above embodiments, this application is not limited to the above embodiments, and may include more other equivalent embodiments without departing from the concept of this application, the scope of which is determined by the scope of the claims.

Claims

1. A method for automatically replying to chat in a game scenario, characterized in that, include: Obtain chat content from the game chat channel, and perform contextual information integration processing based on the chat content to obtain aggregated chat data of the same type; Semantic analysis was performed on the aforementioned aggregated chat data to identify the target problem; The target question is retrieved based on the set knowledge vector library to obtain the corresponding matching knowledge entries, and the corresponding response content is generated based on the knowledge entries; The reply content is published to the game chat channel by setting up a virtual user role.

2. The method according to claim 1, characterized in that, The process of integrating contextual information based on the chat content to obtain similar aggregated chat data includes: Based on the chat content and the set sliding window, a preset number of target chat contents are determined; The first target data is obtained by merging the messages of the same target user based on the target chat content; Based on the target chat content, cross-user interaction identification processing of the corresponding target user is performed to obtain the second target data; The first target data and the second target data are integrated to obtain the similar chat aggregation data.

3. The method according to claim 2, characterized in that, The process of obtaining the second target data by performing cross-user interaction identification based on the target chat content includes: Target users are identified based on the target chat content, and interactive information is determined based on the content of the target users' messages. Based on the interaction information, cross-user interaction identification processing is performed to determine the content of the interactive users' statements, and the content of the interactive users' statements is integrated to obtain the second target data.

4. The method according to any one of claims 1-3, characterized in that, The step of performing semantic analysis on the aggregated chat data of the same type to determine the target problem includes: Semantic vectors are obtained by extracting features from the aggregated chat data of the same type, and text clusters are obtained by clustering the semantic vectors. Keyword extraction and entity extraction are performed on the text cluster to obtain core entities, keywords and game elements, and text classification is performed on the text cluster to obtain core intent tags; The text cluster, the core entity, the keywords, and the core intent label are input into a preset large language model to generate the corresponding target question.

5. The method according to any one of claims 1-3, characterized in that, The process of obtaining chat content from the game chat channel includes: Get the initial chat content from the game chat channel; The initial chat content is subjected to risk control identification processing to identify sensitive and invalid messages; The sensitive speech content is processed according to its corresponding sensitivity level, and the invalid speech content is filtered to obtain the target chat content.

6. The method according to any one of claims 1-3, characterized in that, The step of publishing the reply content to the game chat channel through a set virtual user role includes: Acquire in-game behavior data of the target user and chat feature data in the game chat channel; Construct user attribute feature tags and user behavior status tags based on the in-game behavior data and the chat feature data; The target response strategy is determined by matching the response strategy based on the user attribute feature tags and user behavior status tags. The target response content is obtained by modifying the response content according to the target response strategy; The target's reply content is published to the game chat channel by setting up a virtual user role.

7. The method according to any one of claims 1-3, characterized in that, The step of publishing the reply content to the game chat channel through a set virtual user role includes: The target game character in the current discussion is determined based on the aforementioned similar chat aggregation data; Based on the target game character, the reply content is rewritten in a character-centric manner to obtain personalized reply content; Based on the target game character, retrieve the target virtual user character with the corresponding virtual game personality; The personalized reply content is published to the game chat channel through the target virtual user character.

8. An automatic chat reply device for a game scene, characterized in that, include: The chat content acquisition module is used to acquire chat content from the game's chat channel; The chat data aggregation module is used to integrate and process contextual information based on the chat content to obtain similar chat aggregate data. The target question determination module is used to perform semantic analysis on the similar chat aggregation data to determine the target question. The knowledge entry retrieval module is used to retrieve the target question based on the set knowledge vector library and obtain the corresponding matching knowledge entries; The response content determination module is used to generate corresponding response content based on the knowledge entries; The reply posting module is used to post the reply content to the game chat channel through the set virtual user role.

9. An automatic chat reply device for a game scenario, characterized in that, include: Memory and one or more processors; The memory is used to store one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any one of claims 1-7.

10. A storage medium for storing computer-executable instructions, characterized in that, The computer-executable instructions, when executed by a processor, are used to perform the method as described in any one of claims 1-7.

Images