Information processing methods, devices and storage media
By classifying and categorizing user audio information into events, a structured information set is generated, solving the problem of utilizing massive amounts of data and improving the accuracy of conversations and the intelligence of user interaction.
Patent Information
- Application Number
- CN202611282730.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-21
- Publication Date
- 2026-11-10
AI Technical Summary
Existing technologies struggle to effectively utilize the massive amounts of conversational content and environmental voice data generated by users in their daily lives to provide intelligent and accurate services.
By classifying audio information based on event classification, a structured information set is generated, and a session is output in response to the user's session request.
It significantly improves the accuracy and naturalness of conversations, enhances the intelligence of user interaction, and provides an experience that closely resembles the user's actual experience.
Smart Images

Figure CN123068679A_ABST
Abstract
Description
Technical Field
[0001] This disclosure relates to the field of computer technology, and in particular to an information processing method, apparatus and storage medium. Background Technology
[0002] With the popularization of wearable devices such as artificial intelligence (AI) glasses and smartwatches, users generate a large amount of conversational content and environmental speech transcription input in their daily lives. Therefore, how to help users make good use of relevant data and information to provide more intelligent services is a problem worth solving. Summary of the Invention
[0003] This disclosure provides an information processing method, apparatus, and storage medium, the technical solutions of which are as follows: Firstly, an information processing method is provided, the method comprising: Based on event classification, audio information is categorized to obtain at least one set of information corresponding to an event; In response to a user's session request, a session is generated based on the information set corresponding to the at least one event, and the session is output to the user.
[0004] Secondly, an information processing apparatus is provided, the apparatus comprising: The first acquisition module is used to acquire one or more first text fragments associated with the user identifier; The classification module is used to classify audio information based on event classification to obtain an information set corresponding to at least one event. A generation module is used to respond to a user's session request, generate a session based on the information set corresponding to the at least one event, and output the session to the user.
[0005] Thirdly, a computer device is provided, the computer device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the information processing method described in the first aspect.
[0006] Fourthly, a computer-readable storage medium is provided, the computer-readable storage medium storing a computer program, which, when executed by a processor, implements the information processing method described in the first aspect.
[0007] Fifthly, a computer program product is provided, which, when run on a computer device, causes the computer device to perform the information processing method described in the first aspect.
[0008] In this disclosure, audio information is categorized based on event classification to obtain an information set corresponding to at least one event. In response to a user's session request, a session is generated based on the information set corresponding to at least one event and output to the user. Event classification transforms fragmented audio content into a structured information set organized by events. Generating a session based on the information set corresponding to at least one event significantly improves session accuracy, making responses more closely aligned with the user's actual experiences and needs, and enhancing the naturalness and intelligence of the interaction. Attached Figure Description
[0009] Figure 1 This is a schematic diagram of the structure of a long-term memory system provided in an embodiment of this disclosure; Figure 2 This is an overall architecture diagram of a long-term memory system provided in an embodiment of this disclosure; Figure 3 This is a flowchart of an information processing method provided in an embodiment of this disclosure; Figure 4 This is a flowchart of a user-level delayed batch processing extraction scheduling and user-level lock serialization provided in an embodiment of this disclosure; Figure 5 This is a flowchart and output diagram of a multi-stage incremental boundary detection process and an interval list provided in an embodiment of this disclosure; Figure 6 This is a schematic diagram illustrating the state and boundary of an open / closed memory cell according to an embodiment of this disclosure; Figure 7 This is a diagram illustrating the source relationship between parent-child references and evidence in a hierarchical memory data model provided in this embodiment. Figure 8 This is a flowchart illustrating a user profile information extraction process provided in an embodiment of this disclosure; Figure 9 This is a flowchart of an intelligent agent-based multi-turn retrieval method provided in an embodiment of this disclosure; Figure 10 This is a schematic diagram of the structure of a semantic information extraction device provided in an embodiment of this disclosure; Figure 11 This is a schematic diagram of the structure of a computer device provided in an embodiment of this disclosure. Detailed Implementation
[0010] It should be understood that, when used in this disclosure and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.
[0011] The term "multiple" as used in this disclosure refers to two or more. In the description of this disclosure, unless otherwise stated, " / " signifies "or," for example, A / B can mean A or B. "And / or" in this document is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone.
[0012] To facilitate a clear description of the technical solutions disclosed herein, the terms "first" and "second" are used to distinguish identical or similar items with substantially the same function and effect. Those skilled in the art will understand that the terms "first" and "second" do not limit the quantity or execution order, and that the terms "first" and "second" do not necessarily imply that they are different.
[0013] It is understood that before using the technical solutions disclosed in the various embodiments of this disclosure, users should be informed of the types, scope of use, and usage scenarios of the personal information involved in this disclosure in an appropriate manner in accordance with relevant laws and regulations, and user authorization should be obtained.
[0014] For example, upon receiving a user's active request, a prompt message is sent to the user to explicitly inform them that the requested operation will require the acquisition and use of the user's personal information. This allows the user to independently choose whether to provide personal information to the software or hardware, such as the electronic device, application, server, or storage medium performing the operations of this disclosed technical solution, based on the prompt message.
[0015] As an optional but non-limiting implementation, in response to a user's active request, sending a prompt message to the user can be done via a pop-up window, where the prompt message can be presented in text format. Furthermore, the pop-up window can also include a selection control allowing the user to choose "agree" or "disagree" to provide personal information to the electronic device.
[0016] It is understood that the above notification and user authorization process are merely illustrative and do not constitute a limitation on the implementation of this disclosure. Other methods that comply with relevant laws and regulations may also be applied to the implementation of this disclosure.
[0017] It is understood that the data involved in this technical solution (including but not limited to the data itself, the acquisition or use of the data) shall comply with the requirements of relevant laws, regulations and related provisions.
[0018] The technical solutions and application scenarios involved in the embodiments of this disclosure are described below.
[0019] With the widespread adoption of wearable devices such as artificial intelligence (AI) glasses and smartwatches, users generate a massive amount of continuous, real-time conversational content and environmental speech-to-text input in their daily lives. This real-time, streaming data contains rich personalized information, providing a data foundation for building long-term memory and enabling AI agents to deeply understand users' life experiences, preferences, and intentions.
[0020] To address this, this disclosure provides an information processing method in which audio information is categorized based on event classification to obtain an information set corresponding to at least one event. In response to a user's session request, a session is generated based on the information set corresponding to at least one event and output to the user. Event classification transforms fragmented audio content into a structured information set organized by events. Generating a session based on the information set corresponding to at least one event significantly improves session accuracy, making responses more relevant to the user's actual experiences and needs, and enhancing the naturalness and intelligence of the interaction.
[0021] Figure 1 This is a schematic diagram of the structure of a long-term memory system provided in an embodiment of this disclosure. See also... Figure 1 The long-term memory system 10 may include a data acquisition module 101, a writing and scheduling module 102, a memory construction module 103, a distributed lock module 104, a storage module 105, and a memory retrieval module 106.
[0022] For example, the long-term memory system 10 can operate in a single device, or the data acquisition module 101, writing and scheduling module 102, memory construction module 103, distributed lock module 104, storage module 105, and memory retrieval module 106 in the long-term memory system 10 can operate in different devices. For instance, when the long-term memory system 10 operates in a single device, the device can be smart glasses, a smartphone, or a server. Alternatively, when the long-term memory system 10 operates in multiple devices, these devices can include smart glasses, electronic devices connected to the smart glasses (such as mobile phones, servers, etc.), etc., where the data acquisition module 101 can operate in the smart glasses, and the writing and scheduling module 102, memory construction module 103, distributed lock module 104, storage module 105, and memory retrieval module 106 can operate in the electronic devices.
[0023] The data acquisition module 101 can communicate with the writing and scheduling module 102 via wired or wireless connection.
[0024] The write and scheduling module 102 can communicate with the memory building module 103, the distributed lock module 104, and the storage module 105 (i.e., the database described below) via wired or wireless connections.
[0025] The memory building module 103 can communicate with the memory retrieval module 106 via wired or wireless connection.
[0026] The data acquisition module 101 is used to acquire text fragments and send them to the writing and scheduling module 102. The text fragments are text content segments obtained by transcribing the user's audio information through speech recognition.
[0027] The writing and scheduling module 102 is used to receive text fragments sent by the data acquisition module 101, and send the text fragments to the storage module 105 for storage.
[0028] In one scenario, a text fragment is associated with a user identifier, which identifies the user who generated the text fragment. Alternatively, the user identifier can also identify the user who generated the audio information of the text fragment. The writing and scheduling module 102 can perform storage scheduling on the text fragment based on the user identifier and write the text fragment to the storage module 105, where the storage module 105 stores the text fragment. In this way, a data foundation can be provided for building a user's long-term memory.
[0029] In one scenario, the storage module 105 can store the processing status of the text fragments along with the text fragments. The processing status of the text fragments is used to indicate whether the text fragments have been processed by the memory construction module 103, such as processed or pending processing.
[0030] In another scenario, the data acquisition module 101 can acquire the user's audio information and send it to the writing and scheduling module 102. The writing and scheduling module 102 performs speech recognition on the user's audio information and transcribes it into text fragments. The text fragments are then scheduled to be written to the storage module 105 for storage.
[0031] The memory construction module 103 is used to obtain one or more text fragments from the storage module 105, classify the one or more text fragments based on event classification, and obtain an information set corresponding to at least one event; extract at least one memory information from each information set corresponding to at least one event; and store at least one memory information in the storage module 105.
[0032] The one or more text fragments are input into a large language model, which performs content understanding on the one or more text fragments according to pre-written instructions. Based on the content understanding, the one or more text fragments are categorized according to event classification. The pre-written instructions may include event classification information, descriptive information about the event classification, classification principles, etc.
[0033] After the above classification process, at least one information set corresponding to an event is obtained. The content of the information set still consists of one or more text fragments belonging to the same event after classification. Furthermore, content extraction can be performed on the content of the information set to extract memory information used to describe the event.
[0034] Optionally, the memory information may include one or more of the following: episodic memory information, event log information, and predictive memory information. Episodic memory information is information with a timeline and background, forming a coherent description of a series of related facts that occurred over a period of time. For example, episodic memory information may be scene information extracted from a text fragment, such as "On the afternoon of August 1st, at a coffee shop in Xuhui District, Shanghai, I discussed travel plans for the second half of the year with a friend." Event log information is the smallest indivisible factual statement extracted from a text fragment, such as "Zhang San's phone number is 138xxxx1234" or "The meeting will start at 3 PM on Friday." Predictive memory information is a predicted plan, to-do list, commitment, or expected goal for the future, such as "The project report needs to be finished by next Wednesday."
[0035] In one scenario, after the memory construction module 103 obtains at least one piece of memory information, it can set the status of one or more text fragments stored in the storage module 105 corresponding to the obtained at least one piece of memory information to be processed.
[0036] In one scenario, when the memory construction module 103 retrieves one or more text fragments from the storage module 105, it can acquire a distributed lock from the distributed lock module 104 to lock the text fragment corresponding to the user identifier. This allows for serialization of the extraction of text fragments from the same user across multiple nodes or replicas, avoiding duplicate extractions and concurrent write conflicts.
[0037] Optionally, after the memory construction module 103 extracts at least one memory information from each information set in the information set corresponding to at least one event, it can unlock the text fragment corresponding to the user identifier in the storage module 105.
[0038] In another scenario, the memory construction module 103 can obtain audio information from the storage module 105, identify the audio information, and obtain at least one audio segment corresponding to the audio information; classify the at least one audio segment based on event classification to obtain an information set corresponding to at least one event; extract at least one memory information from each information set in the information set corresponding to at least one event; and store at least one memory information in the storage module 105.
[0039] The distributed lock module 104 is used to lock the text fragments of the user identifier stored in the storage module 105 when the memory construction module 103 retrieves one or more text fragments associated with the user identifier from the storage module 105. This ensures that the memory data processing task for the same user can only be executed by one computing node at a time, preventing concurrent conflicts and data corruption. For example, the distributed lock module 104 can be a Redis user-level lock, etc.
[0040] The storage module 105 is used to store the text fragments acquired by the data acquisition module 101, and the memory information extracted by the memory construction module 103 based on the information set corresponding to at least one event.
[0041] The memory retrieval module 106 is used to receive session requests and generate a session based on the session request and the information set retrieved from the information set corresponding to at least one event. For example, the memory retrieval module 106 can support keyword retrieval, vector retrieval, and hybrid retrieval (such as keyword + vector).
[0042] Optionally, the long-term memory system 10 may also include a vector library for storing vectors of memory information.
[0043] For example, such as Figure 2 As shown, Figure 2 This is an overall architecture diagram of a long-term memory system provided in an embodiment of the present disclosure. Figure 2 Each time the data acquisition module acquires a text fragment, it sends the fragment to the writing and scheduling module. Upon receiving the text fragment, the writing and scheduling module stores it in the storage module. The memory construction module reads at least one text fragment from the storage module in batches; during the reading process, the distributed lock module ensures the uniqueness and stability of the data read by the memory construction module through a locking and debouncing mechanism. Then, the memory construction module triggers memory information retrieval in sequence. When the memory retrieval module receives a session request, it queries the vector library and / or the storage module based on the session request to obtain an information set, and generates a session based on the content of the session request and the information set.
[0044] Figure 3 This is a flowchart of an information processing method provided in an embodiment of this disclosure. The method may include the following steps: Step 301: The computer device classifies the audio information based on event classification to obtain an information set corresponding to at least one event.
[0045] For example, computer equipment can be wearable devices such as smart glasses, mobile devices such as mobile phones, etc.
[0046] Computer equipment is used to collect users' audio information, perform semantic analysis and recognition on the recorded content of the audio information, and classify it according to a set event category. This event category includes: classification according to the application scenario to which the recorded content belongs, such as meeting scenarios, interview dialogue scenarios, personal daily event recording scenarios, etc.; classification according to the objective events described by the recorded content, with one objective event per category, such as "short-term heavy rain event in a certain location on xx year-xx month-xx day" or "going to a certain university to register on xx year-xx month-xx day", etc.; and classification according to the user's interests or topics described by the recorded content, with one interest or topic per category, such as "sports", "technology", "entertainment", etc.
[0047] After the above classification process, we obtain at least one information set corresponding to an event. The content of the information set still consists of one or more text fragments belonging to the same event after classification. In this way, classifying by event ensures that each text fragment has a corresponding category. Furthermore, adopting a multi-dimensional classification method makes the classification criteria clearer and more comprehensive, avoiding the one-sidedness of a single dimension, adapting to the diverse needs of different application scenarios, and improving the semantic richness of the information set.
[0048] Audio information is unstructured, continuous, and multi-topic. Directly using unclassified audio information may lead to problems such as retrieval difficulties and semantic confusion. By classifying audio information, scattered audio content can be organized into information sets organized by event according to categories, thereby providing a data foundation for subsequent responses to user session requests and reducing the complexity of subsequent user session requests.
[0049] Step 302: In response to the user's session request, the computer device generates a session based on the information set corresponding to at least one event and outputs the session to the user.
[0050] For example, the session request can be triggered manually by the user or by a voice assistant.
[0051] Since the information set corresponding to the at least one event includes all the information of each event in the at least one event, a conversation can be generated more fully and accurately based on the information set corresponding to the at least one event, thereby significantly improving the quality of conversations with users and enhancing the naturalness and intelligence of interaction with users.
[0052] In one scenario, the operation of a computer device classifying audio information based on event classification to obtain an information set corresponding to at least one event can include the following two methods.
[0053] Method 1: The computer device performs speech recognition on the audio information to obtain at least one text segment corresponding to the audio information; based on event classification, the at least one text segment is categorized to obtain an information set corresponding to at least one event.
[0054] For example, a computer device can input at least one text fragment into a large language model and obtain a set of information corresponding to at least one event output by the large language model.
[0055] Since at least one text fragment contains all the semantics of the audio information, and text fragments are easy to store, retrieve, and semantically analyze, classifying at least one text fragment can reduce the difficulty of event classification and improve the speed and accuracy of event classification.
[0056] After the one or more text fragments are categorized, they are stored in a database. Each text fragment is associated with a user identifier and its processing status is recorded during storage. This allows for subsequent queries or processing of information within the text fragments, retrieving the fragments from the database based on the user identifier. This approach ensures accurate acquisition of user-related text fragments, crucial for accurate retrieval of user memory information, and avoids redundant processing of already processed data, enabling incremental processing and reducing redundant computation.
[0057] For example, the processing status of a text fragment is used to indicate whether the text fragment has been categorized by a computer device (i.e., whether it has been used to retrieve information from memory), such as pending, processed, etc.
[0058] For example, computer devices can be stored in a database with a low latency, such as 30 milliseconds (ms), 50 ms, or 70 ms.
[0059] This state management approach avoids repetitive processing of text fragments. Furthermore, by receiving and storing text fragments sequentially, the probability of losing text fragments is reduced, ensuring the integrity of the text fragments.
[0060] Method 2: The computer device performs speech recognition on the audio information to obtain at least one audio segment corresponding to the audio information; based on event classification, the at least one audio segment is categorized to obtain an information set corresponding to at least one event.
[0061] At least one audio segment is obtained by performing speech segmentation and / or endpoint detection on the audio information, such as identifying audio information by means of silence, speaker switching, topic boundaries, etc.
[0062] Directly classifying audio segments can preserve the complete information and original acoustic features of the audio, which helps to improve the accuracy of subsequent retrieval based on audio features. This approach is suitable for scenarios that require the analysis of non-textual information.
[0063] In some implementations, before performing classification processing on at least one text fragment in Method 1 above, the text fragment to be processed can be extracted from the database according to the user identifier in the following three ways: The first method: If the number of text fragments associated with the user identifier in the database and the number of text fragments to be processed is greater than or equal to the first number, retrieve the text fragments from the database and process them.
[0064] The initial quantity can be preset, such as 5, 6, or 7. The selection of the initial quantity is based on the delay requirements of the processed text fragments and the semantic completeness of the text fragments, so as to ensure that there are a relatively sufficient number of text fragments to extract relatively complete semantics from the text fragments when retrieving memory information later, while the number of text fragments will not cause a significant delay in processing.
[0065] If the number of text fragments is less than the first number, it may affect the ability to extract relatively complete and accurate semantics from the text fragments when retrieving memory information later. In this case, it is not necessary to process it immediately, but to wait until the number of text fragments reaches the first number before processing.
[0066] The above method can merge multiple text fragments to be processed, acquire and process them all at once, and reduce the frequency and cost of subsequent classification and processing.
[0067] The second method: Every first preset time interval, retrieve the unprocessed text fragments accumulated in the database within that first preset time interval.
[0068] The first preset duration can be set in advance, such as 4 minutes, 5 minutes, or 6 minutes.
[0069] Text fragments are retrieved from the database at first preset intervals, enabling the long-term memory system to trigger processing even when the number of text fragments is small, thereby preventing the time for retrieving memory information from being extended indefinitely.
[0070] The third method: If the number of text fragments associated with the user identifier and to be processed in the database is less than the first number, determine the first duration of the last time the text fragment associated with the user identifier was obtained from the database. If the first duration is greater than or equal to the first preset duration, obtain the text fragment associated with the user identifier and to be processed.
[0071] In this way, the memory system can effectively prevent the occurrence of low memory information retrieval frequency while maintaining real-time processing and an appropriate frequency of text fragment retrieval, thus ensuring the stability of memory information retrieval.
[0072] In one scenario, retrieving a text fragment associated with a user identifier and awaiting processing from the database can be done by retrieving the text fragment from the database while simultaneously locking the retrieved text fragment, provided that the text fragment in the database is currently unlocked.
[0073] A text fragment in the database being in an unlocked state means that no other node is currently accessing that text fragment in the database. Conversely, a text fragment being in a locked state means that another node is currently accessing that text fragment in the database.
[0074] By acquiring distributed locks at the user level (i.e., locking text fragments in the database), the extraction and processing of text fragments corresponding to the same user can be serialized across multiple nodes, thereby avoiding multiple nodes repeatedly extracting the same text fragment and ensuring the correct execution of the processing flow.
[0075] Optionally, the database may also include the retrieval time of the text fragment, which indicates the time of the most recent successful retrieval of the text fragment.
[0076] In one scenario, before locking a text segment in the database, the computer device can determine the target time of the last retrieval of the text segment associated with the user identifier from the database; if the time difference between the target time and the current time is greater than or equal to a second time difference, subsequent locking and retrieval operations are performed; if the time difference between the target time and the current time is less than the second time difference, the current locking and retrieval operation is abandoned; and / or, after locking a text segment in the database, the computer device can determine the target time of the last retrieval of the text segment associated with the user identifier from the database; if the time difference between the target time and the current time is greater than or equal to a second time difference, a retrieval operation is performed; if the time difference between the target time and the current time is less than the second time difference, the current retrieval operation is abandoned and the data is unlocked.
[0077] If the time difference between the target time and the current time is greater than or equal to the second time difference, it means that the conditions for obtaining the text fragment are met, so the subsequent locking and acquiring operations can be performed. If the time difference between the target time and the current time is less than the second time difference, it means that other nodes have successfully obtained the text fragment before or during the process of acquiring the lock, that is, the conditions for obtaining the text fragment are no longer met, so the current locking and acquiring operations can be abandoned.
[0078] By performing retrieval duration checks before and after locking text fragments, the probability of repeatedly retrieving text fragments can be significantly reduced, avoiding excessive extraction of text fragments and ensuring the rhythm of subsequent retrieval of memory information.
[0079] Optionally, after obtaining the text fragments and performing classification processing on the text fragments to obtain the corresponding information set, the information set can be associated with the user identifier and stored in the database, and the text fragments associated with the user identifier in the database can be unlocked.
[0080] To facilitate understanding, the following will be combined with... Figure 4 This document provides an example illustrating the complete process of locking and unlocking text fragments in a database.
[0081] For example, Figure 4 This is a flowchart of user-level delayed batch processing extraction scheduling and user-level lock serialization provided in an embodiment of this disclosure. It includes the following steps 401 to 408.
[0082] Step 401: The computer device receives one or more text fragments and stores the one or more text fragments.
[0083] Step 402: The computer device determines whether the number of text fragments to be processed by the user is greater than or equal to the first number.
[0084] If the number of text fragments to be processed by the user is greater than or equal to the first number, then proceed to step 403; otherwise, proceed to step 401.
[0085] Step 403: The computer device determines whether the waiting time is greater than or equal to the second duration.
[0086] If the waiting time is greater than or equal to the second duration, proceed to step 404; otherwise, proceed to step 401.
[0087] Step 404: The computer device locks the text fragments in the database that are associated with the user identifier.
[0088] Step 405: After locking the text segment associated with the user identifier, the computer device determines whether the waiting time is greater than or equal to the second duration.
[0089] If the waiting time is greater than or equal to the second duration, proceed to step 406; otherwise, proceed to step 408.
[0090] Step 406: The computer device retrieves the text fragment to be processed that is associated with the user identifier from the database.
[0091] Step 407: The computer device performs classification processing on the text fragments to be processed, and obtains the corresponding information set.
[0092] Step 408: The computer device unlocks the text fragment associated with the user identifier.
[0093] In one scenario, a computer device can input at least one text fragment into a large language model, which then divides the text fragment based on event classification to obtain an information set corresponding to at least one event.
[0094] Since the original text fragments are a continuous, unbounded stream, classifying text fragments based on events allows for the combination of text fragments belonging to the same topic to form complete events. This provides a more complete semantic basis for subsequent retrieval of memory information, thereby ensuring the accuracy of event classification and memory information retrieval.
[0095] In one scenario, before a computer device segments one or more text fragments, it may obtain one or more sets of historical information associated with a user identifier. The historical information sets are in a first open state, indicating that the historical information sets lack event end information.
[0096] The information set has one or more states, such as open and closed. The open state can include a first open state and a second open state. The open state is used to indicate that the information set lacks event start information and / or event end information. The second open state is used to indicate that the information set lacks event start information. The closed state is used to indicate that the information set has event start information and event end information.
[0097] For example, “()” and “)” represent open states, with “()” representing the second open state, which is the set of information continuing from the previous part; “)” represents the first open state, which is the set of information continuing from the next part. “[” and “]” represent closed states, with “[” indicating the start of a new event, representing a new event, and “]” indicating the end of an event, representing a closed event.
[0098] Since streaming text input may cause a complete event to be scattered across multiple memory information retrieval cycles, obtaining one or more sets of historical information associated with the user identifier can provide a basis for subsequently combining the historical information sets to form a complete information set.
[0099] In one scenario, the operation of a computer device to extract memory information based on each of one or more information sets can be as follows: for any one of the one or more information sets, if the state of the information set is closed, extract the memory information of the information set; if the state of the information set is a second open state, extract the memory information based on the information set and a target historical information set, where the target historical information set is a set of historical information corresponding to the same event as the information set from one or more historical information sets.
[0100] If the information set is in a closed state, it means that the information set is a complete event, so the memory information of the information set can be directly extracted; if the information set is in a second open state, it means that the information set may be related to the historical information set, so the memory information can be extracted based on the information set and the target historical information set.
[0101] In this way, each time memory information is retrieved, there's no need to wait for an event to finish before starting a new one; events can be processed continuously. During event processing, scattered but related events are automatically integrated into a complete event, thus achieving real-time event processing. Furthermore, event merging can be automatically performed during event processing. This enables incremental, streaming event processing, ensuring event integrity and improving the quality of memory information retrieval.
[0102] In one scenario, a computer device can determine whether a target historical information set exists in one or more historical information sets that corresponds to the same event as the information set; if it does, it extracts memory information based on the information set and the target historical information set; if it does not, it extracts memory information from the information set.
[0103] For example, based on at least one set of historical information and one set of information, it is possible to determine whether a target set of historical information exists using a large language model.
[0104] In one scenario, before the computer device acquires one or more sets of historical information associated with a user identifier, for any set of historical information associated with the user identifier that is in a first open state, if the time difference between the current time and the recording time corresponding to the historical information set is greater than or equal to a second preset time, or if the number of text fragments in the historical information set is greater than or equal to a preset number, the state of the historical information set is updated to a closed state.
[0105] The second preset duration can be set in advance, such as 12 hours, 24 hours, or 36 hours.
[0106] If the time difference between the current time and the recording time corresponding to the historical information set is greater than or equal to the second preset time, it means that a considerable amount of time has passed since the user output the voice data corresponding to the historical information set, and the event corresponding to the historical information set has most likely ended naturally. Therefore, the state of the historical information set can be updated to closed to avoid the event remaining in a waiting state. If the time difference between the current time and the recording time corresponding to the historical information set is less than the second preset time, it means that a relatively short period of time has passed since the user output the voice data corresponding to the historical information set, and the event corresponding to the historical information set may not have ended yet. Therefore, the state of the historical information set can be kept open.
[0107] If the number of text fragments in the historical information set is greater than or equal to the preset number, it means that there are many text fragments and the content of the event corresponding to the historical information set is rich enough. Therefore, the state of the historical information set is updated to closed to avoid continuing to wait and increasing the processing delay and memory burden of the long-term memory system. If the number of text fragments in the historical information set is less than the preset number, it means that there are few text fragments and the content of the event corresponding to the historical information set is less. Therefore, the state of the historical information set is kept open.
[0108] By setting timeout and length criteria for historical information sets, a clear end boundary is established for historical information sets in an "open state" during streaming processing. This not only ensures the real-time performance and stability of pure semantic judgments and reduces resource consumption, but also prevents resource leaks and memory construction stagnation that may occur due to prolonged user silence, ensuring the responsiveness of the long-term memory system. Furthermore, by splitting extremely long dialogues, the model's context window limitations can be circumvented, improving the robustness of single extractions and the semantic purity of the final memory unit.
[0109] To facilitate understanding, the following will be combined with... Figure 5 The process of closing the aforementioned open historical information set and generating a new information set is illustrated by an example.
[0110] Figure 5 A multi-stage incremental boundary detection flowchart provided for embodiments of this disclosure includes stages 0 to 3.
[0111] Phase 0: The computer device acquires one or more sets of historical information.
[0112] Phase 1: The computer device performs rule pre-checks on one or more sets of historical information to close the set of historical information that meets the requirements, thus obtaining a set of historical information that is in an open state.
[0113] Phase 2: The computer device divides at least one text fragment to obtain at least one information set, and determines whether there is a semantic relationship between the open historical information set and at least one information set.
[0114] Phase 3: The computer device generates, continues, or closes the state of the open historical information set and at least one information set based on the semantic association between the open historical information set and at least one information set.
[0115] To facilitate understanding, the following will be combined with... Figure 6 An exemplary illustration is provided of the complete update process of the state of the information set corresponding to an event.
[0116] Figure 6 This is a schematic diagram illustrating the state and boundary of an open / closed memory cell, as provided in an embodiment of this disclosure.
[0117] After receiving the first text fragment message, a message set is created based on the first text fragment. The left boundary (left_bound_message_id) of the message set is determined as the first text fragment, and the left boundary of the message set is in a closed state, while the right boundary of the message set is in an open state.
[0118] Subsequently, for each received text fragment, it is determined whether the text fragment is associated with the information set, and whether the event corresponding to the information set has ended based on the text fragment. If the text fragment is associated with the information set but the event has not ended, the fragment identifier of the text fragment is added to source_message_ids[]; if the text fragment is associated with the information set and the event has ended, the fragment identifier of the text fragment is added to source_message_ids[], and the text fragment is identified as the right boundary (right_bound_message_id) of the information set, and the state and closure time of the information set are updated. After extracting memory information from the text fragment, the memory information is stored in linked_entities[].
[0119] For example, the operation of classifying at least one audio segment based on event classification to obtain the information set corresponding to at least one event can refer to the operation in Method 1 above, and will not be described in detail here.
[0120] In one scenario, the computer device can process the information set to obtain semantic information that identifies the user; and display a first interface on which the semantic information is shown.
[0121] For example, the semantic information that identifies a user can be natural language information used to describe the user's characteristics. For instance, the semantic information that identifies a user can be related to technology, entertainment, etc.
[0122] The first interface is for editing semantic information that identifies users. The first interface may include one or more generated semantic information that identifies users.
[0123] By generating semantic information that identifies users, fragmented events can be elevated to user-level cognition, providing a basis for personalized services to the long-term memory system. Furthermore, visualizing this semantic information allows users to understand the long-term memory system's level of understanding, increasing transparency and trust. Users can also independently verify and correct this semantic information, forming a feedback loop. This enhances user experience and continuously refines and improves the understanding of the user.
[0124] In one scenario, the operation of processing the information set to obtain semantic information identifying the user can be as follows: extracting memory information from the information set, whereby the memory information includes one or more of contextual semantic information, event log information, and predictive memory information; and generating semantic information identifying the user based on the memory information.
[0125] Contextual semantic information is a narrative record with a timeline and background, which coherently describes a series of related facts that occur within a certain period of time. Contextual semantic information focuses on narrative and experiential aspects, and its parent refers to memory units. For example, contextual semantic information could be "On the afternoon of August 1st, in a coffee shop in Xuhui District, Shanghai, wearing a white T-shirt, I discussed travel plans for the second half of the year with a friend."
[0126] Event log information further decomposes contextual semantic information into single-sentence atomic facts and time indices, representing the smallest indivisible granularity of factual statements, with its parent pointing to contextual semantic information. For example, event log information could be "Zhang San's phone number is 138xxxx1234" or "The meeting will begin at 3 PM on Friday."
[0127] Predictive memory information consists of extracted plans, to-do items, commitments, or expected goals for the future, whose parent refers to a set of information or contextual semantic information. Predictive memory information can include multiple generated predictions (e.g., 10, 15, etc.), each prediction including a duration prediction (duration_days) and a start_time / end_time. For example, predictive memory information could be "The project report needs to be finished by next Wednesday."
[0128] Since each set of information corresponds to an independent event, and each event has contextual coherence, retrieving information from each set can yield relatively accurate and complete memory information. Integrating the extracted memory information from various sources can form a comprehensive semantic description of the user, enabling the long-term memory system to better perform functions such as proactive recommendations and reminders for the user.
[0129] In one scenario, the operation of a computer device to retrieve memory information from an information set can be as follows: for any one of one or more information sets, retrieve contextual memory from the information set to obtain contextual semantic information; and / or, retrieve event logs from the contextual semantic information to obtain event log information; and / or, predict the information set or contextual semantic information to obtain predictive memory information.
[0130] By first extracting contextual semantic information with macro-narrative characteristics, then extracting atomic facts from the contextual semantic information to form event log information, and simultaneously independently predicting future plans to form a hierarchical extraction logic for predictive memory information, the long-term memory system only needs to process single-dimensional information at each level, thus improving the accuracy of information extraction. By reusing contextual semantic information as the basis for downstream extraction, computational costs can be reduced. In addition, the hierarchical memory system supports incremental updates and multi-granularity retrieval, enabling the long-term memory system to flexibly meet users' needs for different scenarios such as narrative review, fact query, and schedule reminders.
[0131] In one scenario, where the information set is the first open unit, contextual memory information, event log information, and predictive memory information can be extracted based on the information set and the target historical information set.
[0132] In one scenario, a computer device can generate semantic information identifying a user based on memory information from one or more sets of information associated with the user identifier within a preset time window.
[0133] The period for generating semantic information that identifies users can be 1 day or 2 days, etc.
[0134] For example, the maximum number of memory information corresponding to one or more information sets can be a preset number. The preset number can be set in advance, such as 490, 500, or 510.
[0135] For example, the identifier of at least one set of memory information corresponding to the user can be stored in the semantic information identifying the user, and / or, the identifier of the user's semantic information can be stored in the semantic information of each of one or more sets of information. By storing the user's semantic information in association with the information sets, and storing the information sets in association with the memory information, the user can trace the source of evidence when subsequently viewing the memory information and the semantic information.
[0136] For example, the semantic information identifying a user may include explicit profile information and / or implicit profile information. Explicit profile information is the profile information constituted by information actively provided by the user, such as name, gender, age, occupation, and geographical location filled in during registration. Implicit profile information is the profile information obtained by processing the user's behavioral patterns, such as interests and preferences inferred from browsing history, dwell time, and click sequences.
[0137] The preset time window can be set in advance. For example, the start time of the preset time window is the time when the semantic information of the user identifier was last generated, and the end time is the current time. Alternatively, the start time can be the current time minus a preset multiple of the preset interval time, and the end time can be the current time.
[0138] By aggregating and analyzing the memory information generated by users over a period of time, multidimensional and traceable semantic information can be generated. This avoids biases in user profiles caused by limited memory of a few events, allowing for accurate updates based on user habits and preferences, thus ensuring the stability and accuracy of semantic information updates.
[0139] To facilitate understanding, the following will be combined with... Figure 7 The relationship between the aforementioned contextual semantic information, event log information, predictive memory information, and user profile information is illustrated by an example.
[0140] Figure 7 A diagram showing the parent-child reference and evidence source relationship of the hierarchical memory data model provided in this embodiment of the disclosure.
[0141] Figure 7 In this process, contextual memory information obtained by extracting contextual memories from the information set points to the information set itself, and the contextual memory information stores the identifier of the information set. Event log information obtained by extracting event logs from the contextual memory information points to the contextual memory information. Predictive memory information obtained by predicting contextual memory information points to the contextual memory information. Based on the contextual semantic information, event log information, and predictive memory information corresponding to the information set associated with the user identifier within a preset time window, user profile information is generated.
[0142] In one scenario, after generating semantic information to identify a user, the semantic information of the user generated this time can be marked as the latest, and the semantic information of the user generated last time can be marked as historical.
[0143] By employing versioned management and a timed consolidation mechanism for user semantic information, computational overhead is effectively controlled while ensuring the personalized depth of user semantic information, thereby providing reliable data for understanding users.
[0144] To facilitate understanding, the following will be combined with... Figure 8 The process of extracting the user profile information described above is illustrated by example.
[0145] Figure 8 This is a flowchart illustrating the user profile information extraction process provided in this embodiment of the disclosure.
[0146] Figure 8 In the process, it is determined whether the cumulative time has reached the minimum interval for extracting user profile information. After reaching the minimum interval, one or more memory information associated with the user identifier within a preset time window are obtained and input into the large language model. The number of memory information within the time window is less than or equal to 500 and ordered. The large language model extracts based on one or more memory information to obtain separate explicit user profile information and implicit user profile information. The user profile information is saved, and the user profile information extracted this time is marked as the latest user profile information (i.e., is_latest=true), and the user profile information extracted last time is marked as non-latest user profile information (i.e., is_latest=false).
[0147] In one case, the operation of generating a session based on a set of information corresponding to at least one event in response to a user’s session request may include the following steps (1) to (3).
[0148] Step (1): Identify the first keyword in the session request, and query the information set corresponding to at least one event based on the first keyword.
[0149] The first keyword is the keyword identified from the session request for retrieval. For example, queries can be performed using keyword search, vector search (i.e., semantic similarity), or a hybrid search (i.e., keyword + vector).
[0150] Step (2): When the query retrieves the first set of information that matches the first keyword, a session is generated based on the content of the first set of information and the session request.
[0151] The first information set is the set of information found that can fully respond to the session.
[0152] If the query returns the first set of information, it means that the conditions for generating a session are met. Therefore, a session can be generated based on the content of the first set of information and the session request.
[0153] Optionally, after querying the third information set based on the first keyword, the third information set can be determined by a large language model based on the session request and the third information set to fully respond to the session results; if the third information set can fully respond to the session results, the third information set is determined as the first information set.
[0154] By performing sufficiency checks, the validity of query results can be determined intelligently, providing a basis for whether to conduct further queries. When the third set of information is valid, unnecessary additional calculations can be avoided, saving computing resources.
[0155] Step (3): When no first information set matching the first keyword is found, generate one or more second keywords based on the first keyword; determine the second information set based on the first keyword and one or more second keywords; generate a session based on the content of the second information set and the session request.
[0156] For example, the second keyword can be generated by supplementing the first keyword using a large language model.
[0157] The second information set is the set of information obtained from the query that can fully respond to the session.
[0158] In this way, if the first set of information is not found, an operation is performed to determine a second set of information based on the first keyword and one or more second keywords, resulting in a set of information sufficient to respond to the session request. This ensures the accuracy and completeness of the retrieved information set, meeting user needs.
[0159] To facilitate understanding, the following will be combined with... Figure 9 The retrieval process of memory information provided in the embodiments of this application will be described by way of example. Figure 9 A flowchart of an intelligent body-based multi-turn retrieval provided in an embodiment of this disclosure.
[0160] Figure 9 In the process, after receiving a session request, the query module identifies the first keyword in the session request and encapsulates it into a query request, which is then sent to the retrieval orchestrator.
[0161] The retrieval orchestrator performs the first round of retrieval based on the received query request. The first round of retrieval process is as follows: the retrieval orchestrator sends a retrieval instruction to the hybrid retrieval service; after receiving the retrieval instruction, the hybrid retrieval service uses a hybrid retrieval of keyword + vector based on the first keyword to obtain 50 keyword retrieval results and 50 vector retrieval results. Then, it uses the RRF algorithm (k=60) to initially merge and rearrange these retrieval results, extracts the top 5 candidate retrieval results, and sends them to the retrieval orchestrator; the retrieval orchestrator inputs the first keyword and the 5 retrieval results into the large language model to perform sufficient answer detection. If the large language model indicates that the 5 retrieval results can fully answer the question, the retrieval orchestrator sends the 5 retrieval results to the query module; if the sufficient answer detection indicates that the 5 retrieval results cannot fully answer the question, it generates 3 complementary queries, each of which includes the second keyword.
[0162] The retrieval orchestrator sends three complementary query requests to the hybrid retrieval service to execute the second round of retrieval. Upon receiving the three complementary query requests, the hybrid retrieval service performs a hybrid retrieval, obtaining the retrieval results corresponding to the three complementary queries. These results are then merged and rearranged with five other retrieval results, and the top 20 results are sent to the retrieval orchestrator. The orchestrator then sends these top 20 results to the query module, which generates a session based on them.
[0163] In this embodiment, audio information is categorized based on event classification to obtain an information set corresponding to at least one event. In response to a user's session request, a session is generated based on the information set corresponding to at least one event and output to the user. Event classification transforms fragmented audio content into a structured information set organized by events. Generating a session based on the information set corresponding to at least one event significantly improves session accuracy, making responses more relevant to the user's actual experiences and needs, and enhancing the naturalness and intelligence of the interaction.
[0164] Figure 10 This is a schematic diagram of the structure of an information processing device provided in an embodiment of this disclosure. The device can be implemented as part or all of a computer device by software, hardware, or a combination of both. The computer device can be as described below. Figure 11 The computer equipment shown. See also Figure 10 The device includes a classification module 1001 and a generation module 1002.
[0165] The classification module 1001 is used to classify audio information based on event classification to obtain an information set corresponding to at least one event; The generation module 1002 is used to generate a session based on the information set corresponding to at least one event in response to a user's session request, and output the session to the user.
[0166] Optionally, the event classification may include at least one of the following: the event expressed by the audio information, the application scenario to which the audio information belongs, and the user preferences expressed by the audio information.
[0167] Optionally, the classification module 1001 is used for: Perform speech recognition on the audio information to obtain at least one text segment corresponding to the audio information; classify the at least one text segment based on event classification to obtain an information set corresponding to at least one event; or, Speech recognition is performed on the audio information to obtain at least one audio segment corresponding to the audio information; the at least one audio segment is classified based on event classification to obtain an information set corresponding to at least one event.
[0168] Optionally, the device further includes: The processing module is used to process the information set to obtain semantic information that identifies the user; The display module is used to display the first interface and show semantic information on the first interface.
[0169] Optionally, the generation module 1002 is used for: Identify the first keyword in the session request, and query the information set corresponding to at least one event based on the first keyword; When the query retrieves a first set of information that matches the first keyword, a session is generated based on the content of the first set of information and the session request. When no first information set matching the first keyword is found, one or more second keywords are generated based on the first keyword; a second information set is determined based on the first keyword and one or more second keywords; and a session is generated based on the content of the second information set and the session request.
[0170] Optionally, the processing module is used for: Memory retrieval is performed on the information set to obtain memory information, which includes one or more of episodic memory information, event log information, and predictive memory information. Semantic information identifying users is generated based on memory information.
[0171] Optionally, the device further includes: The tagging module is used to mark the semantic information identifying a user as up-to-date and the semantic information identifying a user generated last time as historical.
[0172] Optionally, the device further includes: The first associated storage module is used to associate and store memory information with information sets.
[0173] Optionally, the device further includes: The second associated storage module is used to associate and store the semantic information that identifies the user with the memory information.
[0174] In this embodiment, audio information is categorized based on event classification to obtain an information set corresponding to at least one event. In response to a user's session request, a session is generated based on the information set corresponding to at least one event and output to the user. Event classification transforms fragmented audio content into a structured information set organized by events. Generating a session based on the information set corresponding to at least one event significantly improves session accuracy, making responses more relevant to the user's actual experiences and needs, and enhancing the naturalness and intelligence of the interaction.
[0175] It should be noted that the semantic information extraction device provided in the above embodiments is only illustrated by the division of the above functional modules when extracting semantic information. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above.
[0176] The functional modules in the above embodiments can be integrated into one processing unit, or each functional module can exist as a separate physical processing unit, or two or more functional modules can be integrated into one processing unit. The processing unit can be implemented in hardware or software. Furthermore, the specific names of the functional modules are only for easy differentiation and are not intended to limit the scope of protection of the embodiments of this disclosure.
[0177] The semantic information extraction device and information processing method embodiments provided in the above embodiments belong to the same concept. The specific working process and technical effects of the functional modules in the above embodiments can be found in the method embodiment section, and will not be repeated here.
[0178] Figure 11 This is a schematic diagram of the structure of a computer device provided in an embodiment of this disclosure. Figure 11 As shown, the computer device 110 includes: a processor 1100, a memory 1101, and a computer program 1102 stored in the memory 1101 and executable on the processor 1100. When the processor 1100 executes the computer program 1102, it implements the steps in the semantic information extraction method in the above embodiments.
[0179] Computer device 110 can be a general-purpose computer device or a special-purpose computer device. In specific implementations, computer device 110 can be a desktop computer, portable computer, network server, handheld computer, mobile phone, tablet computer, wireless terminal device, communication equipment, or embedded device. This disclosure does not limit the type of computer device 110. Those skilled in the art will understand that... Figure 11 The computer device 110 is merely an example and does not constitute a limitation on the computer device 110. It may include more or fewer components than shown, or combine certain components, or different components, such as input / output devices, network access devices, etc.
[0180] Processor 1100 can be a central processing unit (CPU), while processor 110 can be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor.
[0181] In some embodiments, memory 1101 may be an internal storage unit of computer device 110, such as a hard disk or memory of computer device 110. In other embodiments, memory 1101 may be an external storage device of computer device 110, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., provided on computer device 1100. Furthermore, memory 1101 may include both internal and external storage units of computer device 110. Memory 1101 is used to store operating system, applications, boot loader, data, and other programs. Memory 1101 may also be used to temporarily store data that has been output or will be output.
[0182] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this disclosure.
[0183] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.
[0184] This disclosure also provides a computer-readable storage medium storing a computer program that, when executed by a processor, can implement the steps in the various method embodiments described above.
[0185] This disclosure provides a computer program product that, when run on a computer, causes the computer to perform the steps described in the various method embodiments above.
[0186] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the above method embodiments of this disclosure can be implemented by a computer program. The computer program can be stored in a computer-readable storage medium, and when the computer program is executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or some intermediate form. The computer-readable storage medium can include at least: any entity or device capable of carrying computer program code to a computer device, recording media, computer memory, read-only memory (ROM), random access memory (RAM), compact disc read-only memory (CD-ROM), magnetic tape, floppy disk, and optical data storage devices. The computer-readable storage medium mentioned in this disclosure can be a non-volatile storage medium; in other words, it can be a non-transient storage medium.
[0187] It should be understood that all or part of the steps of the above embodiments can be implemented by 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. A computer program product includes one or more computer instructions. The computer instructions can be stored in the above-described computer-readable storage medium.
[0188] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this disclosure.
[0189] In the embodiments provided in this disclosure, it should be understood that the disclosed apparatus / computer devices and methods can be implemented in other ways. For example, the apparatus / computer device embodiments described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices, or units, and may be electrical, mechanical, or other forms. Units described as separate components may or may not be physically separate, and components shown as units may or may not be physical units, i.e., they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this disclosure according to actual needs.
[0190] The embodiments described above are only used to illustrate the technical solutions of this disclosure, and are not intended to limit it. Although this disclosure has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this disclosure, and should all be included within the protection scope of this disclosure.
Claims
1. An information processing method, characterized in that, The method includes: Based on event classification, audio information is categorized to obtain at least one set of information corresponding to an event; In response to a user's session request, a session is generated based on the information set corresponding to the at least one event, and the session is output to the user.
2. The method according to claim 1, characterized in that, The event classification includes at least one of the following: the event expressed by the audio information, the application scenario to which the audio information belongs, and the user preferences expressed by the audio information.
3. The method according to claim 1 or 2, characterized in that, The process of classifying audio information based on event classification yields at least one set of information corresponding to an event, including: The audio information is subjected to speech recognition to obtain at least one text segment corresponding to the audio information; the at least one text segment is classified based on event classification to obtain an information set corresponding to at least one event; or, The audio information is subjected to speech recognition to obtain at least one audio segment corresponding to the audio information; the at least one audio segment is classified based on event classification to obtain an information set corresponding to the at least one event.
4. The method according to claim 1, characterized in that, The method further includes: The information set is processed to obtain semantic information that identifies the user; Display a first interface, on which the semantic information is shown.
5. The method according to claim 1, characterized in that, The step of generating a session in response to a user's session request, based on an information set corresponding to at least one event, includes: The first keyword in the session request is identified, and a query is performed in the information set corresponding to the at least one event based on the first keyword. When a first set of information matching the first keyword is obtained, a session is generated based on the content of the first set of information and the session request. When no first information set matching the first keyword is found, one or more second keywords are generated based on the first keyword; a second information set is determined based on the first keyword and the one or more second keywords; and a session is generated based on the content of the second information set and the session request.
6. The method according to claim 4, characterized in that, The process of processing the information set to obtain semantic information identifying the user includes: The memory information is obtained by memory retrieval from the information set, and the memory information includes one or more of episodic memory information, event log information, and predictive memory information. The semantic information identifying the user is generated based on the memory information.
7. The method according to claim 6, characterized in that, After generating the semantic information identifying the user based on the memory information, the method further includes: The semantic information identifying the user is marked as latest, and the semantic information identifying the user generated last time is marked as historical.
8. The method according to claim 6, characterized in that, After retrieving memory information from the information set, the method further includes: The memory information is associated with and stored in the information set; After generating the semantic information identifying the user based on the memory information, the method further includes: The semantic information identifying the user is associated with and stored in the memory information.
9. A computer device, characterized in that, The computer device includes a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the computer program, when executed by the processor, implements the method as claimed in any one of claims 1 to 8.
10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the method as described in any one of claims 1 to 8.
