Information processing method and device, equipment and medium

Abstract

The embodiment of the invention relates to an information processing method and device, equipment and a medium. The method comprises the steps that conversation content to be processed is acquired; performing semantic analysis processing on the dialogue content to obtain a semantic analysis result; based on the semantic analysis result, obtaining at least one fragment set from the dialogue content; the fragment set comprises at least one dialogue fragment, the dialogue fragments in the same fragment set correspond to the same event, and the dialogue fragments in different fragment sets correspond to different events; and carrying out aggregation processing on the dialogue fragments in the fragment set to obtain event related information corresponding to the fragment set, and storing the event related information corresponding to the fragment set. According to the embodiment of the invention, the dialogue information can be effectively concentrated, the waste of storage resources is avoided, the storage cost is reduced, a complete and coherent memory effect on the event can be ensured, the problem of memory fragmentation is improved, and subsequent efficient and convenient retrieval based on the stored event related information can be facilitated.

Description

Technical Field

[0001] This disclosure relates to the field of computer technology, and in particular to an information processing method, apparatus, device and medium. Background Technology

[0002] With the development of artificial intelligence, language models capable of engaging in AI-powered conversations with users have been widely applied in daily life and various industries. To provide a truly personalized, in-depth, and consistent user experience, it is necessary to store users' historical conversation content so that targeted communication services can be provided to users in the future based on this stored content. However, the inventors have found that the current methods for storing conversation content are inadequate and require improvement. Summary of the Invention

[0003] In order to solve the above-mentioned technical problems, or at least partially solve the above-mentioned technical problems, this disclosure provides an information processing method, apparatus, device and medium.

[0004] This disclosure provides an information processing method, the method comprising: acquiring dialogue content to be processed; performing semantic analysis on the dialogue content to obtain a semantic analysis result; obtaining at least one set of segments from the dialogue content based on the semantic analysis result; wherein the set of segments contains at least one dialogue segment, the dialogue segment being a part of the dialogue content; dialogue segments in the same set of segments correspond to the same event, and dialogue segments in different sets of segments correspond to different events; performing aggregation processing on the dialogue segments in the set of segments to obtain event-related information corresponding to the set of segments, and storing the event-related information corresponding to the set of segments.

[0005] Optionally, the step of performing semantic analysis on the dialogue content to obtain semantic analysis results includes: obtaining prompt information; wherein the prompt information is used to prompt: identifying a first type of content for filtering and a second type of content for retention from the dialogue content, and identifying dialogue fragments belonging to the same event from the second type of content; based on the prompt information, performing semantic analysis on the dialogue content through a target network model to obtain semantic analysis results.

[0006] Optionally, the prompt information is also used to prompt the model output format, wherein the model output format includes: event identifiers and the positions of the dialogue segments corresponding to the event identifiers in the dialogue content; obtaining at least one set of segments from the dialogue content based on the semantic analysis results includes: generating target information that conforms to the model output format through the target network model based on the semantic analysis results; retrieving the dialogue segments corresponding to each event identifier from the dialogue content based on the positions of the dialogue segments corresponding to each event identifier in the dialogue content, and obtaining at least one set of segments based on the dialogue segments corresponding to each event identifier.

[0007] Optionally, the model output format may further include event overview information corresponding to the event identifier; storing event-related information corresponding to the fragment set may include: associating and storing event-related information corresponding to the fragment set with event overview information corresponding to the event identifier.

[0008] Optionally, the step of aggregating the dialogue fragments in the fragment set to obtain event-related information corresponding to the fragment set includes: splicing the dialogue fragments in the fragment set to obtain splicing information corresponding to the fragment set; and obtaining event-related information corresponding to the fragment set based on the splicing information.

[0009] Optionally, obtaining the event-related information corresponding to the fragment set based on the splicing information includes: using the splicing information as the event-related information corresponding to the fragment set; or, performing content compression processing on the splicing information to obtain compressed information, and using the compressed information as the event-related information corresponding to the fragment set.

[0010] Optionally, storing the event-related information corresponding to the fragment set includes: associating the event-related information corresponding to the fragment set with the target account to which the dialogue content belongs and storing it in a preset database.

[0011] Optionally, the method further includes: in response to receiving the content to be answered corresponding to the target account, identifying the target event corresponding to the content to be answered; retrieving event-related information of the target event associated with the target account from the database, and generating response content corresponding to the content to be answered based on the retrieval results.

[0012] This disclosure also provides an information processing apparatus, comprising: a dialogue acquisition module for acquiring dialogue content to be processed; a semantic analysis module for performing semantic analysis processing on the dialogue content to obtain semantic analysis results; a set acquisition module for obtaining at least one fragment set from the dialogue content based on the semantic analysis results; wherein the fragment set contains at least one dialogue fragment, and the dialogue fragment is a part of the dialogue content; dialogue fragments in the same fragment set correspond to the same event, and dialogue fragments in different fragment sets correspond to different events; and an information aggregation and storage module for aggregating the dialogue fragments in the fragment set to obtain event-related information corresponding to the fragment set, and storing the event-related information corresponding to the fragment set.

[0013] This disclosure also provides an electronic device, comprising: a storage device storing a computer program thereon; and a processing device for executing the computer program in the storage device to implement the information processing method provided in this disclosure.

[0014] This disclosure also provides a computer-readable storage medium storing a computer program for performing the information processing method provided in this disclosure.

[0015] This disclosure also provides a computer program product, including a computer program that, when executed by a processor, implements the information processing method provided in this disclosure.

[0016] The technical solution provided in this disclosure can perform semantic analysis on the dialogue content to be processed, thereby obtaining at least one set of fragments from the dialogue content based on the semantic analysis results. Each set of fragments contains at least one dialogue fragment, which is a portion of the dialogue content. Dialogue fragments in the same set correspond to the same event, while dialogue fragments in different sets correspond to different events. The dialogue fragments in the set are then further aggregated to obtain event-related information corresponding to the set of fragments, and this event-related information is saved. This method does not directly store the entire dialogue content or simply segment the dialogue content according to dialogue turns. Instead, it identifies dialogue fragments corresponding to the same event through semantic analysis and aggregates them, storing them as events. This effectively condenses dialogue information, avoids wasting storage resources, reduces storage costs, and ensures a complete and coherent memory of events, improving the problem of fragmented memory. This facilitates efficient and convenient retrieval based on stored event-related information during subsequent communication with users, further improving the accuracy and reliability of subsequent responses.

[0017] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this disclosure, nor is it intended to limit the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0018] The accompanying drawings, which are incorporated in and form a part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure.

[0019] To more clearly illustrate the technical solutions in the embodiments of this disclosure or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 A flowchart illustrating an information processing method provided in an embodiment of this disclosure; Figure 2 This is a schematic diagram of an information processing flow provided in an embodiment of the present disclosure; Figure 3 This is a schematic diagram of an information processing flow provided in an embodiment of the present disclosure; Figure 4 This is a schematic diagram of the structure of an information processing device provided in an embodiment of the present disclosure; Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this disclosure. Detailed Implementation

[0021] To better understand the above-mentioned objectives, features, and advantages of this disclosure, the solutions disclosed herein will be further described below. It should be noted that, unless otherwise specified, the embodiments and features described herein can be combined with each other.

[0022] 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.

[0023] 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.

[0024] 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.

[0025] 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.

[0026] Numerous specific details are set forth in the following description in order to provide a full understanding of this disclosure, but this disclosure may also be implemented in other ways different from those described herein; obviously, the embodiments in the specification are only some, and not all, of the embodiments of this disclosure.

[0027] The inventors discovered that the dialogue content storage method used in related technologies is inadequate, as illustrated below: Some technologies save all of a user's dialogue content. In this storage method, the entire dialogue content between the user and the model in a single complete session is stored as a single memory unit. While this method preserves the most complete context, it also has significant drawbacks: First, it introduces a massive amount of noise information, such as a large amount of casual conversation and invalid questions and answers, which are indiscriminately stored in the database, causing the database to be "contaminated" by low-value information and containing a lot of "noisy memories." This not only results in high storage costs and a waste of storage resources, but these "noisy memories" also interfere with the subsequent retrieval of key information. Second, in a single complete session, a user may discuss multiple completely different events. This solution forcibly combines multiple unrelated events, greatly affecting the accuracy of subsequent retrieval, and consequently, the accuracy and reliability of subsequent responses. In summary, the above methods suffer from low signal-to-noise ratio, storage redundancy, and negative impacts on subsequent retrieval performance. Other technologies store each question-and-answer session between the user and the model (i.e., a single dialogue turn) as an independent memory unit. A significant drawback of this approach is the severe problem of context fragmentation. Understandably, an event typically requires multiple dialogue turns to be fully expressed. Treating each dialogue turn as an isolated memory artificially breaks down a coherent event into numerous discrete fragments. In other words, it mechanically cuts a potentially complete memory event (e.g., a complete discussion about project A) into several isolated, incomplete fragments. This results in subsequent retrievals of memory fragments with disjointed context, potentially retrieving only one fragment and failing to form a coherent and complete understanding of the event. Consequently, answers become one-sided and logically broken, severely impacting the user experience. In summary, based on the aforementioned storage methods, dialogue models will subsequently face two core challenges: low signal-to-noise ratio and fragmented memory content.

[0028] The aforementioned deficiencies in dialogue storage solutions in related technologies are the result of the applicant's practical experience and careful research. Therefore, the discovery process of these deficiencies and the solutions proposed in the embodiments of this disclosure below should be considered as contributions made by the applicant to this application. To improve at least one of the above problems, embodiments of this disclosure provide an information processing method, apparatus, device, and medium, which are described in detail below: Figure 1 This is a flowchart illustrating an information processing method provided in an embodiment of the present disclosure. The method can be executed by an information processing device, which can be implemented using software and / or hardware, and is generally integrated into an electronic device. Figure 1 As shown, the method mainly includes the following steps S102 to S108: Step S102: Obtain the dialogue content to be processed.

[0029] For example, the dialogue content to be processed can be the dialogue content between the user and the dialogue model. This dialogue content can be all the dialogue content from at least one complete session, or it can be the content from the most recent N dialogue rounds, where N is a preset integer. In some specific examples, the dialogue content to be processed can be passively received from the user, where one user corresponds to one account. In other examples, all the dialogue content from a complete session can be actively retrieved upon detecting the end of the user's session, and this content can be used as the dialogue content to be processed. Alternatively, when it is detected that the user has already had N dialogue rounds of interaction with the dialogue model, the content from those N rounds can be used as the dialogue content to be processed. It should be noted that in practical applications, the user can be explicitly prompted that in order to provide better dialogue services, existing dialogue content needs to be retrieved and processed. The user can then choose whether to agree based on the prompt information, and the dialogue content to be processed can be actively retrieved only after the user's authorization.

[0030] For ease of processing, the dialogue content obtained above can be dialogue text. If the dialogue between the user and the dialogue model is a voice dialogue, the voice content can be converted into dialogue text.

[0031] Step S104: Perform semantic analysis on the dialogue content to obtain the semantic analysis results.

[0032] In practical applications, a target network model can be used to perform semantic analysis on the dialogue content to obtain semantic analysis results. The target network model can be a pre-defined neural network model; for example, it can be implemented using a language model. To achieve better semantic analysis results, prompts can be set to indicate matters that the model needs to pay attention to during the semantic analysis process or the goals of the semantic analysis, etc., without limitation. For example, the above semantic analysis results include a first type of content for filtering and a second type of content for retention, as well as information such as dialogue fragments belonging to the same event within the second type of content.

[0033] Step S106: Based on the semantic analysis results, obtain at least one set of fragments from the dialogue content; wherein, the set of fragments contains at least one dialogue fragment, and the dialogue fragment is a part of the dialogue content; the dialogue fragments in the same set of fragments correspond to the same event, and the dialogue fragments in different sets of fragments correspond to different events.

[0034] In some examples, an event can be considered a communication topic. This disclosure does not limit the scope of events; any event involved in the communication is acceptable. The dialogue content between the user and the dialogue model may involve one or more events. Based on semantic analysis results, dialogue fragments corresponding to the same event can be clearly identified, resulting in at least one set of fragments, each set corresponding to one event. One question and one answer constitute one dialogue turn. The dialogue content between the user and the model typically includes multiple dialogue turns. In practical applications, for a given event, a dialogue fragment corresponding to that event can be the complete content of a dialogue turn involving that event, or it can be a portion of a dialogue turn involving that event, such as deleting content unrelated to the event from that turn. Specific settings are flexible and not limited here. Furthermore, multiple dialogue fragments in a fragment set can be discontinuous, but since the dialogue fragments in a fragment set correspond to the same event, the contextual information of that event can be presented relatively coherently.

[0035] Step S108: Aggregate the dialogue fragments in the fragment set to obtain the event-related information corresponding to the fragment set, and store the event-related information corresponding to the fragment set.

[0036] For each fragment set, all dialogue fragments in that fragment set are aggregated. The aggregation result can be used as event-related information corresponding to that fragment set. This disclosure does not limit the aggregation process; for example, it can be a simple concatenation of all dialogue fragments in the fragment set, or it can be information compression of all dialogue fragments in the fragment set. The event-related information obtained through the above methods is the extracted event memory. If the dialogue content involves M events, then M event memories will be obtained, each event memory being an aggregated piece of information related to that event. For ease of understanding, please also refer to... Figure 2 The diagram illustrates an information processing flow. The historical dialogue content is the dialogue content to be processed mentioned above. The information processing flow can be regarded as a memory extraction strategy. Through semantic analysis, intelligent segmentation (i.e., identifying dialogue segments corresponding to the same event), segment aggregation, and other operations, event memory 1, event memory 2, ... event memory M can be extracted from the historical dialogue content. Then, the event memories can be stored in a database (also known as a memory bank) and stored in units of events, forming a complete and logically coherent memory corresponding to each event. The above method does not directly store the entire dialogue content or simply segment the dialogue content according to the dialogue rounds. Instead, it identifies dialogue fragments corresponding to the same event through semantic analysis and aggregates them, storing them as events. This effectively condenses dialogue information, avoids wasting storage resources, reduces storage costs, and ensures a complete and coherent memory of events, improving the problem of fragmented memory. It also helps to efficiently and conveniently retrieve relevant event information when communicating with users later, further improving the accuracy and reliability of subsequent responses.

[0037] In some implementations, step S104 above, that is, the step of performing semantic analysis on the dialogue content to obtain the semantic analysis result, can be performed with reference to steps a and b below: Step a, obtain prompt information; wherein, the prompt information is used to prompt: identify a first type of content for filtering and a second type of content for retention from the dialogue content, and identify dialogue fragments belonging to the same event from the second type of content.

[0038] The aforementioned prompts, also known as model prompts, can be implemented through instruction engineering. In practical applications, the prompts can provide characteristics of the first type of content and the second type of content, and / or provide enumerated examples of the first and second types of content, so that the model can accurately distinguish between the first and second types of content in the dialogue. For example, the first type of content includes low-information-density content such as daily greetings, emotional responses, and meaningless small talk, or content without long-term value. The second type of content includes high-information-density content such as factual information, preferences, important events, and clear conclusions, or content with long-term value. The definition of whether an event is important can be achieved by enumerating or providing the characteristics of important events. For example, events emphasizing commitments, agreements, or proposals that may be retroactively addressed in the future, or requests made by specific roles, can all be considered important events. The above are just examples; in practical applications, the characteristics of the first and second types of content can be flexibly set according to needs. It should be noted that the above prompts not only instruct the model to identify the second type of content that needs to be retained, but also explicitly instruct the model to identify the first type of content that needs to be filtered, that is, to provide a reverse constraint, which ensures from the source that the information finally stored (that is, the generated event memory) has a high signal-to-noise ratio.

[0039] Step b: Based on the prompts, the target network model performs semantic analysis on the dialogue content to obtain the semantic analysis results. The prompts clearly guide the target network model to process the dialogue content in the expected way and obtain the desired results. The target network model can identify which parts of the dialogue content belong to the first category and which belong to the second category through semantic analysis, and can label these contents. Furthermore, it can identify dialogue segments belonging to the same event from the second category content and label the dialogue segments with corresponding event tags.

[0040] In some implementation examples, the prompt information is also used to prompt the model output format, wherein the model output format includes: event identifier and the position of the dialogue segment corresponding to the event identifier in the dialogue content, the position specifically including the start position and end position of the dialogue segment, the above position can be represented by the turn number of the dialogue segment in the dialogue content; based on this, the above step S106, that is, the step of obtaining at least one set of segments from the dialogue content based on the semantic analysis results, can be performed with reference to the following steps (1) and (2): Step (1): Based on the semantic analysis results, target information consistent with the model output format is generated through the target network model. In other words, the target information includes the event identifiers corresponding to each fragment set, and the positions of each dialogue fragment corresponding to the event identifier in the dialogue content, and the positions of multiple dialogue fragments in a fragment set may be discontinuous. The above model output format is a structured format, and the target information is also structured information. For example, the target information is represented as {"event identifier", "position information of dialogue fragment 1", "position information of dialogue fragment 2"... "position information of dialogue fragment n"}, etc. In addition, the model output format can also include other content, such as the event summary information corresponding to the event identifier. In this case, the target information is represented as {"event identifier", "event summary information", "position information of dialogue fragment 1", "position information of dialogue fragment 2"... "position information of dialogue fragment n"}. In practical applications, the model output format can be flexibly set according to needs so that the target network model outputs content that meets the requirements according to the model output format. Through the above method, an accurate and easy-to-operate "data map" is provided for subsequent integration of fragmented memories.

[0041] Step (2): Based on the position of the dialogue fragments corresponding to each event identifier in the target information within the dialogue content, retrieve the dialogue fragments corresponding to each event identifier from the dialogue content, and obtain at least one fragment set based on the dialogue fragments corresponding to each event identifier. It should be noted that this embodiment only requires the model to output the position of the dialogue fragments, without requiring the model to output the complete dialogue fragments. This method can effectively reduce the amount of information processing required by the model and improve model processing efficiency. Based on the position of the dialogue fragments output by the model, the corresponding dialogue fragments can be directly retrieved from the dialogue content, and dialogue fragments corresponding to the same event identifier can be combined into a fragment set.

[0042] If the model output format also includes event overview information corresponding to the event identifier, that is, the target information also includes event overview information corresponding to the event identifier, then the step of storing event-related information corresponding to the fragment set in step S108 above includes: associating the event-related information corresponding to the stored fragment set with the event overview information corresponding to the event identifier. This approach helps to quickly obtain the event overview involved in the fragment set associated with the event identifier based on the event overview information, and obtain specific event-related information. Furthermore, it allows for convenient acquisition of event-related information for the fragment set of events involved in the dialogue based on the event overview information, facilitating subsequent improvement in retrieval efficiency. It should be noted that event-related information is the specific event information obtained by aggregating dialogue fragments, while event overview information can be a brief description of the event by the model, further clarifying the event represented by the event identifier.

[0043] It should be noted that, based on the above approach, this embodiment innovatively uses a neural network model (such as a large language model) itself as the core intelligent engine for extracting the required information from historical dialogue content. By inputting a dialogue content to be processed into the model, and supplementing it with a carefully designed set of instructions containing specific constraints and output format requirements (i.e., the aforementioned prompts), the model is guided to simultaneously complete two key tasks in one processing cycle: first, to make a value judgment on the dialogue content and filter out worthless segments; second, to accurately identify and locate all multiple dialogue segments related to a specific event that may be discontinuous in the historical dialogue content. Finally, based on the structured results returned by the model, the algorithm integrates these fragmented information into several logically complete and information-condensed core event memories.

[0044] In some implementations, the step S108 above, which involves aggregating dialogue fragments in the fragment set to obtain event-related information corresponding to the fragment set, can be performed with reference to steps A and B below: Step A: Concatenate the dialogue fragments in the fragment set to obtain the concatenation information corresponding to the fragment set.

[0045] Step B involves obtaining event-related information corresponding to the fragment set based on the spliced ​​information. The event-related information obtained through this splicing process presents the event context relatively completely and coherently, aiding in subsequent retrieval. In some examples, the spliced ​​information can be directly used as the event-related information corresponding to the fragment set, resulting in more comprehensive event-related information. In other examples, the spliced ​​information can be compressed to obtain compressed information, which can then be used as the event-related information corresponding to the fragment set. This method yields more concise event-related information and further reduces storage costs. In practical applications, the above methods can be flexibly selected according to requirements, without any restrictions.

[0046] The step of storing the event-related information corresponding to the fragment set in step S108 above can be performed as follows: Associate the event-related information corresponding to the fragment set with the target account to which the dialogue content belongs and store it in a preset database. The target account to which the dialogue content belongs is also the account of the user corresponding to the dialogue content. In practical applications, the preset database can also be called a memory, which can be used to store dialogue-related information of multiple users using the dialogue model, that is, to store event-related information corresponding to multiple accounts. By using the above method, the obtained event-related information can be stored under the target account to which the dialogue content belongs, so that when engaging in dialogue with the user of the target account later, the corresponding event-related information can be retrieved in a targeted manner, providing a more accurate dialogue service.

[0047] Based on the foregoing, the method provided in this disclosure further includes the following steps 1) and 2). Step 1) In response to receiving the content to be answered corresponding to the target account, identify the target event corresponding to the content to be answered. In some examples, a dialogue model or a target network model can be used to perform semantic analysis on the content to be answered, thereby identifying the target event corresponding to the content to be answered. The dialogue model and the target network model can be the same model or different models, which is not restricted here.

[0048] Step 2) Retrieve event-related information associated with the target account from the database and generate response content corresponding to the content to be responded to based on the retrieval results. If event-related information associated with the target account is already stored in the database, the response is based on the retrieved event-related information. If event-related information associated with the target account is not already stored in the database, the target event may be an event first raised by the user, and the response can be based on the dialogue context and / or other databases, without any restrictions.

[0049] To facilitate understanding of the information processing method proposed in the embodiments of this disclosure, please refer to... Figure 3The diagram illustrates an information processing flow, showing that the aforementioned target network model is implemented using a large language model. Prompt information and historical dialogue content can be provided to the large language model, constraining its output to produce structured data (i.e., the aforementioned target information), such as event 1 - overview - dialogue fragments {start position: 10, end position: 12}, {start position: 14, end position: 16}, {start position: 20, end position: 28}, etc. The start and end positions of these dialogue fragments are represented by the dialogue round numbers. Subsequently, based on the model output, dialogue fragments corresponding to each event can be retrieved from the dialogue content and combined to obtain corresponding fragment sets. Further aggregation processing of the dialogue fragments within each fragment set yields relevant information for each event, also known as extracted event memory, which is then stored in a database for subsequent retrieval and response.

[0050] The information processing method proposed in this disclosure can be considered an innovative dialogue flow processing strategy (also known as a memory extraction algorithm), capable of deep semantic understanding and structured analysis of the original dialogue content. First, this strategy accurately assesses the information value of each dialogue segment, automatically filtering and discarding meaningless dialogue content such as casual conversation and emotional echoing, retaining only event information that constitutes valid memory. Second, this strategy overcomes the limitations of the natural temporal order of dialogue, intelligently identifying discontinuous segments that belong to the same event but are interrupted by other topics in the dialogue. Then, it associates and aggregates these scattered "memory fragments," integrating them into a logically clear and contextually complete core memory. Through this "filter first, then reassemble" intelligent segmentation mode, it effectively solves the two major problems of low memory signal-to-noise ratio and memory fragmentation in related technologies, efficiently and accurately extracting truly valuable complete memory content from complex and intertwined dialogues. The above processing strategy no longer relies on fixed time or structural units, but dynamically identifies the core event flow that runs through the entire dialogue through deep semantic analysis. It can intelligently filter noise and accurately "stitch together" dialogue fragments that are not continuous in time but belong to the same event. This can ensure that the information is highly condensed and that the semantic and logical integrity is guaranteed, which helps to achieve the expected memory retrieval effect and can effectively enhance long-term memory capacity.

[0051] As mentioned earlier, in order to extract long-term memory from user-model dialogue interactions, related technologies face an inherent contradiction in the selection of "granularity" in the division of memory extraction units: overly fine granularity leads to a lack of context in memory, while overly coarse granularity introduces a large amount of noise and irrelevant information. Related technologies struggle to achieve an effective balance between memory integrity and information signal-to-noise ratio. Therefore, dialogue models face two core challenges: low information signal-to-noise ratio and fragmented memory content. The method provided by the embodiments of this disclosure significantly improves these problems, helping to resolve the core contradictions of difficulty in determining memory extraction granularity and incomplete memory caused by topic overlap. It also helps to solve the two core challenges of low information signal-to-noise ratio and fragmented memory content faced by related technologies when processing real dialogues. It mainly has at least one of the following innovative points: (1) By utilizing target network models such as language models as the intelligent engine for memory extraction, a paradigm shift from "rule-based segmentation" to "semantic understanding" is achieved. In related technologies, whether segmentation is done by round or by conversation, it is essentially a kind of segmentation based on external rules or a shallow, mechanical approach, which cannot truly understand the inherent logical connections of the dialogue content. The embodiments disclosed in this disclosure overturn this approach, leveraging the model's powerful contextual understanding, information summarization, and instruction-following capabilities to directly apply the model to the task of memory extraction itself. In this way, the granularity of memory extraction is no longer preset or fixed, but dynamically determined by the model's semantic analysis results of the dialogue content, thereby enabling a more accurate and flexible capture of the true core of the memory.

[0052] (2) A dual-task instruction (i.e., the aforementioned prompt information) with both "value filtering" and "intelligent segmentation" functions was designed. This instruction not only requires the model to extract memories, but also strictly defines its behavior and output format, mainly including two levels: 1) Value filtering constraint: The instruction will explicitly require the model to only identify and extract content with long-term value, such as factual information, user preferences, important events, and clear conclusions; at the same time, the instruction will give a reverse constraint, requiring the model to ignore low-information-density dialogue content such as daily greetings, emotional echoes, and meaningless chatter, thus ensuring a high signal-to-noise ratio in the final generated memory from the source. 2) Intelligent segmentation and structured output: The instruction requires the model to perform intelligent segmentation based on semantic analysis results, identify dialogue fragments corresponding to the same event, and return the required information in a structured form, and explicitly limits the output format of the model, such as requiring the model to return event identifiers, the position of the dialogue fragments corresponding to the event identifiers in the dialogue content, and event overview information, etc. This mechanism provides a precise and operable "data map" for subsequent integration of fragmented memories.

[0053] (3) Automatic integration and generation of fragmented memories based on "intelligent fragmentation" output is achieved. Traditional solutions cannot solve the problem of incomplete memory caused by overlapping topics, while the present invention effectively overcomes this difficulty through the above-mentioned structured output. After obtaining dialogue fragments corresponding to the same event, these dialogue fragments that are scattered in terms of dialogue time but logically belong to the same event can be automatically aggregated. Even if the user intersperses two topics, event B and event C, during the discussion of event A, the present invention can still accurately link all dialogue content about event A, and finally form a memory about event A with complete context and logical coherence. This fundamentally solves the problem of fragmented memory and ensures that each extracted event memory has a high degree of integrity and usability.

[0054] The methods provided in this disclosure can bring at least one of the following significant beneficial effects to model-based dialogue systems: (1) Significantly reduces memory storage costs and improves management efficiency. This embodiment of the present disclosure, by setting up an intelligent filtering mechanism at the source of memory extraction, can proactively identify and discard casual conversation and polite remarks with low information value density in dialogues, ensuring that only truly meaningful information is stored in the long-term memory bank (i.e., the database), fundamentally avoiding redundant storage of invalid information. Compared to traditional full or near-full storage methods, this embodiment of the present disclosure can significantly reduce the storage space occupied by the memory bank, reducing long-term operating hardware costs and data management complexity.

[0055] (2) Significantly improves the accuracy and recall effect of memory retrieval. This disclosure improves retrieval quality in two dimensions: First, due to the filtering of a large amount of noise information, the signal-to-noise ratio of the memory bank is extremely high. During retrieval, it can effectively avoid interference from irrelevant content, making the recall results more relevant to the user's query intent, thus significantly improving recall accuracy. Second, since this disclosure has pre-integrated fragmented memories of the same event into complete memory units, it helps the retrieval system recall all necessary context about the event at once, avoiding problems such as missing key information due to incomplete memory in related technologies.

[0056] (3) It fundamentally improves the generation quality of downstream models and user experience. Because the memory context provided to the model for generating answers is of higher quality (both pure and complete), the model's performance will be improved in all aspects. Specifically, the model's answers will be more accurate and more consistent with the facts; the logic of the dialogue will be more coherent, and it will be able to reason and respond based on a complete event background; the model can understand the events involved in the dialogue more comprehensively, thereby providing more in-depth and personalized interaction, which can greatly improve the user experience.

[0057] Corresponding to the aforementioned information processing method, this disclosure further provides an information processing apparatus. Figure 4 This is a schematic diagram of the structure of an information processing device provided in an embodiment of the present disclosure. The device can be implemented by software and / or hardware, and is generally integrated into an electronic device, such as... Figure 4 As shown, the information processing device includes: Dialogue acquisition module 402 is used to acquire dialogue content to be processed; Semantic analysis module 404 is used to perform semantic analysis on the dialogue content and obtain semantic analysis results; The set acquisition module 406 is used to obtain at least one set of fragments from the dialogue content based on the semantic analysis results; wherein, the set of fragments contains at least one dialogue fragment, and the dialogue fragment is a part of the dialogue content; the dialogue fragments in the same set of fragments correspond to the same event, and the dialogue fragments in different sets of fragments correspond to different events; The information aggregation and storage module 408 is used to aggregate dialogue fragments in the fragment set, obtain event-related information corresponding to the fragment set, and save the event-related information corresponding to the fragment set.

[0058] The aforementioned device does not directly store the entire dialogue content or simply segment the dialogue content according to dialogue turns before storing it. Instead, it identifies dialogue fragments corresponding to the same event through semantic analysis and aggregates them, storing them as events. This effectively condenses dialogue information, avoids wasting storage resources, reduces storage costs, and ensures a complete and coherent memory of events, improving the problem of fragmented memory. It also helps to efficiently and conveniently retrieve relevant event information when communicating with users later, further improving the accuracy and reliability of subsequent responses.

[0059] In some implementations, the semantic analysis module 404 is specifically used to: obtain prompt information; wherein the prompt information is used to prompt: identify a first type of content for filtering and a second type of content for retention from the dialogue content, and identify dialogue fragments belonging to the same event from the second type of content; based on the prompt information, perform semantic analysis processing on the dialogue content through a target network model to obtain semantic analysis results.

[0060] In some implementations, the prompt information is also used to prompt the model output format, wherein the model output format includes: event identifiers and the positions of the dialogue segments corresponding to the event identifiers in the dialogue content; the set acquisition module 406 is specifically used to: generate target information that conforms to the model output format through the target network model based on the semantic analysis results; retrieve the dialogue segments corresponding to each event identifier from the dialogue content based on the positions of the dialogue segments corresponding to each event identifier in the dialogue content, and obtain at least one set of segments based on the dialogue segments corresponding to each event identifier.

[0061] In some implementations, the model output format further includes event overview information corresponding to the event identifier; the information aggregation and storage module 408 is specifically used to: associate and store event-related information corresponding to the fragment set with the event overview information corresponding to the event identifier.

[0062] In some implementations, the information aggregation and storage module 408 is specifically used to: splice dialogue fragments in the fragment set to obtain splicing information corresponding to the fragment set; and obtain event-related information corresponding to the fragment set based on the splicing information.

[0063] In some embodiments, the information aggregation and storage module 408 is specifically used to: use the spliced ​​information as event-related information corresponding to the fragment set; or, perform content compression processing on the spliced ​​information to obtain compressed information, and use the compressed information as event-related information corresponding to the fragment set.

[0064] In some implementations, the information aggregation and storage module 408 is specifically used to: associate the event-related information corresponding to the fragment set with the target account to which the dialogue content belongs and store it in a preset database.

[0065] In some embodiments, the apparatus further includes: In response to receiving the pending response content corresponding to the target account, the target event corresponding to the pending response content is identified; The database is used to retrieve event-related information associated with the target account and the target event, and response content is generated based on the retrieval results.

[0066] The information processing apparatus provided in this disclosure can execute the information processing method provided in any embodiment of this disclosure, and has the corresponding functional modules and beneficial effects for executing the method.

[0067] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the above-described device embodiments can be referred to the corresponding process in the method embodiments, and will not be repeated here.

[0068] This disclosure provides an electronic device, which includes: a storage device storing a computer program thereon; and a processing device for executing the computer program in the storage device to implement the steps of any method of this disclosure.

[0069] The following is for reference. Figure 5 This document illustrates a structural schematic diagram of an electronic device 500 suitable for implementing embodiments of the present disclosure. The terminal devices in these embodiments may include, but are not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Devices), PMPs (Portable Media Players), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital televisions and desktop computers. Figure 5 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments disclosed herein.

[0070] like Figure 5 As shown, the electronic device 500 may include a processing unit (e.g., a central processing unit, a graphics processing unit, etc.) 501, which can perform various appropriate actions and processes according to a program stored in read-only memory (ROM) 502 or a program loaded from storage device 508 into random access memory (RAM) 503. The RAM 503 also stores various programs and data required for the operation of the electronic device 500. The processing unit 501, ROM 502, and RAM 503 are interconnected via a bus 504. An input / output (I / O) interface 505 is also connected to the bus 504.

[0071] Typically, the following devices can be connected to I / O interface 505: input devices 506 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 507 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 508 including, for example, magnetic tapes, hard disks, etc.; and communication devices 509. Communication device 509 allows electronic device 500 to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 5 An electronic device 500 with various devices is shown; however, it should be understood that it is not required to implement or possess all of the devices shown. More or fewer devices may be implemented or possessed alternatively.

[0072] In particular, according to embodiments of this disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this disclosure include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device 509, or installed from a storage device 508, or installed from a ROM 502. When the computer program is executed by the processing device 501, it performs the functions defined in the methods of embodiments of this disclosure.

[0073] In addition to the methods and devices described above, embodiments of this disclosure can also be computer program products, comprising computer program instructions that, when executed by a processor, cause the processor to perform the methods provided in the embodiments of this disclosure. The computer program product can be written in any combination of one or more programming languages ​​to perform the operations of the embodiments of this disclosure. These programming languages ​​include object-oriented programming languages ​​such as Java and C++, as well as conventional procedural programming languages ​​such as C or similar languages. The program code can be executed entirely on a user computing device, partially on a user device, as a standalone software package, partially on a user computing device and partially on a remote computing device, or entirely on a remote computing device or server.

[0074] Furthermore, embodiments of this disclosure may also be computer-readable storage media storing computer program instructions that, when executed by a processor, cause the processor to perform the methods provided in embodiments of this disclosure.

[0075] The computer-readable storage medium may be any combination of one or more readable media. A readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may, for example, include, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatuses, or devices, or any combination thereof. More specific examples of readable storage media (a non-exhaustive list) include: electrical connections having one or more wires, portable disks, hard disks, random access memory, read-only memory, erasable programmable read-only memory, optical fibers, portable compact disk read-only memory, optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0076] This disclosure also provides a computer program product, including a computer program / instructions that, when executed by a processor, implement the methods described in this disclosure.

[0077] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0078] The above description is merely a specific embodiment of this disclosure, enabling those skilled in the art to understand or implement it. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this disclosure. Therefore, this disclosure is not to be limited to the embodiments described herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. An information processing method, characterized in that, include: Get the content of the dialogue to be processed; The dialogue content is subjected to semantic analysis to obtain semantic analysis results; Based on the semantic analysis results, at least one set of segments is obtained from the dialogue content; wherein, the set of segments contains at least one dialogue segment, and the dialogue segment is a part of the dialogue content; the dialogue segments in the same set of segments correspond to the same event, and the dialogue segments in different sets of segments correspond to different events; The dialogue fragments in the fragment set are aggregated to obtain event-related information corresponding to the fragment set, and the event-related information corresponding to the fragment set is stored.

2. The method according to claim 1, characterized in that, The semantic analysis processing of the dialogue content to obtain the semantic analysis result includes: Obtain prompt information; wherein the prompt information is used to prompt: identify a first type of content for filtering and a second type of content for retention from the dialogue content, and identify dialogue fragments belonging to the same event from the second type of content; Based on the prompt information, the dialogue content is semantically analyzed using a target network model to obtain the semantic analysis results.

3. The method according to claim 2, characterized in that, The prompt information is also used to prompt the model output format, wherein the model output format includes: an event identifier and the position of the dialogue segment corresponding to the event identifier in the dialogue content; The step of obtaining at least one set of fragments from the dialogue content based on the semantic analysis results includes: Based on the semantic analysis results, target information that conforms to the output format of the target network model is generated. Based on the position of the dialogue fragments corresponding to each event identifier in the target information within the dialogue content, the dialogue fragments corresponding to each event identifier are retrieved from the dialogue content, and at least one fragment set is obtained based on the dialogue fragments corresponding to each event identifier.

4. The method according to claim 3, characterized in that, The model output format also includes event overview information corresponding to the event identifier; the storage of event-related information corresponding to the fragment set includes: The event-related information corresponding to the fragment set is associated with the event overview information corresponding to the event identifier.

5. The method according to any one of claims 1 to 4, characterized in that, The aggregation process of dialogue fragments in the fragment set to obtain event-related information corresponding to the fragment set includes: The dialogue fragments in the fragment set are spliced ​​together to obtain the splicing information corresponding to the fragment set. Based on the splicing information, the event-related information corresponding to the fragment set is obtained.

6. The method according to claim 5, characterized in that, The step of obtaining event-related information corresponding to the fragment set based on the splicing information includes: The spliced ​​information is used as the event-related information corresponding to the fragment set; or, the spliced ​​information is compressed to obtain compressed information, and the compressed information is used as the event-related information corresponding to the fragment set.

7. The method according to claim 1, characterized in that, The storage of event-related information corresponding to the fragment set includes: The event-related information corresponding to the fragment set is associated with the target account to which the dialogue content belongs and stored in a preset database.

8. The method according to claim 7, characterized in that, The method further includes: In response to receiving the pending response content corresponding to the target account, the target event corresponding to the pending response content is identified; The database is used to retrieve event-related information associated with the target account and the target event, and response content is generated based on the retrieval results.

9. An information processing device, characterized in that, include: The dialogue acquisition module is used to acquire the dialogue content to be processed. The semantic analysis module is used to perform semantic analysis on the dialogue content and obtain semantic analysis results. The set acquisition module is used to obtain at least one set of segments from the dialogue content based on the semantic analysis results; wherein, the set of segments contains at least one dialogue segment, and the dialogue segment is a part of the dialogue content; the dialogue segments in the same set of segments correspond to the same event, and the dialogue segments in different sets of segments correspond to different events; The information aggregation and storage module is used to aggregate dialogue fragments in the fragment set, obtain event-related information corresponding to the fragment set, and store the event-related information corresponding to the fragment set.

10. An electronic device, characterized in that, The electronic device includes: A storage device on which computer programs are stored; A processing device for executing the computer program in the storage device to implement the steps of the information processing method according to any one of claims 1-8.

11. A computer-readable storage medium, characterized in that, The storage medium stores a computer program for executing the information processing method according to any one of claims 1-8.

12. A computer program product, characterized in that, It includes a computer program that, when executed by a processor, implements the information processing method according to any one of claims 1-8.

Images