Dialogue memory priority system based on multi-dimensional weighting

Abstract

The invention discloses a dialogue memory priority system based on multidimensional weighting, and the system comprises the following modules: a multi-factor scoring module which is used for calculating the importance score of dialogue memory, and multiple factors comprise a time attenuation factor, a semantic correlation factor, an emotion intensity factor and a user feedback factor; the hierarchical storage decision module is used for distributing memories to corresponding storage hierarchies according to importance scores; the dynamic weight adjusting module is used for automatically adjusting the weight of each factor according to the dialogue mode; and the memory retrieval priority ranking module is used for determining the ranking of the retrieval results based on the multi-dimensional scores. According to the invention, core information and secondary information can be distinguished conveniently, and waste of memory resources is avoided; key information of the dialogue context is captured more accurately, so that the memory retrieval efficiency and the system performance are optimized; and the method adapts to different dialogue scenes and user interaction modes.

Description

Technical Field

[0001] The present invention relates to the technical field of artificial intelligence and natural language processing, and in particular to a dialogue memory priority system based on multi-dimensional weighting. Background Art

[0002] Current human-computer dialogue systems generally use fixed-length dialogue history windows or simple queue mechanisms for memory management, such as ring buffers or sliding window techniques. These approaches have significant limitations: First, the mechanical elimination of information can lead to the loss of critical context (such as long-term user preferences or dialogue goals); second, the storage of raw, unfiltered dialogue history introduces significant noise, reducing the accuracy of system responses. While some research has attempted to mitigate this problem by expanding memory capacity or improving vector retrieval, these efforts have not fundamentally broken through the technical paradigm of prioritizing storage over management.

[0003] Memory enhancement schemes proposed in recent years have three core flaws: (1) The importance assessment dimension is single, relying solely on word frequency statistics or time decay factors, which cannot identify implicit intentions in conversations; (2) Priority calculation lacks dynamics. For example, the traditional TF-IDF algorithm has difficulty capturing sudden topic shifts; (3) The storage architecture is rigid and lacks a hierarchical mechanism that matches memory value. As a result, high-value information (such as user taboos) may be overwritten by ordinary chat content. Experiments show that when the number of conversation turns exceeds 50, the recall rate of key information in existing systems drops sharply to below 37%.

[0004] Overall, existing technologies have yet to address four core contradictions in memory management: balancing information integrity and system load, adapting static evaluation criteria to dynamic conversational scenarios, mapping discrete memory units to continuous semantic associations, and integrating universal frameworks with specific domain characteristics. In particular, in specialized scenarios like medical consultation and legal advisory, conversational systems often fail to grasp key terms or misjudge context due to their lack of fine-grained memory management capabilities, severely hindering their practical application. Summary of the Invention

[0005] In order to overcome the above-mentioned shortcomings of the existing technology, the present invention proposes a dialogue memory priority system based on multi-dimensional weighting, which facilitates the distinction between core information and secondary information and avoids the waste of memory resources; more accurately captures the key information of the dialogue context to optimize memory retrieval efficiency and system performance; and adapts to different dialogue scenarios and user interaction modes.

[0006] The technical solution adopted by the present invention to solve the technical problem is: a multi-dimensional weighted dialogue memory priority system, including the following modules:

[0007] A multi-factor scoring module is used to calculate the importance score of conversation memory, wherein the multi-factors include a time decay factor, a semantic relevance factor, a sentiment intensity factor, and a user feedback factor;

[0008] Tiered storage decision module, which allocates memory to the corresponding storage tier based on importance score;

[0009] Dynamic weight adjustment module, automatically adjusts the weight of each factor according to the conversation mode;

[0010] The memory retrieval priority ranking module determines the ranking of retrieval results based on multi-dimensional scoring.

[0011] Furthermore, the multi-factor scoring module includes a time decay factor calculation unit, a semantic relevance factor calculation unit, an emotion intensity factor calculation unit, and a user comprehensive scoring unit; the time decay factor calculation unit is used to calculate the time decay factor of the memory item, and reduce the weight of the memory item that is far away in time; the semantic relevance factor calculation unit is used to calculate the semantic relevance factor between the memory item and the current dialogue context, and increase the weight of the relevant memory item; the emotion intensity factor calculation unit is used to calculate the emotion intensity factor of the memory item, and increase the weight of the memory item with strong emotion. The emotion intensity analysis can be performed online or offline. Offline emotion analysis is suitable for scenarios where historical dialogue data is preprocessed and emotion tags are extracted; the user comprehensive scoring unit performs a weighted combination of the time decay factor, the semantic relevance factor, and the emotion intensity factor, and calculates the comprehensive importance score of the memory item according to the multi-factor scoring algorithm, which is the importance score of the dialogue memory.

[0012] Furthermore, the tiered storage decision module includes:

[0013] Scoring threshold setting unit: used to set scoring thresholds for different storage tiers; scoring thresholds can be set in advance or dynamically adjusted according to system operation conditions;

[0014] Storage tier allocation unit: Allocates memory items to the corresponding storage tier based on their comprehensive importance score and score threshold. The storage tier can use in-memory data structures or persistent storage systems, depending on system performance and data persistence requirements.

[0015] Furthermore, the dynamic weight adjustment module includes:

[0016] Dialogue pattern analysis unit: used to analyze the pattern of the current dialogue;

[0017] Weight adjustment strategy unit: Dynamically adjusts the weights of each factor in the multi-factor scoring module based on the results of conversation pattern analysis. Dynamic weight adjustment can serve as an extended function of the system to improve the flexibility and adaptability of memory management strategies in more complex application scenarios.

[0018] Furthermore, the memory retrieval priority sorting module includes:

[0019] Retrieval request receiving unit: receives memory retrieval requests from the dialogue system.

[0020] Prioritization unit: Based on the comprehensive importance score of the memory items, the search results are prioritized and the memory items with higher scores are returned first.

[0021] Furthermore, the multi-factor scoring module also includes a feedback factor calculation unit; the feedback factor calculation unit is used to calculate the user feedback factor based on the user's feedback on the memory item, so as to increase the weight of the user's recognition of the memory item.

[0022] Furthermore, the multi-factor scoring algorithm is specifically as follows:

[0023] The calculation formula for the comprehensive importance score S is:

[0024] S=wt.T(t)+ws.R(q,m)+we.E(m);

[0025] in:

[0026] S: comprehensive importance score;

[0027] T(t): time decay factor, related to the timestamp t of the memory item;

[0028] R(q,m): semantic relevance factor, related to the semantic relevance between query q and memory item m;

[0029] E(m): emotional intensity factor, which is related to the emotional intensity contained in the memory item m;

[0030] wt,ws,we,wf: weights of each factor, which can be statically configured or dynamically adjusted;

[0031] Furthermore, the weight of each factor is automatically adjusted according to the conversation mode, specifically:

[0032] Frequent topic changes: Increase the weight of the semantic relevance factor ws;

[0033] Severe emotional fluctuations: increase the weight of the emotional intensity factor we;

[0034] Positive user feedback: Increase the weight of the user feedback factor wf;

[0035] Longer conversation time: Increase the weight of the time decay factor wt.

[0036] Furthermore, the storage layers include: a short-term memory layer: using cache to store recent memory items with higher scores to support fast retrieval; a long-term memory layer: using a database to store memory items with lower scores or longer time to reduce storage costs.

[0037] Compared with the prior art, the present invention has the following beneficial effects:

[0038] 1. Refined memory management: Through a multi-dimensional weighted scoring mechanism, the present invention enables the system to finely evaluate and manage conversation memories, distinguish between important and secondary information, and improve the utilization of memory resources.

[0039] 2. More accurate context understanding: This invention comprehensively considers multiple factors such as time, semantics, emotion and user feedback. The system can more accurately capture the key information of the conversation context and improve the coherence and fluency of the conversation.

[0040] 3. Optimize retrieval efficiency: The present invention performs hierarchical storage and retrieval sorting based on memory priority, which can significantly improve memory retrieval efficiency and reduce system response delay.

[0041] 4. Adaptive dialogue scenarios: The dynamic weight adjustment mechanism of the present invention enables the system to adapt to different dialogue scenarios and user interaction modes, maintaining memory management performance in various scenarios.

[0042] 5. Improve user experience: The present invention's smarter memory management and more accurate context understanding will ultimately enhance the user's interaction experience and satisfaction with the dialogue system. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments.

[0044] Figure 1 This is a system architecture diagram of a multi-dimensional weighted conversation memory priority system according to an embodiment of the present invention.

[0045] Figure 2 This is a conceptual diagram of a system core module based on multi-dimensional weighting according to an embodiment of the present invention. DETAILED DESCRIPTION

[0046] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only some of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts shall fall within the scope of protection of the present invention.

[0047] In the description of the present invention, it should be noted that the terms "center", "up", "down", "left", "right", "vertical", "horizontal", "inside", "outside", etc., indicating directions or positional relationships, are based on the directions or positional relationships shown in the accompanying drawings. They are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific direction, be constructed and operated in a specific direction. Therefore, they cannot be understood as limiting the present invention.

[0048] In the description of the present invention, it should be noted that, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be internal communication between two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances. In addition, the technical features involved in the different embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0049] like Figure 1 、 Figure 2 As shown, an embodiment of the present invention provides a multi-dimensional weighted dialogue memory priority system, including the following modules:

[0050] A multi-factor scoring module is used to calculate the importance score of conversation memory, wherein the multi-factors include a time decay factor, a semantic relevance factor, a sentiment intensity factor, and a user feedback factor;

[0051] Tiered storage decision module, which allocates memory to the corresponding storage tier based on importance score;

[0052] Dynamic weight adjustment module, automatically adjusts the weight of each factor according to the conversation mode;

[0053] The memory retrieval priority ranking module determines the ranking of retrieval results based on multi-dimensional scoring.

[0054] The multi-factor scoring module includes a time decay factor calculation unit, a semantic relevance factor calculation unit, a sentiment intensity factor calculation unit, a user feedback factor calculation unit, and a user comprehensive scoring unit. The time decay factor calculation unit is used to calculate the time decay factor of a memory item, reducing the weight of memory items that are far away in time. The semantic relevance factor calculation unit is used to calculate the semantic relevance factor between the memory item and the current conversation context, increasing the weight of related memory items. The sentiment intensity factor calculation unit is used to calculate the sentiment intensity factor of the memory item, increasing the weight of memory items with strong sentiment. Sentiment intensity analysis can be performed online or offline. Offline sentiment analysis is suitable for scenarios where historical conversation data is preprocessed and sentiment tags are extracted. The user feedback factor calculation unit is used to calculate the user feedback factor based on user feedback on the memory item (such as likes, comments, etc.), increasing the weight of user-approved memory items. In some application scenarios, the user feedback factor can be used as an optional enhancement, and its activation can be selected based on specific needs. The user comprehensive scoring unit performs a weighted combination of the time decay factor, semantic relevance factor, sentiment intensity factor, and user feedback factor, and calculates a comprehensive importance score of the memory item according to the multi-factor scoring algorithm, which is the importance score of the conversation memory.

[0055] The tiered storage decision module includes:

[0056] Scoring threshold setting unit: This unit is used to set scoring thresholds for different storage tiers, such as short-term memory thresholds and long-term memory thresholds. Scoring thresholds can be pre-configured or dynamically adjusted based on system performance. For example, a global importance_threshold parameter can be set to distinguish between short-term memory and long-term memory.

[0057] Storage Tier Allocation Unit: This unit allocates memory items to corresponding storage tiers, such as short-term memory and long-term memory, based on their combined importance scores and scoring thresholds. Storage tiers can use in-memory data structures (such as caches) or persistent storage systems (such as databases), depending on system performance and data persistence requirements.

[0058] The dynamic weight adjustment module includes:

[0059] Dialogue pattern analysis unit: used to analyze the pattern of the current dialogue, such as the frequency of topic changes, the degree of emotional fluctuation, and the user interaction style.

[0060] Weight Adjustment Strategy Unit: Dynamically adjusts the weights of various factors in the multi-factor scoring module based on conversation pattern analysis. For example, increasing the weight of the emotion intensity factor in emotional conversations. Dynamic weight adjustment can be used as a system extension to enhance the flexibility and adaptability of memory management strategies in more complex application scenarios. In basic implementations, a fixed weight configuration can be used.

[0061] The memory retrieval prioritization module includes:

[0062] Retrieval request receiving unit: receives memory retrieval requests from the dialogue system.

[0063] Prioritization unit: Based on the comprehensive importance score of the memory items, the search results are prioritized and the memory items with higher scores are returned first.

[0064] The multi-factor scoring module also includes a feedback factor calculation unit (optional). This unit calculates a user feedback factor based on user feedback (such as likes and comments) on a memory item, thereby increasing the weight of the user's approval of the memory item. In some application scenarios, the user feedback factor can be an optional enhancement, depending on specific needs.

[0065] Multi-factor scoring algorithm, specifically:

[0066] The calculation formula for the comprehensive importance score S is:

[0067] S=wt.T(t)+ws.R(q,m)+we.E(m)+wf.F(u,m)(optional);

[0068] in:

[0069] S: comprehensive importance score;

[0070] T(t): time decay factor, related to the timestamp t of the memory item;

[0071] R(q,m): semantic relevance factor, related to the semantic relevance between query q and memory item m;

[0072] E(m): emotional intensity factor, which is related to the emotional intensity contained in the memory item m;

[0073] F(u,m): User feedback factor, related to user u's feedback on memory item m (optional);

[0074] wt,ws,we,wf: weights of each factor, which can be statically configured or dynamically adjusted;

[0075] Furthermore, the weight of each factor is automatically adjusted according to the conversation mode, specifically:

[0076] Frequent topic changes: Increase the weight of the semantic relevance factor ws;

[0077] Severe emotional fluctuations: increase the weight of the emotional intensity factor we;

[0078] Positive user feedback: Increase the weight of the user feedback factor wf;

[0079] Longer conversation time: Increase the weight of the time decay factor wt;

[0080] Storage tiers include:

[0081] Short-term memory layer: Use a cache (such as Redis) to store recent memory items with high scores, supporting fast retrieval.

[0082] Long-term memory layer: Use a database (such as MySQL or MongoDB) to store memory items with low scores or long storage time to reduce storage costs.

[0083] In order to verify the technical effects of the embodiments of the present invention, the following comparative experiments were conducted:

[0084] The embodiment system of the present invention (enhanced memory system) is based on a multi-dimensional weighted dialogue memory priority system; the comparison system is a baseline system (basic system): a dialogue system using FIFO queue memory management.

[0085] Test scenario: simulate multiple rounds of conversations between users and intelligent customer service robots.

[0086] Test Data: 15 long questions were designed to test memory ability, response quality, personalization, and response time.

[0087] Memory: Assess the accuracy of long-term memory through specific questions (e.g., Question 9: "What was my first sentence?").

[0088] Response quality and personalization: Manual evaluation of the system's answers, including their level of detail, structure, and personalized expression.

[0089] Response time: Records the system's response time (in milliseconds) to each question.

[0090] Experimental steps:

[0091] 1. System deployment: Deploy the system of the present invention and the baseline system to simulate the intelligent customer service dialogue scenario.

[0092] 2. Dialogue test: Use 15 preset questions to conduct dialogue tests with the two systems respectively, and record the dialogue process and system responses.

[0093] 3. Metrics Evaluation: Memory Assessment: Analyze the accuracy of the system's response to Question 9 to determine long-term memory capacity. The enhanced memory system was able to accurately recall the user's first sentence, while the basic system was unable to do so.

[0094] Response quality and personalization evaluation: Human evaluators compared and analyzed the responses of the two systems in all questions, and scored and conducted descriptive analysis based on the level of detail, structure, and personalized expression. Due to the large number of questions (15), the evaluation process was more comprehensive and detailed. The evaluation results show that the responses of the enhanced memory system are significantly better than those of the basic system in terms of detail, structure, and personalization. For example, in question 5 (information extraction ability test), the enhanced system carefully divides the user's interests and hobbies into three categories and provides personalized suggestions, while the basic system simply lists the list. In question 7 (user portrait generation), the user portrait generated by the enhanced system is more comprehensive and coherent. In addition, the enhanced system uses personalized titles such as "Mr. Zhang San" more frequently in the conversation, which provides a better user experience.

[0095] Response Time Evaluation: The average response time of the two systems for each problem was calculated and compared, and the difference percentage was calculated. The results showed that the enhanced memory system's response time was slightly longer than the baseline system's on most problems, with an average delay of approximately 23.3%. For complex problems such as Problems 5 and 8, which require extensive memory retrieval, the enhanced system's response time increased significantly, with delays reaching 208.15% and 76.76%, respectively. Detailed response time comparison data is shown in Table 1:

[0096] Table 1

[0097]

[0098]

[0099] Experimental results demonstrate that the multi-dimensional weighted dialogue memory priority system proposed in this paper significantly improves long-term memory capacity, response quality, and personalization compared to traditional FIFO memory management systems. While the enhanced memory system exhibits a slight increase in response time, with an average delay of approximately 23.3%, this delay is more noticeable when retrieving and integrating large amounts of historical information. However, this time cost is acceptable given its advantages in memory capacity and user experience. These experimental results demonstrate the effectiveness and superiority of this invention in enhancing the dialogue system's memory management capabilities and user interaction experience. Future improvements could focus on optimizing memory retrieval efficiency and reducing response times for complex queries.

[0100] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in the present invention should be included in the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.

Claims

1. A multi-dimensional weighted dialogue memory priority system, characterized by: Includes the following modules: A multi-factor scoring module is used to calculate the importance score of conversation memory, wherein the multi-factors include a time decay factor, a semantic relevance factor, a sentiment intensity factor, and a user feedback factor; Tiered storage decision module, which allocates memories to corresponding storage tiers based on importance scores; Dynamic weight adjustment module, automatically adjusts the weight of each factor according to the conversation mode; The memory retrieval priority ranking module determines the ranking of retrieval results based on multi-dimensional scoring.

2. A multi-dimensional weighted conversation memory priority system according to claim 1, characterized in that: The multi-factor scoring module includes a time decay factor calculation unit, a semantic relevance factor calculation unit, an emotion intensity factor calculation unit, and a user comprehensive scoring unit; the time decay factor calculation unit is used to calculate the time decay factor of the memory item, reducing the weight of the memory item that is far away in time; the semantic relevance factor calculation unit is used to calculate the semantic relevance factor between the memory item and the current conversation context, increasing the weight of the related memory item; The emotion intensity factor calculation unit is used to calculate the emotion intensity factor of the memory item and increase the weight of the memory item with strong emotion. The emotion intensity analysis can be performed online or offline. Offline emotion analysis is suitable for scenarios where historical conversation data is preprocessed and emotion tags are extracted. The user comprehensive scoring unit weightedly combines the time decay factor, semantic relevance factor, and emotion intensity factor, and calculates the comprehensive importance score of the memory item according to the multi-factor scoring algorithm, which is the importance score of the conversation memory.

3. The multi-dimensional weighted conversation memory priority system according to claim 1, characterized in that: The tiered storage decision module includes: Scoring threshold setting unit: used to set scoring thresholds for different storage tiers; scoring thresholds can be set in advance or dynamically adjusted according to system operation conditions; Storage tier allocation unit: Allocates memory items to the corresponding storage tier based on their comprehensive importance score and score threshold. The storage tier can use in-memory data structures or persistent storage systems, depending on system performance and data persistence requirements.

4. The multi-dimensional weighted conversation memory priority system according to claim 1, characterized in that: The dynamic weight adjustment module includes: Dialogue pattern analysis unit: used to analyze the pattern of the current dialogue; Weight adjustment strategy unit: Dynamically adjusts the weights of each factor in the multi-factor scoring module based on the results of conversation pattern analysis. Dynamic weight adjustment can serve as an extended function of the system to improve the flexibility and adaptability of memory management strategies in more complex application scenarios.

5. The multi-dimensional weighted conversation memory priority system according to claim 1, characterized in that: The memory retrieval priority sorting module includes: Retrieval request receiving unit: receives memory retrieval requests from the dialogue system; Prioritization unit: Based on the comprehensive importance score of the memory items, the search results are prioritized and the memory items with higher scores are returned first.

6. The multi-dimensional weighted conversation memory priority system according to claim 1, characterized in that: The multi-factor scoring module also includes a feedback factor calculation unit; the feedback factor calculation unit is used to calculate the user feedback factor based on the user's feedback on the memory item, thereby increasing the weight of the user's recognition of the memory item.

7. The multi-dimensional weighted conversation memory priority system according to claim 1, characterized in that: The multi-factor scoring algorithm is specifically: The calculation formula for the comprehensive importance score S is: S=wt.T(t)+ws.R(q,m)+we.E(m); in: S: comprehensive importance score; T(t): time decay factor, related to the timestamp t of the memory item; R(q,m): semantic relevance factor, related to the semantic relevance between query q and memory item m; E(m): emotional intensity factor, which is related to the emotional intensity contained in the memory item m; wt,ws,we,wf: weights of each factor, which can be statically configured or dynamically adjusted.

8. The multi-dimensional weighted conversation memory priority system according to claim 1, characterized in that: Automatically adjust the weight of each factor based on the conversation mode, specifically: Frequent topic changes: Increase the weight of the semantic relevance factor ws; Severe emotional fluctuations: increase the weight of the emotional intensity factor we; Positive user feedback: Increase the weight of the user feedback factor wf; Longer conversation time: Increase the weight of the time decay factor wt.

9. The multi-dimensional weighted conversation memory priority system according to claim 1, characterized in that: The storage layers include: a short-term memory layer: using cache to store recent memory items with higher scores to support fast retrieval; a long-term memory layer: using a database to store memory items with lower scores or older times to reduce storage costs.

Images