Meeting room intelligent reservation method based on DeepSeek large language model

Abstract

The invention discloses a conference room intelligent reservation method based on a DeepSeek large language model, and the method comprises the steps: S10, natural language understanding based on domain knowledge injection and output constraints: S101, dynamically assembling prompt words; s102, performing intention and entity joint extraction on cue words and user query by utilizing a DeepSeek model; s103, performing semantic disambiguation and information completion; s104, carrying out structured output and verification; s20, performing intelligent reservation based on multi-objective optimization and hybrid decision, wherein the intelligent reservation comprises multi-dimensional conference room matching, conflict detection and active solution and priority scheduling; and S30, performing context management and personalized learning based on deep representation learning. According to the method, the complex and fuzzy reservation intention of the user is accurately understood, the resource utilization rate of the conference room is improved, and the conversion from passive alarm to active multi-mode solution providing is realized.

Description

Technical Field

[0001] This invention belongs to the field of intelligent information processing technology, and in particular relates to an intelligent meeting room reservation method based on the DeepSeek large language model. Background Technology

[0002] Currently, conference room management systems have evolved from manual management to electronic and preliminary intelligent stages, but fundamental limitations still exist at the core levels of intelligent interaction and decision-making: Traditional rule-based systems rely on an "if-else" logic engine, resulting in rigid rules that fail to understand and handle users' ambiguous intentions and implicit needs. Their conflict resolution capabilities are limited to passive alerts, unable to proactively generate feasible alternatives, and lead to high system maintenance costs.

[0003] Bottlenecks in the application of existing natural language processing technologies: Although general large language models (LLM) have made progress in natural language understanding, when applied to vertical fields such as conference room management, there are problems such as domain knowledge fragmentation, insufficient accuracy in intent and entity recognition, uncontrollable output, and difficulty in managing the state of multi-turn dialogues, which makes it impossible for them to reliably transition from "understanding" to "action".

[0004] In summary, existing technologies have four major flaws: 1. Insufficient depth of natural language understanding: It relies heavily on keyword matching and fixed templates, lacks true contextual reasoning and common sense judgment, and has poor anti-ambiguity and error correction capabilities.

[0005] 2. Weak ability to handle multi-dimensional constraints: It can only handle rigid conditions that "must be met", and cannot effectively coordinate the handling of flexible preferences that "expect to be met", nor can it understand the priority of different constraints.

[0006] 3. Passive and simplistic conflict resolution mechanisms: Conflict detection is mostly a simple "existence check", and the solutions lack intelligence, cannot proactively provide multimodal and feasible alternatives, and lack a global optimization perspective.

[0007] 4. Lack of user experience and personalization: The interaction process is counterintuitive, the system status is opaque, users need to remember specific system rules, and personalized services based on historical behavior cannot be provided. Summary of the Invention

[0008] To address the aforementioned issues, this invention proposes an intelligent meeting room reservation method based on the DeepSeek large language model. This method solves the problem of low interactive intelligence levels in traditional systems due to insufficient natural language understanding, enabling accurate understanding of users' complex and ambiguous reservation intentions. It also addresses the problem of weak resource matching capabilities in existing systems under multi-dimensional constraints, leading to low meeting room resource utilization. Furthermore, it overcomes the limitations of existing conflict detection mechanisms, which are passive, singular, and lack the ability to automatically generate intelligent solutions, thus transforming the approach from passive alarms to proactively providing multimodal solutions.

[0009] To achieve the above objectives, the technical solution adopted by this invention is: a smart meeting room reservation method based on the DeepSeek large language model, comprising the following steps: S10, Natural Language Understanding Based on Domain Knowledge Injection and Output Constraints, includes the following steps: S101, Dynamically assembled prompt words; S102, For prompt words and user queries, use the DeepSeek model to jointly extract intent and entities; S103, perform semantic disambiguation and information completion; S104, Structured Output and Verification; S20, based on multi-objective optimization and hybrid decision-making, features intelligent reservation, including: multi-dimensional meeting room matching, conflict detection and proactive resolution, and priority scheduling. S30, context management and personalized learning based on deep representation learning.

[0010] Furthermore, based on the current dialogue state, relevant information is extracted in real time from the domain knowledge base and user profiles and dynamically assembled into a structured system prompt word; the system prompt word includes task instructions, domain rules, user context, and mandatory JSON output format requirements.

[0011] Furthermore, the assembled system prompts and user queries are input into the DeepSeek model; the DeepSeek model, based on its deep semantic understanding of natural language, performs end-to-end intent classification and entity extraction.

[0012] Furthermore, semantic disambiguation: using the context provided by the dialogue state tracking module to resolve referential ambiguity; information completion: for requests with incomplete information, the system can intelligently complete the information based on default rules or user history data.

[0013] Furthermore, it outputs structured instructions in a predefined JSON format.

[0014] Furthermore, it determines whether the information in the instruction is complete and logically sound. If so, it passes the quality control to the next step; otherwise, it generates a list of clarification questions and uses the NLG natural language generation module to generate follow-up questions and responses to provide feedback to the user.

[0015] Furthermore, the multi-dimensional meeting room matching includes: Level 1: Rigid Filtering: The structured reservation instructions input are initially screened based on rigid conditions to exclude meeting rooms that do not meet the conditions; if a perfectly matching resource exists, conflict detection and proactive resolution are performed; if no matching resource exists, flexible scoring is performed. Level 2: Flexible scoring: For meeting rooms that pass the initial screening, a multi-dimensional weighted score is applied based on the user's flexible preferences; the weight of flexible preferences is dynamically adjusted based on the user profile, thereby upgrading from Boolean matching to quantifiable scoring and ranking.

[0016] Furthermore, the conflict detection and proactive resolution include: Conflict detection: Employs a time overlap algorithm to perform minute-accurate checks in the time dimension; Solution strategy: When a conflict occurs, a hybrid decision-making architecture is activated; a rule engine is used to detect whether the solution meets the hard constraints, while a multi-objective optimization algorithm is used to operate in the soft constraint space to generate a set of alternative solutions based on different compromise dimensions.

[0017] Furthermore, a priority scheduling mechanism is implemented: in resource competition scenarios, a priority calculation model is constructed, taking into account factors such as the organizational level of the user or department, the importance of the meeting topic, the order of appointment submission, and historical resource usage, in order to achieve a balance between fairness and efficiency in resource allocation.

[0018] Furthermore, context management and personalized learning based on deep representation learning include: Dialogue status tracking: Through dynamic dialogue status boxes, it continuously records confirmed entities, issues to be clarified, and the current dialogue stage, ensuring the continuity of multi-round dialogues and the accuracy of referential resolution. User preference learning: Employing deep representation learning, we automatically learn and summarize users' explicit and implicit preferences from their historical reservation records, and dynamically update user profiles. Historical record analysis: Aggregates and analyzes global reservation data to identify usage hotspots, high-frequency devices, and conflict patterns, and feeds the analysis results back to the resource matching and conflict resolution module to optimize system default parameters and decision-making strategies.

[0019] The beneficial effects of adopting this technical solution are: This invention employs an innovative system architecture to deeply integrate the general cognitive capabilities of the DeepSeek large language model with the professional knowledge and decision-making logic of the meeting room reservation domain. Through a structured constraint mechanism involving domain knowledge injection and output, the system transforms the large language model into a reliable domain-specific intelligent interaction hub. This hub collaborates with a backend multi-objective optimization hybrid decision engine to complete the entire process from natural language instructions to optimal reservation decisions, thereby achieving deep semantic understanding, intelligent resource matching, and proactive conflict resolution.

[0020] This invention presents a semantic understanding method based on domain knowledge injection and output structure constraints. Through dynamic prompt word engineering and strict JSON output constraints, it constructs a "guided" semantic understanding framework, effectively solving the two major challenges of "knowledge limitation" and "output uncertainty" in the application of general-purpose large language models in vertical domains. It significantly improves the accuracy of intent recognition and entity extraction in meeting room reservation scenarios, ensuring the reliability and stability of the conversion from natural language to structured business instructions.

[0021] This invention proposes a hybrid decision-making architecture that combines a rule engine and optimization algorithm to resolve conflicts based on multi-objective optimization and hybrid decision-making. It also establishes a dialogue-driven dynamic adjustment closed loop, achieving a leap from simple conflict detection to proactively generating multimodal optimization solutions. This significantly reduces the proportion of complex conflicts requiring manual intervention, improves user satisfaction and participation in decision-making in conflict scenarios, and enhances the overall utilization efficiency of enterprise spatial resources from a global optimization perspective.

[0022] This invention presents a personalized recommendation algorithm based on deep representation learning. It dynamically models user preferences using deep representation learning and designs a context-aware recommendation ranking mechanism. Simultaneously, it introduces a learning strategy to balance historical preference matching with the exploration of new resources. This achieves highly personalized meeting room resource recommendations, significantly improving user decision-making efficiency, reducing resource waste, and enabling the system to continuously learn and evolve. Attached Figure Description

[0023] Figure 1 This is a schematic diagram of a meeting room intelligent reservation method based on the DeepSeek large language model according to the present invention. Figure 2 This is a flowchart of natural language processing in an embodiment of the present invention; Figure 3 This is a flowchart of the intelligent appointment decision-making process in an embodiment of the present invention; Figure 4 This is a flowchart of the context function management in an embodiment of the present invention. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described below with reference to the accompanying drawings.

[0025] In this embodiment, see Figure 1 As shown, this invention proposes a smart meeting room reservation method based on the DeepSeek large language model, including the following steps: S10, Natural Language Understanding Based on Domain Knowledge Injection and Output Constraints, includes the following steps: S101, Dynamically assembled prompt words; S102, For prompt words and user queries, use the DeepSeek model to jointly extract intent and entities; S103, perform semantic disambiguation and information completion; S104, Structured Output and Verification; S20, based on multi-objective optimization and hybrid decision-making, features intelligent reservation, including: multi-dimensional meeting room matching, conflict detection and proactive resolution, and priority scheduling. S30, context management and personalized learning based on deep representation learning.

[0026] As an optimization of the above embodiments, such as Figure 2 As shown, based on the current dialogue state, relevant information is extracted in real time from the domain knowledge base and user profile, and dynamically assembled into a structured system prompt word; the system prompt word includes task instructions, domain rules, user context, and mandatory JSON output format requirements.

[0027] The assembled system prompts and user queries are input into the DeepSeek model. Based on its deep semantic understanding of natural language, the DeepSeek model performs end-to-end intent classification and entity extraction. This process is not a simple keyword matching, but rather an understanding of the "modify appointment" intent implied in "I want to change the location of the meeting".

[0028] Semantic disambiguation: Resolves referential ambiguity by utilizing the context provided by the dialogue state tracking module.

[0029] Information completion: For requests with incomplete information, the system can intelligently complete the information based on default rules or user history data.

[0030] Output structured instructions in a predefined JSON format.

[0031] The system determines whether the information in the instruction is complete and logically sound. If so, it passes the quality control to the next step. Otherwise, it generates a list of clarification questions and uses the NLG (Natural Language Generation) module to generate follow-up questions and responses to provide feedback to the user.

[0032] As an optimization of the above embodiments, such as Figure 3 As shown, the multi-dimensional meeting room matching includes: Level 1: Rigid Filtering: The structured reservation instructions are initially screened based on rigid conditions (such as time availability, minimum capacity, and required equipment) to exclude meeting rooms that do not meet the conditions; if a perfectly matching resource exists, conflict detection and proactive resolution are performed; if not, flexible scoring is performed. Level 2: Flexible scoring: For meeting rooms that pass the initial screening, a multi-dimensional weighted score is applied based on the user's flexible preferences (such as "quiet" or "near the elevator"). The weight of flexible preferences is dynamically adjusted based on the user profile, thereby upgrading from Boolean matching to quantifiable scoring and ranking.

[0033] The conflict detection and proactive resolution include: Conflict detection: Employs a time overlap algorithm to perform minute-accurate checks in the time dimension; Solution strategy: When a conflict occurs, a hybrid decision-making architecture is activated; a rule engine is used to detect whether the solution meets the hard constraints, while a multi-objective optimization algorithm is used to operate in the soft constraint space to generate a set of alternative solutions based on different compromise dimensions.

[0034] For example: Time offset solution: Suggest users fine-tune meeting times. Resource alternative solution: Recommend other meeting rooms with similar functions and available at the original time. Solution splitting solution: For large meetings, it is recommended to split them into multiple smaller meetings that will be held simultaneously.

[0035] The system transforms passive alerts into proactive "choices," shifting the cognitive burden of conflict resolution from the user to the system.

[0036] The priority scheduling mechanism: In a resource competition scenario, a priority calculation model is constructed, taking into account factors such as the organizational level of the user or department, the importance of the meeting topic, the order of appointment submission, and historical resource usage, in order to achieve a balance between fairness and efficiency in resource allocation.

[0037] As an optimization of the above embodiments, such as Figure 4 As shown, context management and personalized learning based on deep representation learning enable the system to continuously evolve and provide personalized services, including: Dialogue status tracking: Through dynamic dialogue status boxes, it continuously records confirmed entities, issues to be clarified, and the current dialogue stage, ensuring the continuity of multi-round dialogues and the accuracy of referential resolution. User preference learning: Employing deep representation learning, we automatically learn and summarize users' explicit and implicit preferences from their historical reservation records, and dynamically update user profiles. Historical record analysis: Aggregates and analyzes global reservation data to identify usage hotspots, high-frequency devices, and conflict patterns, and feeds the analysis results back to the resource matching and conflict resolution module to optimize system default parameters and decision-making strategies.

[0038] This invention has the following advantages: Enhanced comprehension: Through domain knowledge enhancement and output constraints, a deep understanding and accurate analysis of complex and ambiguous user needs are achieved, resulting in a significant improvement in intent recognition accuracy.

[0039] Intelligent decision-making: Through multi-objective optimization and hybrid decision-making, intelligent and refined matching of resources and proactive, multimodal conflict resolution are achieved, which is expected to reduce the need for manual intervention by more than 70%.

[0040] Resource efficiency optimization: By adopting a global optimization perspective and personalized recommendations, the overall utilization rate of meeting room resources has been improved, reducing resource idleness and waste.

[0041] Improved user experience: Through natural and smooth multi-turn dialogue interaction, the user decision-making and operation path is significantly shortened, reducing the threshold for use and the learning cost.

[0042] System self-evolution: It has the ability to evolve itself through continuous learning, adapt to changes in user behavior, and provide long-term, stable, high-quality, personalized services.

[0043] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.

Claims

1. A method for intelligent reservation of a conference room based on a DeepSeek large language model, characterized in that, Comprising steps of: S10, natural language understanding based on domain knowledge injection and output constraints, comprising steps of: S101, dynamically assembling prompt words; S102, jointly extracting intent and entity using DeepSeek model on prompt words and user query; S103, semantic disambiguation and information completion; S104, structured output and verification; S20, intelligent reservation based on multi-objective optimization and hybrid decision, including multi-dimensional conference room matching, conflict detection and active resolution, and priority scheduling; S30, context management and personalized learning based on deep representation learning.

2. The conference room intelligent reservation method based on the DeepSeek large language model according to claim 1, characterized in that, According to the current dialogue state, relevant information is extracted from the domain knowledge base and user portrait in real time, and a structured system prompt word is dynamically assembled; the system prompt word includes task instructions, domain rules, user context and mandatory JSON output format requirements.

3. The conference room intelligent reservation method based on the DeepSeek large language model according to claim 2, characterized in that, The assembled system prompt word is input into the DeepSeek model together with the user query; the DeepSeek model has deep semantic understanding ability of natural language, and performs end-to-end intent classification and entity extraction.

4. The conference room intelligent reservation method based on the DeepSeek large language model according to claim 3, characterized in that, Semantic disambiguation: use the context provided by the dialogue state tracking module to solve the reference ambiguity; information completion: for incomplete information requests, the system can intelligently complete based on default rules or user historical data.

5. The conference room intelligent reservation method based on the DeepSeek large language model according to claim 4, characterized in that, Output structured instructions according to the predefined JSON format.

6. The conference room intelligent reservation method based on the DeepSeek large language model according to claim 5, characterized in that, Determine whether the information in the instructions is complete and logically reasonable, if so, pass the quality to the next step, if not, generate a list of clarification questions, and generate a follow-up response feedback to the user through the NLG natural language generation module.

7. The conference room intelligent reservation method based on the DeepSeek large language model according to claim 1, characterized in that, The multi-dimensional conference room matching includes: First level: rigid filtering: according to rigid conditions, the structured reservation instructions input are preliminarily screened to exclude conference rooms that do not meet the conditions; if there is a completely matched resource, conflict detection and active resolution are performed, if not, flexible scoring is performed; Second level: flexible scoring: for conference rooms that pass the preliminary screening, multi-dimensional weighted scoring is performed according to the user's flexible preferences; the flexible preference weight is dynamically adjusted based on the user portrait, so as to realize the upgrade from Boolean matching to quantifiable scoring and sorting.

8. The conference room intelligent reservation method based on the DeepSeek large language model according to claim 7, characterized in that, The conflict detection and active resolution includes: Conflict detection: time overlap algorithm is used to check in time dimension accurately to minute; Solution strategy: when conflict occurs, the hybrid decision architecture is started; the rule engine is used to detect that the scheme meets the hard constraints, and the multi-objective optimization algorithm is used to operate in the soft constraint space to generate a set of alternative schemes based on different compromise dimensions.

9. The conference room intelligent reservation method based on the DeepSeek large language model according to claim 1, characterized in that, Priority scheduling mechanism: in the resource competition situation, a priority calculation model is constructed, including user or department organizational level, meeting theme importance, reservation submission time sequence and historical resource use, to realize the balance between fairness and efficiency of resource allocation.

10. The conference room intelligent reservation method based on the DeepSeek large language model according to claim 1, characterized in that, Context management and personalized learning based on deep representation learning includes: Dialogue state tracking: through a dynamic dialogue state box, confirmed entities, topics to be clarified and current dialogue stages are recorded continuously to ensure the coherence of multi-round dialogue and the accuracy of reference resolution; User preference learning: automatically learn and induce the explicit and implicit preferences of users from their historical booking records, and dynamically update the user profile based on deep representation learning; History analysis: aggregate and analyze the global booking data to identify the hotspots, high-frequency devices, and conflict patterns, and feed the analysis results back to the resource matching and conflict resolution module to optimize the system default parameters and decision-making strategies.

Images