Conference information intelligent management method and system

Abstract

The invention discloses an intelligent conference information management method and system, and relates to the technical field of information management, and the method comprises the steps: carrying out the more precise identity authentication of participants through multi-modal information, generating an authority matrix through combining the historical behavior data of the participants, guaranteeing that important participants obtain the preferential document access authority, and improving the user experience. Information leakage is avoided, document distribution is performed according to importance of conference documents, benefit conflicts among participants are detected in real time, permissions of the participants are effectively adjusted, efficiency and safety of document distribution are improved, interaction behaviors among the participants can be deeply analyzed through a graph neural network technology, an interaction graph is generated, and the method and the system have the advantages of being high in practicability and the like. The speaking authority of the participants is dynamically adjusted, reasonable distribution of speaking resources is ensured, the interaction efficiency of the conference is improved, the conference abstract can be dynamically generated according to the change of the speaking content in combination with the intelligent voice recognition technology and pushed to the participants, and the conference efficiency, the discussion quality and the intelligent level of conference management are remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of information management, and particularly relates to a conference information intelligent management method and system. BACKGROUND

[0002] In recent years, with the increasing demand for conference and conference management, the application of information technology in conference management has been continuously expanding, and conference management has gradually evolved from traditional manual management to intelligent and digital direction. Traditional conference management usually relies on manual methods for participant identity verification, conference material distribution, permission management, and interactive discussion, which not only takes time and effort, but also is prone to management omissions and information leakage. In recent years, with the maturity of artificial intelligence, big data, natural language processing and other technologies, intelligent conference management systems have been widely used. Through these systems, conference process automation can be achieved, such as sending meeting notices, verifying participant identities, automatically generating meeting records, and the like, thereby greatly improving the efficiency of conference management. However, the existing intelligent conference management system still has certain technical limitations in participant identity authentication, permission control, interactive analysis and conference record generation. SUMMARY

[0003] In view of the problems existing in the prior art conference information intelligent management method and system, the present application is proposed.

[0004] Therefore, the problem to be solved by the present application is that the existing intelligent conference management system still has certain technical limitations in participant identity authentication, permission control, interactive analysis and conference record generation.

[0005] To solve the above technical problems, the present application provides the following technical scheme: a conference information intelligent management method, comprising: authenticating the identity of a participant through multi-modal information, generating a permission matrix based on the historical behavior data of the participant after identity authentication;

[0006] Extracting core keywords and topics of all conference documents and classifying them according to document importance, calculating the association degree between documents and participants based on the permission matrix and document classification results, assigning corresponding conference documents to each participant, and detecting interest conflicts between participants in real time during the document allocation process and adjusting the participant permissions in real time according to the conflict detection results;

[0007] Real-time monitoring of the interactive situation of each participant and analysis of the interactive behavior of each participant through a graph neural network to generate an interactive map, identify the interactive relationship between participants and high-value discussion nodes, and dynamically adjust the speaking permissions of participants according to real-time interactive data;

[0008] The keywords of the speech content of the participants are extracted through intelligent voice recognition technology, and the extracted keywords are dynamically adjusted according to the change of the speech content, a real-time conference abstract is generated, and the generated conference abstract is pushed to the corresponding participants in real time according to the permission of the participants.

[0009] As a preferred scheme of the conference information intelligent management method, the identity of the participant is authenticated through multi-modal information, and a permission matrix is generated according to historical behavior data of the participant after the identity authentication.

[0010] The deep features of the preprocessed face image of the participant are extracted using a convolutional neural network, the similarity between the current face features and the pre-stored face features of the participant in the database is calculated using cosine similarity, a similarity threshold W is set, if the face similarity is greater than or equal to the similarity threshold W, the face recognition is passed, otherwise the recognition fails, and the participant is prompted to re-collect the image;

[0011] After the face recognition verification is passed, a one-time verification code is sent according to the pre-set security information of the participant for participant identity authentication, the identity of the participant is confirmed after the participant passes the face recognition and the verification code verification, the historical participation data and the speech record of the participant are retrieved according to the identity information of the participant, and the historical behavior data of the participant is extracted;

[0012] The conference types of discussion conferences, decision-making conferences and training conferences are preset, the organizer is required to select the corresponding conference type from the preset types according to the conference target when the conference is created, and different behavior weights are set for the behavior data of the participants according to different conference types;

[0013] The behavior score of the participant is calculated according to the historical behavior data through a weighted scoring formula;

[0014] The preset permission levels are L1: the highest permission, L2: the medium permission, and L3: the lowest permission, each permission level corresponds to the document range and the speech permission level that the participant can access, the participant is assigned to the corresponding permission level according to the behavior score of the participant, the permission matrix of each participant is generated, and the document range that the participant can access is preliminarily defined.

[0015] As a preferred scheme of the conference information intelligent management method, the core keywords and the theme of all the documents of the conference are extracted and classified according to the importance of the documents, the association degree between the documents and the participants is calculated based on the permission matrix and the document classification result, and the corresponding conference documents of each participant are assigned.

[0016] The TF-IDF value of each word is obtained by calculating the frequency of each word in the document and using the inverse document frequency to measure the importance of each word in all documents, each document is sorted in descending order according to the TF-IDF value to generate a keyword list, and a document keyword set is generated according to the keyword list; a participant keyword set is generated according to the historical behavior and permission matrix of the participant;

[0017] The association degree of a single document to all participants is calculated using the Jaccard similarity coefficient, and the calculated association degrees are fused by weighted summation to generate a comprehensive association degree value, the association degree threshold is set as T1 and T2, and the association degree threshold is compared with the association degree value to classify the document:

[0018] If the association degree value is greater than or equal to T1, the document is a core document, and the document is assigned to the L1 level participant;

[0019] If T2 is less than or equal to the association degree value and less than T1, the document is a general document, and the document is assigned to the L2 level participant;

[0020] If the association degree value is less than T2, the document is a secondary document, and the document is assigned to the L3 level participant;

[0021] All documents in the conference are classified by using the Jaccard similarity coefficient calculation and threshold assignment method.

[0022] As a preferred scheme of the conference information intelligent management method, wherein: the interest conflict between participants is detected in real time during the document allocation process, and the participant permission is adjusted in real time according to the conflict detection result; the participant attribute is defined and the attribute set is generated, the corresponding attribute of each document is defined and the document attribute set is generated, and the participant attribute is matched;

[0023] Potential interest conflict factors are extracted by analyzing the participant attribute and the document attribute;

[0024] The conflict factors include discrete attribute conflict factors and continuous attribute conflict factors;

[0025] The conflict of the discrete attribute is processed using logical judgment, a binary result is assigned to each conflict factor, and a conflict score is accumulated based on the judgment result of each factor;

[0026] The Euclidean distance is used to calculate the continuous attribute conflict value after normalizing each continuous attribute;

[0027] The discrete attribute and the continuous attribute conflict value are weighted and added to obtain the final conflict value C of the participant and the document ij ;

[0028] Calculate the conflict value for each participant and document, set a separate conflict value threshold U for each document, compare the conflict value with the conflict value threshold, and determine the conflict situation:

[0029] If C ij >U, it indicates that the conflict risk of participant i is high, and the access permission of participant i to document j is immediately adjusted;

[0030] If C ij ≤U, it indicates that the conflict risk of participant i is low, and participant i is allowed to continue accessing the document;

[0031] Adjust the permission matrix in real time according to the changed participant permission.

[0032] As a preferred scheme of the intelligent conference information management method, wherein: the interaction of each participant is monitored in real time, and the interaction behavior of each participant is analyzed through a graph neural network to generate an interaction graph, identify the interaction relationship between participants and high-value discussion nodes, and dynamically adjust the speaking permission of participants according to real-time interaction data; after setting the permissions of the participants, the interaction data of each participant is collected in real time through sensors, microphones and cameras, the interaction edge weight w il is calculated between each pair of participants, and after the calculation is completed, the edge weights between all participants are summarized to form an interaction intensity matrix, and the interaction intensity between each pair of participants is determined by the corresponding weight;

[0033] An initial feature vector is generated for each participant According to the initial feature vector and the interaction edge weight, a graph G=(V,J) is constructed, wherein V represents a participant node, each node represents a participant, and J represents an interaction edge between participants;

[0034] The feature vector of the participant is iteratively updated through a graph neural network, and in each iteration, the feature vector of the participant node is updated through the information of its neighbor nodes;

[0035] After the iteration of the graph neural network is completed, a complete interaction graph is generated, the interaction graph is visualized, and the node color is identified according to the relevance score of the speaking content and the conference theme;

[0036] Based on the interaction graph, the centrality score of each participant is calculated using the feature vector centrality, and the initial speaking score of the participant is set according to the centrality score of the participant in the interaction graph;

[0037] The real-time speaking score increment is obtained by weighted calculation of the speaking frequency, duration and relevance of the participant, and the real-time speaking score is calculated by combining the initial speaking score and the real-time speaking score increment;

[0038] Rank all participants according to the real-time speech score of each participant and dynamically update the ranking list, if the participant ranking is in the top 20%, it belongs to high-level participants, if the participant ranking is in 20%-80%, it belongs to medium-level participants, and if the participant ranking is in the last 20%, it belongs to low-level participants.

[0039] As a preferred scheme of the intelligent conference information management method, wherein: the keyword extraction of the speech content of the participant by the intelligent speech recognition technology and the dynamic adjustment of the extracted keywords according to the change of the speech content, and the generation of the real-time conference abstract point to the real-time capture of the speech audio stream of the participant, the conversion of the preprocessed voice signal into text through the DeepSpeech model, the cutting of the text to generate text paragraphs and the preprocessing through the natural language processing technology, the theme training of the preprocessed text paragraphs using the LDA model, and the expression of each text paragraph as a set of potential theme probability distribution;

[0040] Generate N themes for each text paragraph, each theme contains multiple keywords, and arrange the keywords according to the keyword frequency, extract the top n keywords with the highest weight as the core content of the text paragraph, and calculate the keyword distribution of the new segment when a new speech segment is generated, and update the keyword weight B y ;

[0041] The theme distribution of the LDA model clusters similar keywords into the same theme, and generates a conference abstract according to the clustering results and keyword weights.

[0042] As a preferred scheme of the intelligent conference information management method, wherein: the real-time push of the generated conference abstract to the corresponding participant according to the participant's authority indicates that the generated conference abstract is matched according to the authority level of each participant, and the exclusive conference abstract of each participant is generated according to the authority matrix, and the best push mode is selected according to the device type and preference of each participant to push the conference abstract to the corresponding participant.

[0043] Another object of the present application is to provide an intelligent conference information management system, which comprises,

[0044] The authority management module is used for identity authentication of the participants, generating an authority matrix, and distributing authority according to the historical behavior data of the participants.

[0045] The document distribution module is used for collecting conference documents and distributing the documents based on the participant authority and historical behavior.

[0046] The conflict detection module is used for real-time detection of the interest conflicts between participants and adjustment of the participant authority according to the detection results.

[0047] The right adjustment module is used for monitoring the interaction of the participants in real time, analyzing the interaction behavior through a graph neural network, generating an interaction graph, and dynamically adjusting the speaking right;

[0048] The abstract pushing module is used for processing the speaking content of the participants through intelligent voice recognition technology, extracting keywords, generating a conference abstract, and pushing personalized conference abstracts according to the right level of the participants.

[0049] A computer device comprises a memory and a processor, the memory stores a computer program, and the processor implements the steps of the meeting information intelligent management method when executing the computer program.

[0050] A computer readable storage medium stores a computer program, and the computer program implements the steps of the meeting information intelligent management method when executed by a processor.

[0051] The present application has the advantages that: the present application can accurately identify the identity of the participants through multi-modal information, generate a right matrix combined with the historical behavior data of the participants, ensure that important participants have priority access to documents, and avoid information leakage; the present application can allocate documents according to the importance of the conference documents and detect the interest conflicts between the participants in real time, effectively adjust the rights of the participants, improve the efficiency and security of document allocation, deeply analyze the interaction behavior between the participants through the graph neural network technology, generate an interaction graph, dynamically adjust the speaking right of the participants, ensure the reasonable allocation of speaking resources, improve the interaction efficiency of the conference, and dynamically generate conference abstracts according to the changes of the speaking content through the intelligent voice recognition technology and push the conference abstracts to the participants, thereby significantly improving the conference efficiency, discussion quality, and intelligent level of conference management. BRIEF DESCRIPTION OF DRAWINGS

[0052] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiment description. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0053] Figure 1 It is a flowchart of the meeting information intelligent management method.

[0054] Figure 2 It is a structural schematic diagram of the meeting information intelligent management system. DETAILED DESCRIPTION

[0055] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the specific embodiments of the present application will be described in detail in combination with the drawings of the specification.

[0056] In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present application. However, it will be appreciated that the present application can be practiced in a variety of ways beyond the specific details set forth herein, having regard to the content of the following description, and thus the present application should not be construed as being limited to the following description.

[0057] Secondly, the "one embodiment" or "embodiment" referred to herein means a specific feature, structure or characteristic that can be included in at least one implementation of the present application. "In one embodiment" appearing in different places in the specification does not mean the same embodiment, nor does it mean an embodiment that is independent of or mutually exclusive with other embodiments.

[0058] Embodiment 1, Reference Figure 1 As a first embodiment of the present application, the embodiment provides an intelligent conference information management method, the intelligent conference information management method comprises,

[0059] S1, identity authentication of the participant is performed through multi-modal information, and a permission matrix is generated according to historical behavior data of the participant after the identity authentication;

[0060] Specifically, the identity authentication of the participant is performed through multi-modal information, and the permission matrix is generated according to the historical behavior data of the participant after the identity authentication, which means that the facial image of the participant is collected in real time through the camera of the device when the participant logs in;

[0061] The preprocessing means removing noise, adjusting illumination and contrast, and normalizing the image to ensure that the quality of the image is suitable for model processing;

[0062] The deep features of the preprocessed facial image of the participant are extracted using a convolutional neural network, the similarity between the current facial features and the pre-stored facial features of the participant in the database is calculated using cosine similarity, a similarity threshold W is set, if the facial similarity is greater than or equal to the similarity threshold W, the facial recognition is passed, otherwise the recognition fails, and the participant is prompted to re-collect the image;

[0063] The similarity threshold W is obtained by statistical analysis of the similarity values of the facial features, obtaining the distribution characteristics of the similarity values, by observing the similarity values of the known matching and non-matching samples, selecting the position of the similarity value in the known matching samples as the threshold W.

[0064] After the face recognition verification is passed, a one-time verification code is sent according to the preset security information (mobile phone number or email address) of the participant, the consistency of the verification code input by the participant and the sent verification code is checked, if the input is correct, the verification is passed, if the input is incorrect, the participant is prompted to request a new verification code, if the participant inputs incorrectly for three times in succession, the account of the participant is automatically locked and the user is required to contact the administrator for unlocking;

[0065] After the participant passes the face recognition and verification code verification, the identity of the participant is confirmed, the historical participation data and speech records of the participant are retrieved according to the identity information of the participant, and the historical behavior data of the participant is extracted;

[0066] The historical behavior data includes the number of participation times, the speech frequency and the conference contribution of the past conference;

[0067] The conference types of the preset discussion conference, decision-making conference and training conference are required to be selected by the organizer according to the conference target from the preset types when the conference is created, and different behavior weights are set for the behavior data of the participants according to different conference types;

[0068] The goal of the discussion conference is to promote discussion and exchange, and the focus is on the speech frequency and interactive participation of the participants;

[0069] The goal of the decision-making conference is to make decisions, and the focus is on the contribution degree (such as proposal, decision-making suggestion) of the participants;

[0070] The goal of the training conference is knowledge transfer and learning, and the focus is on the participation degree and learning feedback of the participants;

[0071] The behavior score of the participant is calculated according to the historical behavior data through a weighted scoring formula;

[0072] The preset permission level corresponds to the document range and speech permission level that the participant can access, and the permission level is set as follows:

[0073] L1: the highest permission, can access all documents, has the priority to speak, and can make speeches in any stage and any topic of the conference;

[0074] L2: medium permission, can access part of important documents, make speeches after L1 participants speak, and can express opinions on part of important topics;

[0075] L3: the lowest permission, only can access public documents, make speeches after L1 and L2 participants speak, and can only speak in secondary topics or open discussion sessions;

[0076] According to the behavior score of the participants, the participants are assigned to the corresponding permission level, and a permission matrix is generated for each participant to preliminarily define the range of documents accessible by the participants.

[0077] This step authenticates the identity of the participants through multi-modal information, relying not only on a single information source (such as facial recognition), but also combining verification methods such as mobile phone numbers and email addresses. This provides a more comprehensive and secure identity authentication solution, preventing authentication failures due to the limitations of single verification methods and improving the reliability of participant identity verification. During the identity verification process, a convolutional neural network (CNN) is used to extract the deep features of the participants' facial images, ensuring the accuracy of image feature extraction. Compared to simple image comparison, CNN can capture more dimensional features, especially suitable for facial recognition under less than ideal conditions such as light changes and angle differences, greatly improving the accuracy of facial recognition. The generation of the permission matrix based on the historical behavior data of the participants ensures the dynamic and personalized nature of the permission allocation. The introduction of historical behavior data not only allows for the integration of historical performance such as the number of participations and the frequency of speeches, but also allows for the setting of different weights based on different types of meetings. For example, in decision-making meetings, the system pays more attention to the quality of the participants' proposals and decision-making suggestions; while in discussion meetings, the frequency of speeches and the level of interaction become the main basis for measuring the participants' permissions. This scoring mechanism based on historical behavior data and behavior weights makes the permission allocation more targeted, reflecting the participants' actual contributions in different types of meetings, ensuring that the allocation of permissions is more reasonable and can fully motivate participants to perform well in different types of meetings according to the meeting goals. Through the weighted scoring formula, the behavior score of the participants is calculated based on their historical behavior data, and the participants are assigned to different permission levels according to the score, improving the fairness and participation of the meeting and ensuring that the most contributing and active participants receive higher permissions. The generation of the permission matrix is based on the behavior score of the participants to determine the range of documents and speaking permissions they can access. This mechanism can effectively manage conference resources and ensure that key documents and speaking opportunities are allocated to the appropriate participants. The system can effectively control the conference process, ensuring the reasonable use and allocation of important resources.

[0078] S2, extract the core keywords and topics of all conference documents and classify them according to their importance, calculate the association degree between documents and participants based on the permission matrix and document classification results, assign corresponding conference documents to each participant, and in the process of document allocation, real-time detect the conflicts of interest between participants and adjust the permissions of participants in real-time according to the conflict detection results;

[0079] Specifically, the core keywords and topics of all documents of the meeting are extracted and classified according to the importance of the documents, the relevance between the documents and the participants is calculated based on the permission matrix and the classification results of the documents, and the corresponding meeting document collection is allocated to each participant. Collect all documents related to the meeting, unify the document format, and clean the text of the document by removing unnecessary characters, punctuation marks, pictures and tables. The text in the document is cut into individual words through a word segmentation algorithm to generate a word set for each document.

[0080] The frequency of each word in the document is calculated, and the importance of each word in all documents is measured using the inverse document frequency to obtain the TF-IDF value of each word. The keywords list is generated by sorting each document in descending order according to the TF-IDF value.

[0081] The document keyword set is generated according to the keyword list, and the participant keyword set is generated according to the historical behavior and permission matrix of the participant.

[0082] The relevance of a single document to all participants is calculated using the Jaccard similarity coefficient:

[0083] The intersection and union of the document keyword set and the participant keyword set are generated.

[0084] The intersection represents the keywords common to the document and the participant. Assuming that there are P keywords common to the document and the participant, the value of the intersection is P.

[0085] The union represents the total number of all unique keywords in the two sets, i.e. the set of all keywords of the document and the participant. If the keyword set of the document has M words, the keyword set of the participant has Q words, and they have Y common keywords, then the size of the union is M+Q-Y.

[0086] The relevance of the document and the participant is calculated by the ratio of the intersection and the union, represented as the Jaccard similarity coefficient, ranging from 0 to 1, with a higher value closer to 1 indicating a higher relevance.

[0087] The comprehensive relevance value is generated by weighted summation of all the calculated relevance values. Set the relevance threshold values T1 and T2, and compare the relevance threshold value with the relevance value to classify the documents:

[0088] If the relevance value is greater than or equal to T1, the document is a core document, and the document is assigned to L1 level participants.

[0089] If T2 is less than or equal to the relevance value and less than T1, the document is a general document, and the document is assigned to L2 level participants.

[0090] If the relevance value is less than T2, the document is a secondary document, and the document is assigned to L3 level participants.

[0091] All documents in the conference are divided by using the method of Jaccard similarity coefficient calculation and threshold assignment;

[0092] All conference-related documents include topics, background reports, and attachments.

[0093] The relevance threshold is obtained by user feedback on document classification through small-scale pilot conferences, so that T1 and T2 can be continuously adjusted according to actual experience, verified using different data sets, and optimized through multiple experiments. Through the extraction of core keywords, the content of the document can be more accurately reflected, thereby laying a solid foundation for subsequent document allocation. When generating the keyword set of the document, the TF-IDF value provides an automated weight measurement method that can accurately reflect the actual content of the document. By calculating the intersection and union of the document keyword set and the participant keyword set, the relevance of the participant and the document can be effectively evaluated. Compared with the traditional allocation mechanism based on fixed permissions, the keyword set generated by actual behavior and interest area is more dynamic and flexible, which can more accurately match the needs of participants and document content, avoid the allocation of irrelevant documents, and improve the conference experience of participants. The Jaccard similarity coefficient is used to measure the similarity of the document and the participant keyword set. Based on this relevance, the system can flexibly allocate documents to participants of different levels, accurately allocate relevant documents to the most suitable participants according to their interests and needs, improve the efficiency of conference resource allocation, and ensure that core documents are allocated to the most relevant personnel. By weighting and summing the relevance of all documents, a comprehensive relevance value can be generated, and by setting threshold T1 and T2, the documents can be classified. The system can automatically process a large number of conference documents, reducing human intervention while ensuring that the importance of the document is fully evaluated and correctly allocated. By setting relevance thresholds T1 and T2, documents are divided into core documents, general documents, and secondary documents according to the relevance value, and are allocated to participants of different levels to improve the accuracy of document allocation, so that core documents are preferentially allocated to the most relevant participants, thereby improving conference efficiency, ensuring the confidentiality and security of conference content, and preventing unnecessary information sharing.

[0094] Further, in the document allocation process, the interest conflicts between participants are detected in real time, and the participant permission definition is adjusted in real time according to the conflict detection results;

[0095] The participant attributes include the organization to which they belong, their role (manager, ordinary participant, speaker), their cooperation history (whether they have a cooperation history with the relevant projects involved in the conference), and their behavior history (the extent of their participation in past conferences, such as speaking frequency, contribution, etc.);

[0096] define corresponding attributes for each document and generate a set of document attributes, which are matched with the attributes of the participants;

[0097] The document attributes include document sensitivity, document topic, and associated organization;

[0098] Extract potential conflict factors by analyzing the attributes of the participants and the attributes of the documents;

[0099] The conflict factors include discrete attribute conflict factors and continuous attribute conflict factors;

[0100] The discrete attribute conflict factors include that the participant and the document belong to the same organization, the participant and the document involve a project, and the participant and the document have a cooperation history;

[0101] The continuous attribute conflict factor refers to the frequency of the participant's speech in past meetings;

[0102] Use logical judgment to process the conflict of discrete attributes, assign a binary result (0 or 1) to each conflict factor, and accumulate the conflict score based on the judgment result of each factor:

[0103]

[0104] where F s (i,j) is the conflict value of the discrete attribute, F k (i,j) is the value of participant i and document j on the kth discrete attribute, I() is an indicator function that returns a value of 0 or 1, i.e., returns 1 if a certain condition (such as the same organization, cooperation history, etc.) is met, otherwise returns 0, and m is the number of discrete attributes;

[0105] After normalizing each continuous attribute, use the Euclidean distance to calculate the conflict value of the continuous attribute:

[0106]

[0107] where F′ c (i,j) is the conflict value of the continuous attribute, F′ k ‘(i,j) is the normalized value of participant i and document j on the kth continuous attribute, D k ’(j) is the normalized value of document j on the kth continuous attribute, and p is the total number of continuous attributes;

[0108] Weighted addition of the discrete attribute and continuous attribute conflict values gives the final conflict value C ij between the participant and the document;

[0109] The benefit conflict value is calculated for each participant and document, a dynamic risk score is generated by setting different attributes (such as speech frequency, institutional relationship, etc.) to fuzzy rules of conflict risk through a fuzzy logic system, and a separate conflict value threshold U is set for each document according to the conflict value, the conflict value is compared with the conflict value threshold, and the conflict situation is judged:

[0110] If C ij >U, it indicates that the conflict risk of participant i is high, and the access right of participant i to document j is immediately adjusted;

[0111] If C ij ≤U, it indicates that the conflict risk of participant i is low, and participant i is allowed to continue to access the document;

[0112] The permission matrix is adjusted in real time according to the changed participant permission.

[0113] By monitoring the attributes of participants and documents in real time, the system can more accurately identify potential conflicts of interest and prevent sensitive information from being leaked due to conflicts of interest. For example, when a participant has a competing relationship with an organization related to a meeting document, the system can automatically restrict their access rights to reduce risks. This enhances the flexibility and security of the document distribution process, enabling intelligent early warning and immediate response to conflicts of interest. By defining attribute sets for participants and documents, the system can match the relationship between participants and documents based on these attributes, enabling personalized permission allocation. Compared to traditional fixed role allocation methods, this method is more flexible and can set access rights for each document individually, rather than relying solely on pre-set roles for participants. By dynamically generating attribute sets for participants and documents, the system can more accurately allocate permissions, ensuring that participants can only access documents related to their functions. This significantly reduces the risk of information leakage, especially for complex meeting scenarios with multiple levels and roles, providing higher security. In handling discrete attribute conflicts, the system assigns a binary result (0 or 1) to each conflict factor through logical judgment and accumulates conflict scores based on the judgment results of each factor. This improves the efficiency of judgment, enabling quick identification and handling of discrete attribute conflicts, avoiding complex calculations and manual processing, and ensuring that the system can adjust permissions in real time. Using the Euclidean distance to calculate the conflict of normalized continuous attributes (such as the speaking frequency of participants) can quantify the difference between participant attributes and document attributes, especially for continuous variables such as speaking frequency, enhancing the accuracy of conflict detection. Through a refined calculation model, the system can more flexibly handle continuous attribute conflicts, improving the rationality of permission adjustments. By calculating the conflict of interest value of participants and documents, setting a conflict threshold for each document, and determining whether the conflict risk of participants exceeds the threshold, the system can adjust access rights. This enables automatic and real-time adjustment of permissions, allowing the system to dynamically adjust permissions based on the interest relationship between participants and documents, ensuring the security and confidentiality of sensitive information in meetings.

[0114] S3, real-time monitoring of each participant's interaction and analysis of each participant's interaction behavior through a graph neural network to generate an interaction map, identify the interaction relationship between participants and high-value discussion nodes, and dynamically adjust the speaking rights of participants according to real-time interaction data;

[0115] Specifically, real-time monitoring of each participant's interaction and analysis of each participant's interaction behavior through a graph neural network to generate an interaction map, identify the interaction relationship between participants and high-value discussion nodes, and dynamically adjust the speaking rights of participants according to real-time interaction data. After setting the permissions of participants, real-time collection of interaction data of each participant through sensors, microphones and cameras;

[0116] The interaction data includes the number of speeches, the length of speeches, the relevance of speech content to the conference theme (calculated by a real-time semantic analysis model), and the interaction frequency and response situation with other participants (including whether the speech triggered the response of others and whether it was discussed with others);

[0117] The interaction edge weight w is calculated according to the interaction data between each participant il :

[0118]

[0119] In the formula, H il is the total number of interactions between participants i and l (unitless), which is obtained by recording the interaction behavior of the system, R il is the total number of responses of participant l to i (unitless), which is also obtained by recording the system, T total is the total time of the meeting, which is a time parameter for normalization, to standardize the interaction frequency and response intensity to "per unit time", t il is the time when the interaction or response occurs, and λ is the time decay coefficient, which controls the decay of the interaction weight with time;

[0120] After the calculation is completed, the edge weights between all participants are summarized to form an interaction intensity matrix, and the interaction intensity between each pair of participants is determined by the corresponding weight;

[0121] An initial feature vector is generated for each participant

[0122]

[0123] In the formula, S i is the number of speeches of participant i, T' i is the length of speeches of participant i, X i is the relevance of speech content of participant i to the conference theme;

[0124] The graph G=(V,J) is constructed according to the initial feature vector and the interaction edge weight, wherein V represents the participant node, each node represents a participant, and J represents the interaction edge between participants;

[0125] The feature vector of the participant is updated iteratively through the graph neural network, and in each iteration, the feature vector of the participant node is updated through the information of its neighbor nodes:

[0126]

[0127] In the formula, is the feature vector of participant node q after the w+1 iteration, is the feature vector of the wth layer participant node a, A(q) is the neighbor node set of participant node q, d q is the number of other participant nodes interacting with participant node q, W is a weight matrix used to integrate the feature information of neighbor nodes, d a is the number of other participant nodes interacting with participant node a, σ is the ReLU activation function.

[0128] After the iteration of the graph neural network is completed, a complete interaction map is generated, the interaction map is visualized, the node color is identified according to the relevance score of the speech content and the conference theme, and the deeper the node color, the higher the relevance.

[0129] Based on the interaction map, the centrality score of each participant is calculated using the feature vector centrality, and the initial speech score of the participant is set according to the centrality score of the participant in the interaction map.

[0130] The real-time speech score increment is calculated by weighting the speech frequency, duration and relevance of the participant, and the real-time speech score is calculated by combining the initial speech score and the real-time speech score increment.

[0131] According to the real-time speech score of each participant, all participants are ranked and the ranking list is dynamically updated, if the participant ranking is in the top 20%, it belongs to high-level participants, and can speak at any stage of the meeting, if the participant ranking is between 20% and 80%, it belongs to medium-level participants, and needs to queue for speaking, that is, can only apply for speaking when the meeting is paused or allowed to express opinions, after applying for speaking, the participants are arranged in order, and the participants speak in order, if the participant ranking is in the last 20%, it belongs to low-level participants, and needs to apply for speaking in the allowed speaking stage and be granted speaking opportunities by the host;

[0132] A fixed time window is set, and the speech score and ranking of all participants are recalculated in the time window and the speaking rights are adjusted in real time.

[0133] By collecting real-time data including the number of speeches, speech duration, content relevance, and interactive responses, the system calculates the interactive edge weights between participants, dynamically adapts to real-time changes during the meeting, identifies high-frequency interactive participants, and provides reliable basic data for subsequent interactive graph generation, thereby improving the accuracy of meeting management. Through the iterative updating of the initial feature vectors of the participants and the interactive edge weights by Graph Neural Network (GNN), the system can generate an interactive graph that fully reflects the interaction between participants. The core advantage of GNN is that it can effectively capture the complex multi-dimensional interaction between participants, and each iteration can integrate the feature information of neighboring nodes to dynamically update the feature vectors of each participant. Compared with the existing technology that relies on simple statistical methods to construct an interactive graph, GNN can more efficiently process complex meeting interaction networks, generate more accurate interactive graphs, and provide strong data support for real-time adjustment of speaking rights and dynamic management of meeting processes. After the generation of the interactive graph, the system calculates the centrality score of each participant in the network based on their feature vector centrality. The feature vector centrality can reflect the influence of each participant in the interactive network, and participants with higher centrality scores often have more interactive connections and stronger relevance. In this way, the system can identify key participants in the meeting and provide a scientific basis for subsequent speaking right setting and dynamic adjustment. This speaking right management method based on centrality scores can ensure that participants with higher meeting contribution have more speaking opportunities, thereby improving the quality and efficiency of the meeting discussion. The calculation of the incremental score combines the real-time performance of each participant during the meeting, including the number of speeches, speech duration, content relevance, and interactive responses. This mechanism ensures the rational allocation of meeting resources, avoids the waste of speaking resources, and encourages participants to actively participate in the meeting discussion, improving the overall meeting effectiveness. By ranking all participants according to their real-time speaking scores, the system can effectively identify high-contributing participants. This mechanism ensures that high-level participants can freely speak at any stage of the meeting, thereby improving the smoothness and efficiency of the discussion. Compared with the random speaking method in traditional meetings, this method reduces disordered speaking, making the speaking process more standardized. For the top 80% of medium-level participants, the mechanism requires them to apply for speaking when the meeting is paused or allowed, encouraging them to be more proactive in the discussion. By sorting the participants who apply for speaking, the system can effectively manage the speaking order and avoid chaos during speaking. For low-level participants, speaking requires the opportunity granted by the moderator, which provides the moderator with additional control, ensuring that the meeting can focus on major issues and important discussions. The moderator can flexibly decide which speeches are necessary based on the progress of the meeting, avoiding unnecessary interference and ensuring the efficiency of the meeting.

[0134] S4, extracting keywords of the speech content of the participant through intelligent speech recognition technology and dynamically adjusting the extracted keywords according to the change of the speech content, generating a real-time conference abstract, and pushing the generated conference abstract to the corresponding participant in real time according to the permission of the participant;

[0135] Specifically, the real-time conference abstract is generated by capturing the speech audio stream of the participant in real time, converting the preprocessed voice signal into text through a DeepSpeech model, cutting and preprocessing the text through natural language processing technology, and using an LDA model to train the theme of the preprocessed text paragraph and representing each text paragraph as a set of potential theme probability distribution:

[0136]

[0137] In the formula, E(r|z) is the probability of the occurrence of the word r under the theme z, which is a probability value, and the range is [0, 1], E(r|θ f ) is the probability of the occurrence of the word r in the fth theme, E(z f |α) is the probability of the theme z f in all possible themes, and alpha is a hyperparameter, which is used to control the theme distribution in each document, F is the number of themes, and the parameter controls the smoothness of the theme distribution, ensures that the model does not excessively deviate to certain themes, and is obtained by manual setting or learning during model training, and is usually set to a small value (such as 0.1 or 0.01) to ensure uniform distribution of different themes;

[0138] N themes are generated for each text paragraph, each theme contains multiple words, and the words are arranged according to the word frequency, and the top n keywords with the highest weight are extracted as the core content of the text paragraph, and the keyword weight B y is updated when a new speech segment is generated.

[0139]

[0140] In the formula, is the old weight of the keyword r, and gamma is a regulating factor used to control the amplitude of the keyword weight adjustment, which is set by grid search optimization, E(r|z new ) is the theme distribution probability of the keyword r in the newly generated text segment, which represents the probability of the keyword in the theme distribution of the current latest speech content, according to the analysis result of the LDA model on the new speech content, the probability of each keyword under different themes is calculated, and avg(E(r|z) is the average theme distribution probability of the keyword r during the entire conference.

[0141] The topic distribution of the LDA model clusters similar keywords into the same topic, and generates conference abstracts based on the clustering results and keyword weights;

[0142] The abstract content includes conference topics, key discussion points (the system extracts the core discussion content of the conference based on the weight and frequency of keywords), and decision information (extracts the speech content related to the decisions in the conference and generates a concise decision summary);

[0143] Classify the conference abstracts into public content, sensitive content, and content directly related to the participants according to content sensitivity, theme, and importance.

[0144] Through intelligent speech recognition technology, the system can capture the audio stream of the participants' speech in real time and convert it into text. This process uses the DeepSpeech model, which can handle complex speech signals in the speech and achieve accurate text output. Compared with traditional conference recording methods, the introduction of intelligent speech recognition not only greatly improves the efficiency of conference recording, but also reduces the errors in manual recording. It can provide high-quality text data for subsequent natural language processing and LDA model topic analysis, ensuring the accuracy and coherence of subsequent steps. Through the LDA model, the system can analyze the theme of each text paragraph, extract the potential theme in the text, and generate keywords according to the theme distribution. This process ensures that the core theme of the conference content can be effectively extracted and represented. Unlike existing keyword extraction methods, the LDA model can aggregate similar keywords into the same theme through probability distribution, thereby avoiding the theme ambiguity problem that occurs when keywords are extracted based on word frequency alone. The LDA model provides a more accurate text theme extraction method, ensuring that the system can extract the most representative theme and keywords from a large amount of speech content. By recalculating the keyword weight of the newly generated speech segment, the system can adjust the weight value of each keyword at any time. Ensuring the continuity and robustness of the keyword extraction process, the system can always extract the most relevant keywords when facing dynamic changes in conference content. Based on the theme and keywords generated by the LDA model, the system automatically generates conference abstracts according to keyword weights. The abstract content not only includes conference topics and key discussion points, but also extracts information related to conference decisions to generate concise decision summaries, greatly improving the efficiency and accuracy of conference minutes generation, and providing a tool for decision-makers to quickly review conference highlights. In particular, in complex decision-making conference scenarios, abstracts can help decision-makers efficiently grasp key conference information, and by classifying abstract content according to sensitivity, theme, and importance, the system can provide customized abstracts to different participants.

[0145] Further, the generated meeting summary is pushed to the corresponding participant in real time according to the participant's authority, or the generated meeting summary is matched according to the authority level of each participant, the exclusive meeting summary is generated for each participant according to the authority matrix, the presentation order of the summary content is further optimized according to the historical preference and interaction order of the participant, and the theme that the participant often pays attention to is preferentially displayed at the beginning of the summary;

[0146] According to the device type and preference of each participant, the best push mode is selected to push the meeting summary to the corresponding participant;

[0147] The push mode includes email push, mobile application notification and Web push;

[0148] According to the conference process and the update frequency of the content, the timing of the push is determined. For example, real-time push: during the conference process, the system can push the summary of the conference process in real time, such as important decision information; post-conference push: after the conference ends, the complete meeting summary is pushed to the participants, ensuring that they can review the conference content.

[0149] According to the authority level of the participant, the corresponding meeting summary is matched to ensure that the distribution of key information is more accurate and effective. Through the authority matrix, the system can effectively prevent the leakage of sensitive information and ensure the security of the conference document. The historical preference and interaction order of the participant directly affect the sorting of the meeting summary. In this way, the participant can see the issues they care about at the first time, reducing the interference of irrelevant information and significantly improving the user experience. By preferentially displaying the theme that is often concerned, the participant can more efficiently obtain and process information, reducing the time spent searching for relevant content in a large number of conference documents. The system adapts the push according to the device type of the participant, ensuring that each participant can receive the meeting summary on the most convenient device.

[0150] Embodiment 2, refer to Figure 2 For the second embodiment of the present application, which is different from the previous embodiment, an intelligent meeting information management system is provided, which comprises,

[0151] The authority management module is used for identity authentication of the participant, generating an authority matrix, and assigning authority according to the historical behavior data of the participant;

[0152] The document distribution module is used for collecting conference documents and distributing the documents based on the participant's authority and historical behavior;

[0153] The conflict detection module is used for real-time detection of interest conflicts between participants and adjustment of participant authority according to the detection result;

[0154] The permission adjustment module is used for monitoring the interaction of the participants in real time, analyzing the interaction behavior through a graph neural network, generating an interaction graph, and dynamically adjusting the speaking permission.

[0155] The abstract pushing module is used for processing the speaking content of the participants through intelligent voice recognition technology, extracting keywords, generating a conference abstract, and pushing personalized conference abstracts according to the permission levels of the participants.

[0156] If the functions are realized in the form of software function units and sold or used as independent products, they can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application essentially or the parts that make contributions to the prior art or parts of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device) to execute all or part of the steps of the method described in the various embodiments of the present application. The aforementioned storage medium includes a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various media that can store program codes.

[0157] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered a list of executable instructions for implementing logical functions and can be specifically embodied in any computer-readable medium for use by or in connection with an instruction execution system, apparatus, or device, such as a computer-based system, a processor-based system, or other system that can fetch the instructions from the instruction execution system, apparatus, or device and execute the instructions, or in conjunction with these instructions. For the purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device, or in conjunction with these instructions.

[0158] More specific examples (a non-exhaustive list) of the computer-readable medium include the following: an electrical connection having one or more wires (electrical devices), a portable computer diskette (magnetic devices), a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disc read-only memory (CDROM). In addition, the computer-readable medium can even be paper or other suitable medium on which the program can be printed, because the program can be electronically obtained, for example, by optical scanning of the paper or other medium, followed by electronic editing, interpretation, or necessary processing, and then stored in a computer memory if necessary.

[0159] It should be understood that aspects of the application can be implemented in hardware, software, firmware or combinations thereof. In the embodiments described above, various steps or methods can be implemented, for example, in software or firmware that is stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, and in another embodiment, any of the following technology, known in the art, or combinations thereof, can be used: discrete logic circuitry having logic gates for implementing logic functions upon an application of data signals, application specific integrated circuits having appropriate combinational logic gates, programmable gate arrays (PGA), field programmable gate arrays (FPGA), and the like.

Claims

1. A method for intelligent management of meeting information, characterized in that: The method comprises the following steps: Identity authentication of participants is performed through multi-modal information, and a permission matrix is generated according to historical behavior data of the participants after the identity authentication; Core keywords and topics of all conference documents are extracted and classified according to document importance, the association degree between the documents and the participants is calculated based on the permission matrix and the document classification results, and the corresponding conference documents are allocated to each participant, the interest conflicts between the participants are detected in real time during the document allocation process, and the participant permissions are adjusted in real time according to the conflict detection results; The interaction of each participant is monitored in real time, the interaction behavior of each participant is analyzed through a graph neural network, an interaction graph is generated, the interaction relationship between the participants and high-value discussion nodes are identified, and the speaking permission of the participants is dynamically adjusted according to real-time interaction data; The keywords of the speech content of the participants are extracted through intelligent speech recognition technology, the extracted keywords are dynamically adjusted according to the changes in the speech content, real-time conference abstracts are generated, and the generated conference abstracts are pushed to the corresponding participants in real time according to the permissions of the participants.

2. The intelligent management method of meeting information according to claim 1, characterized in that: The method comprises the following steps: When the participants log in, the face images of the participants are collected in real time through the camera of the device, and the collected face images are preprocessed; The deep features of the preprocessed face images of the participants are extracted using a convolutional neural network, the similarity between the current face features and the pre-stored face features of the participants in the database is calculated using cosine similarity, a similarity threshold W is set, if the face similarity is greater than or equal to the similarity threshold W, the face recognition is passed, otherwise the recognition fails, and the participants are prompted to re-collect the images; After the face recognition verification is passed, a one-time verification code is sent according to the pre-set security information of the participants for identity authentication of the participants, the identity of the participants is confirmed after the participants pass the face recognition and the verification code verification, the historical participation data and the speech records of the participants are retrieved according to the identity information of the participants, and the historical behavior data of the participants are extracted; The conference types of discussion conferences, decision-making conferences and training conferences are preset, the organizers are required to select the corresponding conference type from the preset types according to the conference goal when the conference is created, and different behavior weights are set for the behavior data of the participants according to different conference types; The behavior score of the participants is calculated according to the historical behavior data through a weighted scoring formula; The preset permission levels are L1: the highest permission, L2: the medium permission, and L3: the lowest permission, each permission level corresponds to the document range and the speaking permission level that can be accessed by the participants, the participants are allocated to the corresponding permission level according to the behavior score of the participants, and the permission matrix of each participant is generated, and the document range that can be accessed by the participants is preliminarily defined.

3. The intelligent management method of meeting information according to claim 2, characterized in that: The core keywords and topics of all documents of the meeting are extracted, and the documents are classified according to the importance of the documents; the relevance between the documents and the participants is calculated based on the permission matrix and the classification results of the documents, and the corresponding meeting document collection is allocated to each participant; all documents related to the meeting are collected, the documents are uniformly formatted, and the text is cleaned; the text in the documents is cut into individual words through a word segmentation algorithm, and a word set of each document is generated; The frequency of each word in the document is calculated, and the importance of each word in all documents is measured using the inverse document frequency to obtain the TF-IDF value of each word; the TF-IDF values of each document are sorted in descending order to generate a keyword list, and a document keyword set is generated according to the keyword list; a participant keyword set is generated according to the historical behavior of the participant and the permission matrix; The relevance of a single document to all participants is calculated using the Jaccard similarity coefficient, and the calculated relevance is fused by weighted summation to generate a comprehensive relevance value; the relevance threshold is set as T1 and T2, and the relevance threshold is compared with the relevance value to classify the documents: If the relevance value is greater than or equal to T1, the document is a core document, and the document is allocated to the L1-level participant; If T2 is less than or equal to the relevance value and less than T1, the document is a general document, and the document is allocated to the L2-level participant; If the relevance value is less than T2, the document is a secondary document, and the document is allocated to the L3-level participant; All documents in the meeting are divided by using the Jaccard similarity coefficient calculation and threshold allocation method.

4. The intelligent management method of meeting information according to claim 3, characterized in that: The interest conflicts between the participants are detected in real time during the document allocation process, and the participant permissions are adjusted in real time according to the conflict detection results; the attributes of the participants are defined and an attribute set is generated, the attributes of each document are defined and a document attribute set is generated, and the participant attributes are matched; Potential interest conflict factors are extracted by analyzing the participant attributes and the document attributes; The conflict factors include discrete attribute conflict factors and continuous attribute conflict factors; The conflict of the discrete attribute is processed using logical judgment, a binary result is allocated to each conflict factor, and a conflict score is accumulated based on the judgment result of each factor; After normalization processing of each continuous attribute, the Euclidean distance is used to calculate the continuous attribute conflict value; The final conflict value C of the participant and the document is obtained by weighted addition of the discrete attribute conflict value and the continuous attribute conflict value ij ; The interest conflict value is calculated for each participant and document, a separate conflict value threshold U is set for each document, the interest conflict value is compared with the conflict value threshold, and the conflict situation is judged: If C ij >U, then it indicates that the participant i has a high conflict risk, and the access right of the participant i to the document j is immediately adjusted. If C ij ≤ U, then it indicates that the conflict risk of the participant i is low, and the participant i is allowed to continue accessing the document; The permission matrix is adjusted in real time according to the changed participant permissions.

5. The intelligent management method of meeting information according to claim 4, characterized in that: The real-time monitoring of the interaction of each participant and the analysis of the interaction behavior of each participant through the graph neural network generate an interaction map, identify the interaction relationship between the participants and the high-value discussion nodes, and dynamically adjust the speaking right of the participants according to the real-time interaction data il After setting the speaking right of the participants, the interaction data of each participant is collected in real time through sensors, microphones and cameras, the interaction edge weight w between each pair of participants is calculated according to the interaction data between each pair of participants, and after the calculation is completed, the edge weights between all participants are summarized to form an interaction intensity matrix, and the interaction intensity between each pair of participants is determined by the corresponding weight. Generating an initial feature vector for each participant According to the initial feature vector And the interaction edge weight constructs the graph G=(V, J), where V represents the participant node, each node represents a participant, and J represents the interaction edge between participants; The feature vectors of the participants are iteratively updated by the graph neural network; in each iteration, the feature vector of the participant node is updated by the information of its neighbor nodes; After the iteration of the graph neural network is completed, a complete interaction map is generated, the interaction map is visualized, and the node color is identified according to the relevance score of the speech content and the meeting theme; The centrality score of each participant is calculated based on the interaction map using the feature vector centrality, and the initial speech score of the participant is set according to the centrality score of the participant in the interaction map; The number of speeches, the duration and the relevance of the speeches of the participants are weighted to obtain an increment of real-time speech score, and the initial speech score and the increment of real-time speech score are combined to obtain the real-time speech score; The participants are ranked according to the real-time speech score of each participant, and a ranking list is dynamically updated, if the participant ranking is in the top 20%, the participant is a high-level participant, if the participant ranking is in 20%-80%, the participant is a medium-level participant, and if the participant ranking is in the last 20%, the participant is a low-level participant.

6. The intelligent management method of meeting information according to claim 5, characterized in that: The keyword extraction of the speech content of the participant by the intelligent speech recognition technology and the dynamic adjustment of the extracted keyword according to the change of the speech content are used to generate a real-time conference abstract, which refers to capturing the speech audio stream of the participant in real time, converting the preprocessed voice signal into text through the DeepSpeech model, cutting the text to generate text paragraphs through the natural language processing technology, and preprocessing the text paragraphs, using the LDA model to train the theme of the preprocessed text paragraphs, and representing each text paragraph as a set of potential theme probability distribution; Generate N topics for each text paragraph, each topic contains multiple vocabularies and arranges the vocabularies according to the frequency of vocabulary occurrence, extracts the top n keywords with the highest weight as the core content of the text paragraph, calculates the vocabulary distribution of the new segment when a new speech segment is generated, and updates the keyword weight B y ; The theme distribution of the LDA model is used to cluster similar keywords into the same theme, and the conference abstract is generated according to the clustering results and the keyword weight.

7. The intelligent management method of meeting information according to claim 6, characterized in that: The generated conference abstract is pushed to the corresponding participant in real time according to the participant's authority, which refers to matching the generated conference abstract according to the authority level of each participant, generating a dedicated conference abstract for each participant according to the authority matrix, and selecting the best push mode according to the device type and preference of each participant to push the conference abstract to the corresponding participant.

8. An intelligent meeting information management system based on the intelligent meeting information management method according to any one of claims 1-7, characterized in that: It includes, The authority management module is used for identity authentication of the participants, generating an authority matrix, and assigning authority according to the historical behavior data of the participants; The document allocation module is used for collecting conference documents and allocating documents based on the participant's authority and historical behavior; The conflict detection module is used for real-time detection of interest conflicts between participants and adjustment of participant authority according to the detection results; The authority adjustment module is used for real-time monitoring of the interaction of the participants, analyzing the interaction behavior through the graph neural network, generating an interaction map, and dynamically adjusting the speaking authority; The abstract pushing module is used for processing the speech content of the participants through intelligent speech recognition technology, extracting keywords, and generating conference abstracts, and pushing personalized conference abstracts according to the authority level of the participants.

9. A computer device comprising: The memory and the processor; the memory stores a computer program, characterized in that: the processor executes the computer program to realize the steps of the meeting information intelligent management method in any one of claims 1-7.

10. A computer readable storage medium having stored thereon a computer program, characterized in that: The computer program is executed by the processor to realize the steps of the meeting information intelligent management method in any one of claims 1-7.

Images