A smart management system for a service consultation platform

Abstract

The present invention discloses a smart management system for a service consultation platform, which includes: processing data through a data collection and processing module; constructing a feature matrix for the multi-source data after cleaning by using a data fusion unit; generating a prediction result of user consultation needs based on a demand prediction module; identifying user input through a semantic analysis and question-answering module; transmitting the parsed semantic information to a knowledge management module, retrieving matching content from a knowledge base based on a semantic similarity matching algorithm, generating an answer after sorting by priority; collecting information during the service process for scoring through a service quality dynamic evaluation module, formulating an optimization plan through a service optimization strategy generation module, and feeding back the optimization plan to the evaluation module to achieve closed-loop management. The present invention significantly improves the prediction ability and service efficiency of the platform, providing users with a more efficient and personalized consultation experience.

Description

Technical Field

[0001] The present invention relates to the technical field of platform intelligent management, and particularly to an intelligent management system for a service consultation platform. Background Art

[0002] With the rapid development of Internet technology and data processing capabilities, service consultation platforms have become an important way for users to obtain professional knowledge and solve problems. However, traditional service consultation platforms face significant challenges in demand prediction and resource matching. Existing technologies usually rely on static rules or single data dimensions for demand judgment, and are unable to dynamically capture the time characteristics, group characteristics, and cross-dimensional interaction relationships of user behaviors. In addition, traditional collaborative filtering methods are only limited to simple user-service interactions, lacking in-depth analysis of the time dimension and user behavior characteristics; while clustering analysis models are mostly based on fixed features, ignoring the dynamics and relevance of user needs. The above deficiencies have led to low accuracy and efficiency of service recommendations, unable to meet the personalized and diverse needs of users. Therefore, there is an urgent need for an innovative technical solution that can integrate time dynamics, behavior characteristics, and group demand characteristics to achieve accurate prediction of user consultation needs and efficient resource matching. Summary of the Invention

[0003] In view of the above existing problems, the present invention is proposed.

[0004] Therefore, the present invention provides an intelligent management system for a service consultation platform, which can solve the problems that existing platforms insufficiently capture the time dynamic characteristics of user behaviors, cannot reflect the short-term changes, periodic laws, and long-term trends of user needs; the interaction relationship modeling between users and services is insufficient, lacking the high-order expression ability for complex interaction patterns; traditional clustering methods do not accurately mine the group demand characteristics of users, ignoring the dynamic relevance of user characteristics. At the same time, the result fusion mechanism of existing prediction models lacks dynamic adjustment ability and is difficult to adapt to different scenario requirements. Through the improved time dynamic weight function, high-order tensor decomposition method, and user demand clustering model based on spectral clustering of the present invention, combined with the multi-model dynamic weight fusion mechanism, the dynamics, accuracy, and adaptability of user demand prediction are effectively improved, providing more efficient prediction and resource matching technical support for the service consultation platform.

[0005] To solve the above technical problems, the present invention provides the following technical solution. An intelligent management system for a service consultation platform includes:

[0006] Through a data collection and processing module, user behavior data, historical consultation data, and external industry trend data are collected, and the collected data is cleaned and formatted to obtain multi-source data;

[0007] The data fusion unit constructs a feature matrix based on the cleaned multi-source data to generate a multi-dimensional feature vector of user requirements;

[0008] Based on the collaborative filtering algorithm and clustering analysis model in the demand prediction module, generate a prediction result of user consultation requirements;

[0009] Through the semantic parsing and question-answering module, perform word segmentation, grammar analysis, and intention recognition on the natural language input by the user;

[0010] Transmit the parsed semantic information to the knowledge management module, retrieve matching content from the knowledge base based on the semantic similarity matching algorithm, and generate answers after sorting by priority;

[0011] Through the service quality dynamic evaluation module, collect user feedback data, expert behavior data, and platform operation data during the service process, and perform scoring. Based on the scoring results, formulate an optimization plan through the service optimization strategy generation module, use the resource scheduling module to adjust the expert resource configuration, optimize the recommended content, and feedback the optimization plan to the evaluation module to achieve closed-loop management.

[0012] As a preferred solution of the intelligent management system of the service consultation platform described in the present invention, wherein: the data collection and processing module includes a user behavior collection unit, a historical consultation data collection unit, an industry trend collection unit, a data cleaning unit, and a feature extraction unit;

[0013] The user behavior collection unit is used to collect user behavior data of users, including: browsing records, click behaviors, and search keywords; the historical consultation data collection unit is used to collect historical consultation data, including recording the consultation topics of users, expert selections, and corresponding scoring records; the industry trend collection unit is used to obtain external industry trend data, including industry hot topics, time distribution data, and external public information; the data cleaning unit is used to remove invalid or duplicate data and unify the data format; the feature extraction unit is used to perform dimensional regularization and feature enhancement processing on the collected user behavior data, historical consultation data, and external industry trend data.

[0014] As a preferred solution of the intelligent management system of the service consultation platform described in the present invention, wherein: the data fusion unit includes numerically processing non-numerical data through a feature embedding algorithm, and unifying user behavior, time dimension, and historical records into a multi-dimensional feature representation of user requirements by constructing a feature matrix, where the non-numerical data includes keywords and behavior tags.

[0015] As a preferred solution of the intelligent management system of the service consultation platform described in the present invention, wherein: the construction of the feature matrix includes, for text data Use a context-aware embedding model:

[0016] ;

[0017] Where:

[0018] ;

[0019] represents the input embedding matrix, E is the word embedding matrix, and R is the set of real numbers. is the text length. is the dimension. represents the BERT-based embedding function for generating context-aware text embeddings.

[0020] are the weight matrices of queries, keys, and values in the BERT model. Q, K, and V represent the query, key, and value matrices respectively. MultiHead represents the multi-head attention mechanism, and the output ; represents the text embedding matrix.

[0021] For the behavior label use graph convolutional network embedding: ;

[0022] Where represents the input feature matrix of the th layer, represents the label correlation matrix; m represents the number of label nodes. represents the input feature matrix of the th layer; represents the weight matrix of graph convolution; represents the activation function, and the output is the label embedding matrix ; m represents the label feature dimension; k represents the dimension of the label embedding space, which is the number of dimensions for each label embedding.

[0023] ;

[0024] Where represents the hidden state at time ; represents the input data at time ; represents the weight matrix of the hidden state; represents the weight matrix of the input data; b represents the bias term. represents the activation function, and the final output is the time series embedding matrix ; h represents the dimension of the state vector at time step t; t represents the time series feature dimension.

[0025] Extended time weight function , polynomial fitting is introduced for dynamic adjustment:

[0026] ;

[0027] Among them, represents the behavior frequency, represents the behavior category, represents the behavior category weight coefficient of, represents the time decay factor, is the decay parameter, represents the logarithmic processing of the behavior frequency, represents the high-order polynomial term of the time series, represents the order of the polynomial, represents the coefficient of the time polynomial term;

[0028] ;

[0029] Among them, represents the fused feature tensor, represents the number of samples in the dataset, represents the weighted value of the th data point, represents the text feature matrix, represents the label feature matrix, represents the time feature matrix, is the number of samples, is the dimension of the time feature,

[0030] As a preferred solution of the intelligent management system of the service consultation platform described in the present invention, wherein: the construction of the feature matrix further includes using HOSVD to extract the core interaction relationship of multimodal features and reducing the high-dimensional tensor to a dense representation;

[0031] Perform high-order singular value decomposition on the multimodal embedding tensor : ;

[0032] Among them, represents the fused feature tensor, with a dimension of , represents the number of samples, represents the dimension of the text feature, represents the dimension of the label feature, represents the dimension of the time feature, represents the core tensor after tensor decomposition, The factor matrix for text features, representing the low-dimensional representation on the text dimension obtained through tensor decomposition, with the dimension being , is the decomposition rank of the text features, The factor matrix for label features, representing the low-dimensional representation on the label dimension obtained through tensor decomposition, with the dimension being , is the decomposition rank of the label features, The factor matrix for time features, representing the low-dimensional representation on the time dimension obtained through tensor decomposition, with the dimension being , is the decomposition rank of the time features, The outer product operation of the tensor, representing multiplying the core tensor separately with the factor matrices to obtain the low-dimensional representation after dimensionality reduction;

[0033] Concatenate the core tensor after tensor decomposition with the time series features: ;

[0034] Among them, represents unfolding the core tensor along the first dimension into a matrix; represents the concatenated feature matrix, representing the multi-dimensional features of user requirements;

[0035] Optimize the final feature representation through the contrastive learning mechanism, and the loss function is:

[0036] ;

[0037] Among them, represents the loss function of contrastive learning, used to maximize the similarity between positive sample pairs and minimize the similarity between negative sample pairs, is the embedded feature vector of the th sample, is the positive sample feature vector paired with it, is the feature vector of the negative sample, is the temperature parameter, is the exponential function, is the dot product of the positive sample pair, represents the summation operation for all negative samples, is the total number of negative samples.

[0038] As a preferred solution of the intelligent management system of the service consulting platform described in the present invention, wherein: the demand prediction module includes a collaborative filtering algorithm and a clustering analysis model;

[0039] The collaborative filtering algorithm is used to predict the potential needs of users based on the behavior patterns of similar users; the clustering analysis model is used to extract demand features and divide user groups according to the demand features.

[0040] As a preferred solution of the intelligent management system of the service consultation platform according to the present invention, wherein: the semantic parsing and question answering module includes a word segmentation unit, a syntax analysis unit, a semantic similarity calculation unit, and a context management unit;

[0041] The word segmentation unit is used to segment the natural language questions input by users; the syntax analysis unit is used to extract the syntactic structure and semantic content in the questions through dependency syntax analysis; the semantic similarity calculation unit is used to quantify the similarity between the question semantics and the content in the knowledge base; the context management unit is used to record the context information of the multi-round question interaction of users to dynamically adjust the parsing results.

[0042] As a preferred solution of the intelligent management system of the service consultation platform according to the present invention, wherein: the semantic similarity matching algorithm includes parsing the semantics input by users through a multi-head attention mechanism, and fusing multi-modal data by combining image features, time decay weights, and user portrait information to construct a user demand representation. At the same time, in the semantic matching process, a similarity calculation method is adopted, and based on the text content, the graph relationship and context information in the knowledge base are combined to improve the accuracy of matching, and through the graph neural network GNN, the semantic association between entries is optimized, and context-aware similarity is introduced to aggregate the semantic information in the knowledge base neighborhood;

[0043] The entries in the knowledge base are sorted by priority through a multi-factor scoring mechanism, comprehensively considering the historical usage frequency, expert scores, and time decay factors of the entries, and a contrastive learning method is used to optimize the answering process to ensure that the system can select the knowledge entries that best match the user's needs.

[0044] As a preferred solution of the intelligent management system of the service consultation platform according to the present invention, wherein: the semantic parsing and question answering module combines the retrieval results of the knowledge management module, generates answers through priority sorting, supports the decomposition processing of the questions input by users, and realizes multi-round interactive question answering by combining the context memory function.

[0045] As a preferred solution of the intelligent management system of the service consultation platform according to the present invention, wherein: the service optimization strategy generation module includes an expert resource optimization unit, a user guidance optimization unit, a resource scheduling optimization unit, and a feedback unit;

[0046] The expert resource optimization unit is used to dynamically adjust the matching relationship between experts and consultation topics; the user guidance optimization unit is used to generate recommended service content according to the user demand prediction results; the resource scheduling optimization unit is used to adjust the platform resource allocation strategy; the feedback unit is used to re-enter the optimized data into the service quality dynamic evaluation module to form a closed-loop management process.

[0047] Advantages of the present invention: The present invention introduces an improved time dynamic weight function to fully capture the time characteristics of user behavior, covering short-term changes, periodic patterns, and long-term trends, thereby improving the accuracy of demand prediction. Secondly, based on the high-order tensor decomposition method, a joint model of the three-dimensional interaction relationship among users, services, and time is constructed, breaking through the limitations of traditional methods in expressing complex demand patterns. Thirdly, the spectral clustering is used to optimize the user clustering model to deeply explore the common characteristics of group demands, providing a solid foundation for accurate resource matching. In addition, the present invention uses a multi-model dynamic weight fusion mechanism to flexibly adjust the contribution weights of time series, collaborative filtering, and clustering analysis to meet the requirements of diverse application scenarios. In summary, the present invention significantly improves the prediction ability and service efficiency of the platform, providing users with a more efficient and personalized consultation experience. Description of the Drawings

[0048] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for the description of the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0049] Figure 1 It is a schematic diagram of the framework of a service consultation platform intelligent management system provided by an embodiment of the present invention. Detailed Embodiments

[0050] To make the above objects, features, and advantages of the present invention more obvious and understandable, the specific embodiments of the present invention will be described in detail below with reference to the drawings of the specification. Obviously, the described embodiments are some embodiments of the present invention, rather than all embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0051] Many specific details are set forth in the following description to facilitate a thorough understanding of the present invention. However, the present invention may be practiced in other ways different from those described herein. Those skilled in the art can make similar extensions without departing from the spirit of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0052] Second, the "one embodiment" or "embodiment" referred to herein means a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The appearances of "in one embodiment" in different places in this specification do not all refer to the same embodiment, nor are they separate or alternative embodiments that exclude each other from other embodiments.

[0053] The present invention will be described in detail with reference to the schematic diagrams. When describing the embodiments of the present invention in detail, for the convenience of explanation, the cross-sectional views showing the device structure will be enlarged locally out of the general scale, and the schematic diagrams are only examples and should not limit the scope of protection of the present invention here. In addition, in actual production, three-dimensional spatial dimensions of length, width, and depth should be included.

[0054] Meanwhile, in the description of the present invention, it should be noted that the orientation or positional relationship indicated by terms such as "upper, lower, inner, and outer" is based on the orientation or positional relationship shown in the drawings, and is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be construed as a limitation to the present invention. In addition, the terms "first, second, or third" are only used for descriptive purposes and cannot be construed as indicating or implying relative importance.

[0055] Unless otherwise clearly defined and limited in the present invention, the terms "installed, connected, and coupled" should be understood in a broad sense. For example, it may be a fixed connection, a detachable connection, or an integral connection; it may also be a mechanical connection, an electrical connection, or a direct connection, and may also be indirectly connected through an intermediate medium, or may be the communication inside two elements. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0056] Embodiment 1, referring to Figure 1 , which is the first embodiment of the present invention. This embodiment provides a smart management system for a service consultation platform, including:

[0057] Through the data collection and processing module, user behavior data, historical consultation data, and external industry trend data are collected, and the collected data is cleaned and formatted to obtain multi-source data. The data fusion unit constructs a feature matrix based on the cleaned multi-source data to generate a multi-dimensional feature vector of user needs. Based on the collaborative filtering algorithm and clustering analysis model in the demand prediction module, a prediction result of user consultation needs is generated. The natural language input by the user is segmented, grammatically analyzed, and intention recognized through the semantic parsing and question answering module. The parsed semantic information is transmitted to the knowledge management module, and matching content is retrieved from the knowledge base based on the semantic similarity matching algorithm, and answers are generated after sorting by priority. Through the service quality dynamic evaluation module, user feedback data, expert behavior data, and platform operation data during the service process are collected and scored. Based on the scoring results, an optimization plan is formulated through the service optimization strategy generation module, and the expert resource configuration is adjusted and the recommended content is optimized using the resource scheduling module, and the optimization plan is fed back to the evaluation module to achieve closed-loop management.

[0058] The data collection and processing module includes a user behavior collection unit, a historical consultation data collection unit, an industry trend collection unit, a data cleaning unit, and a feature extraction unit;

[0059] The user behavior collection unit is used to collect user behavior data of users, including browsing records, click behaviors, and search keywords. The historical consultation data collection unit is used to collect historical consultation data, including recording the consultation topics of users, expert selection, and corresponding scoring records. The industry trend collection unit is used to obtain external industry trend data, including industry hot topics, time distribution data, and external public information. The data cleaning unit is used to remove invalid or duplicate data and unify the data format. The feature extraction unit is used to perform dimensional regularization and feature enhancement processing on the collected user behavior data, historical consultation data, and external industry trend data.

[0060] The data fusion unit includes numerically processing non-numerical data through a feature embedding algorithm, and unifying user behavior, time dimension, and historical records into a multi-dimensional feature representation of user needs by constructing a feature matrix, where the non-numerical data includes keywords and behavior tags.

[0061] The construction of the feature matrix includes using a context-aware embedding model:

[0062] ;

[0063] where:

[0064] ;

[0065] Denote the input embedding matrix, E is the word embedding matrix, and R is the set of real numbers. is the text length, and is the dimension; denotes the BERT-based embedding function for generating context-aware text embeddings;

[0066] are the weight matrices of queries, keys, and values in the BERT model. Q, K, and V respectively represent the query, key, and value matrices; MultiHead represents the multi-head attention mechanism, and the output ; denotes the text embedding matrix;

[0067] For the behavior label use graph convolutional network embedding: ;

[0068] where denotes the input feature matrix of the th layer, denotes the label correlation matrix; m represents the number of label nodes; denotes the input feature matrix of the th layer; denotes the weight matrix of graph convolution; denotes the activation function, and the output is the label embedding matrix ; m represents the label feature dimension; k represents the dimension of the label embedding space, which is the number of dimensions for each label embedding;

[0069] ;

[0070] where denotes the hidden state at time t; denotes the input data at time t; denotes the weight matrix of the hidden state; denotes the weight matrix of the input data; b represents the bias term; denotes the activation function, and the final output is the time series embedding matrix ; h represents the dimension of the state vector at time step t; t represents the time series feature dimension;

[0071] Expand the time weight function and introduce polynomial fitting for dynamic adjustment:

[0072] ;

[0073] where denotes the behavior frequency, Indicates the behavior category, Indicates the behavior category The weight coefficient of, Indicates the time decay factor, Is the decay parameter, Indicates the logarithmic processing of the behavior frequency, Indicates the high-order polynomial terms of the time series, Indicates the order of the polynomial, Indicates the coefficient of the time polynomial term;

[0074] ;

[0075] Among them, Indicates the fused feature tensor, Indicates the number of samples in the dataset, Indicates the Weighted value of the th data point, Indicates the text feature matrix, Indicates the label feature matrix, Indicates the time feature matrix, Is the number of samples, Is the dimension of the time feature, Tensor outer product operation.

[0076] The construction of the feature matrix further includes extracting the core interaction relationship of multimodal features using HOSVD to reduce the high-dimensional tensor to a dense representation;

[0077] For the multimodal embedding tensor Perform high-order singular value decomposition:

[0078] ;

[0079] Among them, Indicates the fused feature tensor, with dimension , Indicates the number of samples, Indicates the dimension of the text feature, Indicates the dimension of the label feature, Indicates the dimension of the time feature, Indicates the core tensor after tensor decomposition, Is the factor matrix of the text feature, representing the low-dimensional representation on the text dimension obtained by tensor decomposition, with dimension , Is the decomposition rank of the text feature, Is the factor matrix of the label feature, representing the low-dimensional representation on the label dimension obtained by tensor decomposition, with dimension , is the decomposition rank of the label feature, is the factor matrix of the time feature, representing the low-dimensional representation in the time dimension obtained through tensor decomposition, with the dimension of , is the decomposition rank of the time feature, is the outer product operation of the tensor, indicating that the core tensor is respectively multiplied with the factor matrix to obtain the low-dimensional representation after dimensionality reduction;

[0080] The core tensor after tensor decomposition is concatenated with the time series feature: ;

[0081] Among them, represents that the core tensor is unfolded into a matrix along the first dimension; represents the concatenated feature matrix, representing the multi-dimensional features of the user's needs;

[0082] The final feature representation is optimized through the contrastive learning mechanism, and the loss function is:

[0083] ;

[0084] Among them, represents the loss function of contrastive learning, which is used to maximize the similarity between positive sample pairs and minimize the similarity between negative sample pairs, is the embedded feature vector of the th sample, is the positive sample feature vector paired with it, is the feature vector of the negative sample, is the temperature parameter, is the exponential function, is the dot product of the positive sample pair, represents the summation operation for all negative samples, is the total number of negative samples.

[0085] The demand prediction module includes a collaborative filtering algorithm and a clustering analysis model;

[0086] The collaborative filtering algorithm is used to predict the possible potential demands of users based on the behavior patterns of similar users; the clustering analysis model is used to extract demand features and divide user groups according to the demand features.

[0087] The semantic parsing and question answering module includes a word segmentation unit, a syntax analysis unit, a semantic similarity calculation unit, and a context management unit;

[0088] The word segmentation unit is used to segment the natural language question input by the user; the syntactic analysis unit is used to extract the syntactic structure and semantic content in the question through dependency syntactic analysis; the semantic similarity calculation unit is used to quantify the similarity between the question semantics and the content in the knowledge base; the context management unit is used to record the context information of the user's multi-round question interaction to dynamically adjust the parsing result.

[0089] The semantic similarity matching algorithm includes parsing the semantics of the user input through a multi-head attention mechanism, and fusing multi-modal data by combining image features, time decay weights, and user portrait information to construct a representation of the user's needs. At the same time, in the semantic matching process, a similarity calculation method is adopted. Based on the text content, combined with the graph relationship and context information in the knowledge base, the accuracy of the matching is improved, and through the graph neural network GNN, the semantic association between entries is optimized, introducing context-aware similarity, and aggregating the semantic information in the knowledge base neighborhood;

[0090] The entries in the knowledge base are sorted by priority through a multi-factor scoring mechanism, comprehensively considering the historical usage frequency, expert ratings, and time decay factors of the entries, and a contrastive learning method is used to optimize the answering process to ensure that the system can select the knowledge entry that best matches the user's needs.

[0091] The semantic parsing and question answering module combines the retrieval results of the knowledge management module, generates answers through priority sorting, supports the decomposition processing of the questions input by the user, and realizes multi-round interactive question answering in combination with the context memory function.

[0092] The service optimization strategy generation module includes an expert resource optimization unit, a user guidance optimization unit, a resource scheduling optimization unit, and a feedback unit;

[0093] The expert resource optimization unit is used to dynamically adjust the matching relationship between experts and consultation topics; the user guidance optimization unit is used to generate recommended service content according to the predicted results of user needs; the resource scheduling optimization unit is used to adjust the platform resource allocation strategy; the feedback unit is used to re-enter the optimized data into the service quality dynamic evaluation module to form a closed-loop management process.

[0094] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered within the scope of the claims of the present invention.

[0095] Embodiment 2, the second embodiment of the present invention, which is different from the previous embodiment:

[0096] When the above-mentioned functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs that can store program codes.

[0097] This application is described with reference to the flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to the embodiments of this application. It should be understood that each flow and / or block in the flowchart and / or block diagram, as well as the combination of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate a device for implementing the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0098] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in this computer-readable memory generate a manufactured article including an instruction device, and this instruction device implements the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0099] These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process. Thus, the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in Figure 1 one flow or multiple flows and / or blocks Figure 1 one block or multiple blocks.

[0100] Although the preferred embodiments of the present application have been described, additional changes and modifications can be made to these embodiments by those skilled in the art once they learn the basic creative concept. Therefore, the appended claims are intended to be construed to include the preferred embodiments as well as all changes and modifications that fall within the scope of the present application.

[0101] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalent technologies, the present application is also intended to include these modifications and variations.

Claims

1. A service consulting platform intelligent management system, characterized by: include, Through the data collection and processing module, user behavior data, historical consulting data and external industry trend data are collected, and the collected data is cleaned and formatted to obtain multi-source data; The data fusion unit is used to construct a feature matrix based on the cleaned multi-source data to generate a multi-dimensional feature vector of user needs; Generate forecast results of user consultation needs based on collaborative filtering algorithm and cluster analysis model in demand forecast module; The semantic parsing and question-answering modules are used to segment, parse and identify the natural language input by users. The parsed semantic information is passed to the knowledge management module, and matching content is retrieved from the knowledge base based on the semantic similarity matching algorithm, and answers are generated after sorting by priority; Through the service quality dynamic evaluation module, user feedback data, expert behavior data and platform operation data in the service process are collected and scored. Based on the scoring results, the service optimization strategy generation module formulates an optimization plan, uses the resource scheduling module to adjust the expert resource allocation, optimizes the recommended content, and feeds the optimization plan back to the evaluation module to achieve closed-loop management; The data fusion unit includes: processing non-numeric data into digitized form through feature embedding algorithm, and unifying user behavior, time dimension and historical record into multi-dimensional feature representation of user demand through constructing feature matrix, wherein non-numeric data includes keywords and behavior labels; The constructing feature matrix includes: Using context-aware embedding models: in: Q=W Q E,K=W K E,V=W V E; E∈R n×d represents the input embedding matrix, E is the word embedding matrix, R is the set of real numbers, n is the text length, and d is the dimension; f bert (·) represents the BERT-based embedding function for generating context-aware text embeddings; W Q , W K , W V is the weight matrix of query, key and value in BERT model, Q, K, V represent query, key and value matrix respectively; MultiHead represents multi-head attention mechanism, output X text ∈R n×d ;X text represents the text embedding matrix; Behavior Labels Using graph convolutional network embedding: in, Represents the input feature matrix of the l+1th layer, A∈R m×m represents the label association matrix; m represents the number of label nodes; represents the input feature matrix of the lth layer; W (l) represents the weight matrix of graph convolution; σ represents the activation function, and the output is the label embedding matrix X label ∈R m×k ; k represents the dimension of the label embedding space, which is the number of dimensions for each label embedding; h t =σ(W h h t-1 +W x x t +b); Among them, h t ∈R h represents the hidden state at time t; x t ∈R d represents the input data at time t; W h The weight matrix representing the hidden state; W x represents the weight matrix of the input data; b represents the bias term; σ represents the activation function, and the final output is the time series embedding matrix X time ∈R t×h ; h represents the dimension of the state vector at time t; t represents the time series feature dimension; Expand the time weight function w(t, f, c) and introduce polynomial fitting for dynamic adjustment: Among them, f represents the behavior frequency, c represents the behavior category, α c represents the weight coefficient of behavior category c, e -λt represents the time attenuation factor, λ is the attenuation parameter, log(1+f) represents the logarithmic processing of the behavior frequency, represents the high-order polynomial term of the time series, p represents the order of the polynomial, β i represents the coefficients of the time polynomial terms; in, represents the fused feature tensor, n1 represents the number of samples in the dataset, w(t q , f q , c q ) represents the weighted value of the qth data point, is a tensor outer product operation; The constructing of the feature matrix also includes extracting the core interaction relationship of the multimodal features using HOSVD to reduce the high-dimensional tensor into a dense representation; right Perform a higher-order singular value decomposition: in, Represents the decomposed feature tensor with a dimension of n1×d text ×d label ×d time , d text Represents the dimension of text features, d label Represents the dimension of label features, d time The dimension representing the time feature, Represents the core tensor after tensor decomposition, U1 is the factor matrix of text features, which represents the low-dimensional representation of the text dimension obtained by tensor decomposition, with a dimension of d text ×r1, r1 is the decomposition rank of the text feature, U2 is the factor matrix of the label feature, which represents the low-dimensional representation on the label dimension obtained by tensor decomposition, with a dimension of d label ×r2, r2 is the decomposition rank of the label feature, U3 is the factor matrix of the time feature, which represents the low-dimensional representation on the time dimension obtained by tensor decomposition, with a dimension of d time ×r3, r3 is the decomposition rank of the time feature, ×1, ×2, ×3 are the outer product operations of the tensor, which means that the core tensor Perform product operations with factor matrices U1, U2, and U3 respectively to obtain low-dimensional representations after dimensionality reduction; The core tensor after decomposing the tensor Concatenate with time series features: in, Represents the core tensor expanded into a matrix along the first dimension; Represents the concatenated feature matrix, which represents the multi-dimensional features of user needs; The final feature representation is optimized by contrastive learning mechanism, and the loss function is: in, represents the loss function of contrastive learning, which is used to maximize the similarity between positive sample pairs and minimize the similarity between negative sample pairs, z w is the embedding feature vector of the wth sample, z j is the paired positive sample feature vector, z o is the eigenvector of the negative sample, τ is the temperature parameter, exp(x) is the exponential function, z w , z j is the dot product of the positive sample pair, represents the summation operation of all negative samples, and N is the total number of negative samples.

2. A service consultation platform intelligent management system as claimed in claim 1, characterized in that: The data collection and processing module includes a user behavior collection unit, a historical consultation data collection unit, an industry trend collection unit, a data cleaning unit, and a feature extraction unit; The user behavior collection unit is used to collect the user behavior data of the user, including browsing history, click behavior and search keywords; the historical consultation data collection unit is used to collect historical consultation data, including recording the user's consultation topic, expert selection and corresponding rating records; The industry trend collection unit is used to obtain external industry trend data, including industry hot topics, time distribution data and external public information; the data cleaning unit is used to remove invalid or duplicate data and unify the data format; the feature extraction unit is used to perform dimension normalization and feature enhancement processing on the collected user behavior data, historical consulting data and external industry trend data.

3. A service consultation platform intelligent management system as claimed in claim 2, characterized in that: The demand forecasting module includes collaborative filtering algorithm and cluster analysis model; The collaborative filtering algorithm is used to predict the potential needs of users based on the behavior patterns of similar users; the cluster analysis model is used to extract demand characteristics and divide user groups according to the demand characteristics.

4. A service consultation platform intelligent management system as claimed in claim 3, characterized in that: The semantic parsing and question-answering module includes a word segmentation unit, a grammatical analysis unit, a semantic similarity calculation unit, and a context management unit; The word segmentation unit is used to segment the natural language question input by the user; the grammatical analysis unit is used to extract the grammatical structure and semantic content in the question through dependency syntactic analysis; The semantic similarity calculation unit is used to quantify the similarity between the question semantics and the knowledge base content; The context management unit is used to record the context information of multiple rounds of user question interactions to dynamically adjust the parsing results.

5. A service consultation platform intelligent management system as claimed in claim 4, characterized in that: The semantic similarity matching algorithm includes parsing the semantics of user input through a multi-head attention mechanism, and combining image features, time decay weights and user portrait information to fuse multimodal data to construct a user demand representation. At the same time, in the semantic matching process, a similarity calculation method is used to improve the matching accuracy based on text content and combined with graph relationships and context information in the knowledge base. The semantic association between items is optimized through the graph neural network GNN, and context-aware similarity is introduced to aggregate the semantic information of the neighborhood of the knowledge base. The entries in the knowledge base are prioritized through a multi-factor scoring mechanism, and the historical usage frequency, expert ratings and time decay factors of the entries are comprehensively considered. The comparative learning method is used to optimize the answering process to ensure that the system can select the knowledge entries that best match user needs.

6. A service consultation platform intelligent management system as claimed in claim 5, characterized in that: The semantic analysis and question-answering module combines the search results of the knowledge management module, generates answers through priority sorting, supports the decomposition and processing of questions input by users, and realizes multiple rounds of interactive question answering in combination with the context memory function.

7. A service consultation platform intelligent management system as claimed in claim 6, characterized in that: The service optimization strategy generation module includes an expert resource optimization unit, a user guidance optimization unit, a resource scheduling optimization unit, and a feedback unit; The expert resource optimization unit is used to dynamically adjust the matching relationship between experts and consulting topics; the user guidance optimization unit is used to generate recommended service content according to the user demand prediction results; The resource scheduling optimization unit is used to adjust the platform resource allocation strategy; The feedback unit is used to re-input the optimized data into the service quality dynamic evaluation module to form a closed-loop management process.

Images