Information technology service consultation platform based on intelligent software

Abstract

The invention discloses an information technology service consultation platform based on intelligent software, which is characterized in that when keywords of a user consultation text are extracted, the user consultation text and semantic information of each word in the text are introduced into a weight iteration optimization processing flow of each node in a word graph model; according to the method, a user consultation text and semantic information of words in the text are utilized to allocate the restart probability and the edge weight of each node, so that a word graph model can be combined with semantic features of the text and the words to extract keywords; on the basis, the accuracy of keyword extraction can be improved, so that the accuracy of subsequent keyword-based information consultation retrieval matching is ensured; therefore, compared with manual service, the information consultation efficiency is improved, and compared with a traditional intelligent consultation system, the accuracy and reliability of an information consultation result are greatly improved, so that the system is very suitable for large-scale application and popularization in the field of information consultation service.

Description

Technical Field

[0001] The present invention belongs to the technical field of information consulting, and particularly relates to an information technology service consulting platform based on intelligent software. Background Art

[0002] With the rapid development of society, the growth rate of knowledge is getting faster and faster, and currently it has shown an explosive growth trend; and the explosively growing knowledge makes it impossible for people to be experts in every knowledge field. Therefore, when people want to understand the professional knowledge of some unfamiliar knowledge fields, how to quickly find experts in that knowledge field and obtain consulting services from experts in that knowledge field has become a key technology in information consulting.

[0003] Currently, traditional information consulting platforms are served by a single human or a single intelligent system. Among them, for human services, it is necessary to connect to a human customer service, which has low efficiency, and the professional levels of each human customer service are different, resulting in certain deviations in consulting results, and these deviations are also likely to cause losses to the consultants; at the same time, most intelligent systems use keyword search technology to answer information consulting. Traditional word segmentation technology mostly segments words based on features such as the word frequency of words in the text, and it does not consider the semantic information of the text. In this way, the accuracy of keyword extraction is not high, which further reduces the accuracy of information consulting answers; thus, based on the above deficiencies, how to provide an information technology service consulting platform with high efficiency and high information consulting accuracy has become an urgent problem to be solved. Summary of the Invention

[0004] The purpose of the present invention is to provide an information technology service consulting platform based on intelligent software to solve the problems of low efficiency in using human services and poor consulting accuracy in using intelligent consulting systems in the prior art.

[0005] To achieve the above purpose, the present invention adopts the following technical solutions:

[0006] In a first aspect, an information technology service consulting platform based on intelligent software is provided, including:

[0007] A consulting client for obtaining a user consulting text and sending the user consulting text to a consulting server;

[0008] A consulting server for constructing a word graph model based on the user consulting text, wherein each node in the word graph model is used to represent a word in the user consulting text, and the nodes with co-occurrence relationships are connected by edges;

[0009] A consultation server, configured to calculate the restart probability and edge weight of each node in the word graph model according to the user consultation text and the semantic features of the corresponding words of each node, and perform weight optimization processing on each node in the word graph model based on the restart probability and edge weight of each node, so as to obtain the optimal weight of each node after the optimization processing;

[0010] A consultation server, configured to determine a consultation keyword set corresponding to the user consultation text according to the optimal weight of each node;

[0011] The consultation server is further configured to perform information matching in the information technology service knowledge base according to the consultation keyword set, obtain a plurality of retrieval texts, and send the plurality of retrieval texts to the consultation client;

[0012] The consultation client is further configured to visually display the plurality of retrieval texts.

[0013] Based on the above disclosed content, after the present invention obtains the user consultation text based on the consultation client, it sends the user consultation text to the consultation server, and the consultation server can, by means of natural language processing technology, extract the consultation keyword set corresponding to the user consultation text, so as to perform result matching of information consultation based on the keyword set subsequently; specifically, the consultation server first constructs a word graph model based on the user consultation text, and then calculates the restart probability and edge weight of each node in the word graph model according to the user consultation text and the semantic features of the corresponding words of each node; then, the consultation server can perform weight optimization on each node based on the restart probability and edge weight of each node, so as to obtain the optimal weight of each node after the weight optimization; then, the consultation keyword set corresponding to the user consultation text can be extracted based on the optimal weight of each node; finally, information matching can be performed in the information technology service knowledge base according to the foregoing consultation keyword set, so as to obtain the consultation retrieval result corresponding to the user consultation text.

[0014] Through the above design, when the present invention extracts the keywords of the user consultation text, it introduces the user consultation text and the semantic information of each word in the text into the weight iterative optimization process of each node in the word graph model, that is, uses the user consultation text and the semantic information of each word in the text to allocate the restart probability and edge weight of each node. In this way, the word graph model can be combined with the semantic features of the text and words to extract keywords; based on this, the accuracy of keyword extraction can be improved, so as to ensure the accuracy of subsequent information consultation retrieval matching based on keywords; thus, compared with manual services, the present invention improves the information consultation efficiency, and compared with traditional intelligent consultation systems, greatly improves the accuracy and reliability of information consultation results. Therefore, it is very suitable for large-scale application and promotion in the field of information consultation services.

[0015] In a possible design, a consulting server is configured to perform word segmentation on the user's consultation text to obtain a set of segmented words, and construct the word graph model based on the set of segmented words, where each node in the word graph model is respectively used to represent a word in the set of segmented words;

[0016] The consulting server is configured to use the Word2vec model to generate word semantic vectors corresponding to each word in the set of segmented words, and use the Doc2vec model to generate a text semantic vector corresponding to the user's consultation text;

[0017] The consulting server is configured to calculate the similarity between the word semantic vector corresponding to each word and the text semantic vector, and determine the restart probability and edge weight of each node based on the similarity between the word semantic vector corresponding to each word and the text semantic vector;

[0018] The consulting server is configured to calculate the word feature information corresponding to the word of each node;

[0019] The consulting server is further configured to perform weight optimization processing on each node in the word graph model according to the restart probability, edge weight and word feature information of each node, so as to obtain the optimal weight of each node after the optimization processing.

[0020] In a possible design, the consulting server is configured to calculate the restart probability of any node by using the following formula (1);

[0021]

[0022] In the above formula (1), C(u i ) represents the restart probability of the any node, s(u i ,T) is used to represent the similarity between the word semantic vector corresponding to the word of the any node and the text semantic vector, s(u m ,T) is used to represent the similarity between the word semantic vector corresponding to the word of the m-th node and the text semantic vector, and M is the total number of nodes.

[0023] In a possible design, for any node, the consulting server is configured to filter out all nodes connected to the any node based on the word graph model, so as to form a target node set by using all the filtered nodes;

[0024] The consulting server is further configured to calculate the edge weights between the any node and each target node in the target node set by using the following formula (2), so as to use the edge weights between the any node and each target node as the edge weight of the any node;

[0025] e(u i ,u k ) = s(u i ,T) + s(u k ,T) (2)

[0026] In the above formula (2), e(u i ,u k ) represents the edge weight value between any one of the nodes and the k-th target node in the target node set, s(u i ,T) is used to represent the similarity between the word semantic vector of the word corresponding to any one of the nodes and the text semantic vector, and s(u k ,T) is used to represent the similarity between the word semantic vector of the word corresponding to the k-th target node and the text semantic vector, where k = 1, 2, 3,..., K, and K is the total number of target nodes.

[0027] In a possible design, the consulting server is used to calculate the word frequency feature, part-of-speech factor feature, word position feature, and word span feature of the word corresponding to each node, and use the word frequency feature, part-of-speech factor feature, word position feature, and word span feature of the word corresponding to each node to form the word feature information of the word corresponding to each node.

[0028] In a possible design, the consulting server is used to calculate the word span feature of the word corresponding to any one of the nodes according to the following formula (3);

[0029]

[0030] In the above formula (3), p(u i ) represents the word span feature of the word corresponding to any one of the nodes, last(u i ) represents the last occurrence position of the word corresponding to any one of the nodes in the user consultation text, first(u i ) represents the first occurrence position of the word corresponding to any one of the nodes in the user consultation text, and M is the total number of nodes.

[0031] In a possible design, the consulting server is used to obtain the target node set corresponding to each node, where the target node set of any one of the nodes contains all the nodes connected to the any one of the nodes in the word graph model;

[0032] The consulting server is used to initialize the iteration number t to 1 and obtain the weight of each target node in the target node set corresponding to each node at the (t - 1)-th iteration, where when t is 1, the weight of any one of the target nodes at the (t - 1)-th iteration is the initial weight of the any one of the target nodes;

[0033] A consulting server, which is used to calculate the weight of each node at the t-th iteration according to the word feature information of the words corresponding to each node, the restart probability of each node, the edge weight value of each node, and the weight of each target node in the target node set corresponding to each node at the (t - 1)-th iteration;

[0034] A consulting server, which is used to determine whether the iteration stop condition is satisfied based on the weight of each node at the t-th iteration;

[0035] If not, the consulting server is used to increment t by 1 and re-obtain the weight of each target node in the target node set corresponding to each node at the (t - 1)-th iteration until the iteration stop condition is satisfied, and then obtain the optimal weight of each node.

[0036] In a possible design, the word feature information corresponding to any node includes: the word frequency feature, the part-of-speech factor feature, the word position feature, and the word span feature of the word corresponding to the any node;

[0037] Among them, for any node, the consulting server is used to generate a weight control factor for the any node according to the word frequency feature, the part-of-speech factor feature, the word position feature, and the word span feature in the word feature information corresponding to the any node;

[0038] The consulting server is used to calculate the weight of the any node at the t-th iteration by using the following formula (4);

[0039]

[0040] In the above formula (4), Q t (u i ) represents the weight of the any node at the t-th iteration, OP(u i ) represents the weight control factor corresponding to the any node, β represents the damping coefficient, C(u i ) represents the restart probability of the any node, e(u i ,u k ) represents the any node and the k-th target node u k in the target node set corresponding to the any node, the edge weight value between them, G represents the target node set corresponding to the any node, out(u k ) represents the edge weight value of the k-th target node, Q t-1 (u k ) represents the weight of the k-th target node at the (t - 1)-th iteration.

[0041] In a possible design, a consultation server is configured to perform information matching processing in an information technology service knowledge base according to the set of consultation keywords, so as to obtain a number of initial retrieval texts after the information matching processing;

[0042] The consultation server is configured to sort the optimal weights corresponding to each consultation keyword in the set of consultation keywords in descending order, so as to generate a user consultation text feature vector after the descending order sorting;

[0043] The consultation server is configured to generate a retrieval text feature vector corresponding to each initial retrieval text, wherein the retrieval text feature vector of any one initial retrieval text includes the optimal weights of each keyword corresponding to the any one initial retrieval text, and the arrangement of the optimal weights of each keyword corresponding to the any one initial retrieval text is in descending order;

[0044] The consultation server is configured to calculate the information matching degree between each initial retrieval text and the user consultation text based on the retrieval text feature vectors corresponding to each initial retrieval text and the user consultation text feature vector;

[0045] The consultation server is further configured to determine a number of retrieval texts from the number of initial retrieval texts according to the information matching degree between each initial retrieval text and the user consultation text, and send the number of retrieval texts to the consultation client.

[0046] In a possible design, for any one initial retrieval text, the consultation server is configured to calculate a first matching degree between the any one initial retrieval text and the user consultation text by using the following formula (5);

[0047]

[0048] In the above formula (5), D1 represents the first matching degree between the any one initial retrieval text and the user consultation text, w z represents the z-th optimal weight in the user consultation text feature vector, w hz represents the z-th optimal weight in the retrieval text feature vector corresponding to the any one initial retrieval text, and Z represents the length of the user consultation text feature vector;

[0049] The consultation server is configured to calculate a second matching degree between the any one initial retrieval text and the user consultation text by using the following formula (6);

[0050]

[0051] In the above formula (6), D2 represents the second matching degree between the any one initial retrieval text and the user consultation text;

[0052] The consultation server is further configured to calculate an information matching degree between any one of the initial retrieval texts and the user consultation text according to the first matching degree and the second matching degree.

[0053] Beneficial effects:

[0054] (1) When extracting the keywords of the user consultation text, the present invention introduces the semantic information of the user consultation text and each word in the text into the weight iterative optimization process of each node in the word graph model, that is, uses the semantic information of the user consultation text and each word in the text to allocate the restart probability and edge weight of each node. In this way, the word graph model can be combined with the semantic features of the text and words to extract keywords; based on this, the accuracy of keyword extraction can be improved, thereby ensuring the accuracy of subsequent information consultation retrieval matching based on keywords; thus, compared with manual services, the present invention improves the information consultation efficiency, and compared with traditional intelligent consultation systems, greatly improves the accuracy and reliability of information consultation results. Therefore, it is very suitable for large-scale application and promotion in the field of information consultation services. Description of the Drawings

[0055] Figure 1 It is a system architecture diagram of an information technology service consultation platform based on intelligent software provided by an embodiment of the present invention. Detailed Embodiments

[0056] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the present invention in combination with the drawings and the descriptions of the embodiments or the prior art. Obviously, the following descriptions of the structures of the drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained according to these drawings without creative efforts. It should be noted here that the descriptions of these embodiment modes are used to help understand the present invention, but do not constitute a limitation to the present invention.

[0057] It should be understood that although terms such as first and second may be used herein to describe various units, these units should not be limited by these terms. These terms are only used to distinguish one unit from another. For example, the first unit can be called the second unit, and similarly, the second unit can be called the first unit, without departing from the scope of the exemplary embodiments of the present invention.

[0058] It should be understood that for the term "and / or" that may appear in this text, it is merely a description of the association relationship between associated objects, indicating that three relationships may exist. For example, A and / or B may represent: A exists alone, B exists alone, and both A and B exist simultaneously. For the term " / and" that may appear in this text, it describes another association object relationship, indicating that two relationships may exist. For example, A / and B may represent: A exists alone, and both A and B exist. Additionally, for the character " / " that may appear in this text, it generally indicates that the associated objects before and after are in an "or" relationship.

[0059] Embodiment:

[0060] Refer to Figure 1 As shown, the information technology service consulting platform based on intelligent software provided in this embodiment may, but is not limited to, include: a consulting client (N may be set) and a consulting server. Among them, the consulting client provides an information technology service consulting interface for the user to input the required user consulting text. Of course, it also provides voice input and, after receiving the voice information, converts the voice information into the user consulting text. The consulting server, on the other hand, provides a consulting retrieval service. Specifically, it stores an information technology service knowledge base, which contains data in different information technology service fields. At the same time, the consulting server is also used to extract keywords from the user consulting text by using natural language processing technology (specifically, an improved graph sorting algorithm), and then use the extracted keywords to perform information matching in the aforementioned information technology service knowledge base to obtain a consulting result. Finally, the consulting server can send the aforementioned consulting result to the consulting client for visual display.

[0061] Among them, the following discloses the specific working processes of the aforementioned consulting client and consulting server:

[0062] In specific implementation, the consulting client is used to obtain the user consulting text and send the user consulting text to the consulting server. The consulting server, based on the user consulting text, can construct a word graph model. Specifically, the consulting server is used to perform word segmentation processing on the user consulting text to obtain a word segmentation set. Then, based on the word segmentation set, the word graph model can be constructed. In this embodiment, the word segmentation processing may, but is not limited to, first perform stop word removal processing on the user consulting text, and then perform sentence segmentation. Finally, a word segmentation tool (such as the jieba word segmentation tool) can be used to perform word segmentation on the segmented sentences to obtain the word segmentation set.

[0063] Further, the word graph model is one way to convert text into a graph model. That is, the words obtained by segmenting the text are used as nodes (i.e., each node is used to represent a word in the user's consultation text, that is, a word in the word segmentation word set). Then, based on the co-occurrence relationship between each node, edges are connected (edges are used to connect nodes with co-occurrence relationships). In this way, by iteratively scoring each node in the word graph model, words with high importance can be screened out, thereby completing the extraction of keywords.

[0064] In actual application, to improve the accuracy of keyword extraction, in this embodiment, when iteratively scoring each node in the word graph model, the semantic information of the user's consultation text and the corresponding words of each node is introduced, that is, the semantic information of the user's consultation text and each word in the text is used to allocate the restart probability and edge weight of each node, so that in the subsequent node iterative scoring, by adding semantic information, the accuracy of keyword extraction can be improved.

[0065] Specifically, the consultation server is used to calculate the restart probability and edge weight of each node in the word graph model according to the semantic features of the user's consultation text and the corresponding words of each node; then, based on the restart probability and edge weight of each node, weight optimization processing is performed on each node in the word graph model to obtain the optimal weight of each node after the optimization processing; among them, the higher the weight of a node, the more important the word corresponding to the node in the user's consultation text. Therefore, by using the improved graph sorting algorithm to update the node weights, the most important words in the user's consultation text can be screened out; specifically, the above-mentioned allocation process of the restart probability and edge weight based on semantics, and the weight optimization processing process will be elaborated in detail in the following embodiments.

[0066] After obtaining the optimal weights of each node, the consultation server can be used to determine the consultation keyword set corresponding to the user's consultation text according to the optimal weight of each node; in this embodiment, for example, the words corresponding to each node can be sorted in descending order of the optimal weight, and then the top 3 or 5 words can be selected as the consultation keyword set.

[0067] After obtaining the consultation keyword set, information matching can be performed, that is: the consultation server is also used to perform information matching in the information technology service knowledge base according to the consultation keyword set to obtain several retrieval texts, and send the several retrieval texts to the consultation client, and the consultation client can be used to visually display the several retrieval texts, thereby completing the information technology consultation service for the user; in this embodiment, using keywords for information retrieval is a common technique for information query, and its principle will not be elaborated here.

[0068] Thus, through the foregoing description, this embodiment utilizes the user consultation text and the semantic information of each word in the text to allocate the restart probability and edge weights of each node in the word graph model. In this way, the word graph model can be combined with the semantic features of the text and words to extract keywords; based on this, the accuracy of keyword extraction can be improved, thereby ensuring the accuracy of subsequent information consultation retrieval matching based on keywords.

[0069] In a possible design, this embodiment provides the foregoing consultation server to perform the specific process of optimizing the weight of each node in the word graph model as follows:

[0070] Among them, the consultation server is used to generate the word semantic vectors corresponding to each word in the word segmentation word set by using the Word2vec model, and is used to generate the text semantic vector corresponding to the user consultation text by using the Doc2vec model; among them, using the Doc2vec model to vectorize the user consultation text not only obtains the semantic information of the text, but more importantly, also extracts the word order information of the text. In this way, through the word order information of the text, the component constraints between sentences can be represented. Based on this, considering the word order information, the semantic information of the text can be mined at a deeper level; at the same time, by using the Word2vec model to map each word to a semantic vector, it is convenient to allocate the edge weights and restart probabilities of each node in the word graph model according to the semantic similarity between the word and the text.

[0071] Furthermore, in this embodiment, for example, the dimensions of the word semantic vectors of each word are the same as those of the text semantic vector, both of which are 512. Therefore, ensuring that the dimensions of the vectors are the same is convenient for subsequent calculation of semantic similarity; among them, the semantic similarity calculation process is as follows: the consultation server is used to calculate the similarity between the word semantic vector corresponding to each word and the text semantic vector, and based on the similarity between the word semantic vector corresponding to each word and the text semantic vector, determine the restart probability and edge weight of each node; in this embodiment, for example, the similarity between any word semantic vector and the text semantic vector is represented by the cosine distance.

[0072] In specific implementation, taking any node as an example, the calculation process of its restart probability is described. That is, for example, the consultation server can but is not limited to use the following formula (1) to calculate the restart probability of the foregoing any node.

[0073]

[0074] In the above formula (1), C(u i ) represents the restart probability of the any node, s(u i, T) is used to represent the similarity between the word semantic vector corresponding to any of the nodes and the text semantic vector, s(u m , T) is used to represent the similarity between the word semantic vector corresponding to the m-th node and the text semantic vector, and M is the total number of nodes.

[0075] As can be seen from the above formula, the semantic similarity between the node and the user's consultation text is used as the weight of the restart probability, that is, the higher the semantic similarity between the word corresponding to the node and the user's consultation text, the greater its restart probability, and vice versa. Thus, in the random walk process, the words with higher relevance to the text theme are more likely to be selected as keywords; based on this, introducing semantic information into the restart probability can improve the accuracy of keyword extraction.

[0076] After calculating the restart probability of each node, the edge weight value of each node can be calculated, and the calculation process is as follows:

[0077] For any node, the consultation server first filters out all the nodes connected to the any node based on the word graph model to form a target node set by using all the filtered nodes; then, the following formula (2) is used to calculate the edge weight value between the any node and each target node in the target node set, so as to use the edge weight value between the any node and each target node as the edge weight value of the any node; where formula (2) is:

[0078] e(u i , u k ) = s(u i , T) + s(u k , T) (2)

[0079] In the above formula (2), e(u i , u k ) represents the edge weight value between the any node and the k-th target node in the target node set, s(u i , T) is used to represent the similarity between the word semantic vector corresponding to the any node and the text semantic vector, s(u k , T) is used to represent the similarity between the word semantic vector corresponding to the k-th target node and the text semantic vector, where k = 1, 2, 3,..., K, and K is the total number of target nodes.

[0080] In this embodiment, in the word graph model, any node has nodes that it points to, that is, the nodes connected to this any node. Therefore, the weight of the edge between this any node and the node connected to this any node (i.e., the aforementioned target node) can be calculated. Then, based on this, the edge weight value of this any node is formed. That is, assuming any node is A, and the nodes connected to A are B and C, then the edge weight value of this any node includes the edge weight value between node A and node B, and the edge weight value between node A and node C.

[0081] Meanwhile, the calculation formula for the weight of the edge between this any node and the aforementioned target node can refer to the aforementioned formula (2), that is, calculate the similarity between the word semantic vector of the word corresponding to this any node and the text semantic vector, and the similarity between the word semantic vector of the word corresponding to the target node and the text semantic vector, and then sum them to obtain the edge weight value between this any node and the target node; thus, as can be seen from the aforementioned formula (2), when the sum of the similarities between any node and the target node and the user consultation text is higher, that is, the semantic relationship between this word and the document is higher, then the proportion of this word in the document is larger.

[0082] Through the aforementioned formula (1), after introducing the semantic information of the text and the words into the restart probability and edge weight value allocation of each node in the word graph model, the word feature information of the words can be extracted, so as to subsequently combine the aforementioned restart probability and edge weight value to perform weight optimization for each node, that is: the consultation server is used to calculate the word feature information corresponding to each node, so as to perform weight optimization processing on each node in the word graph model according to the restart probability, edge weight value and word feature information of each node, so as to obtain the optimal weight of each node after the optimization processing.

[0083] In this embodiment, in the traditional graph sorting algorithm, usually only the co-occurrence relationship between words is considered to affect the weight score. In fact, in addition to the word co-occurrence relationship, there are several factors that will affect the importance score of words, such as word frequency, word position, part of speech, and word span. Therefore, in this embodiment, the aforementioned factors are quantified to form word feature information and introduced into the graph sorting algorithm; optionally, taking the consultation server as an example, it can be used but not limited to calculate the word frequency feature, part of speech factor feature, word position feature, and word span feature corresponding to each node; then, use the word frequency feature, part of speech factor feature, word position feature, and word span feature corresponding to each node to form the word feature information corresponding to each node.

[0084] Further, taking the word corresponding to any node as an example, the calculation process of the aforementioned four features is elaborated. Among them, for the word frequency feature, for example, it can be calculated but not limited to using the following formula (7).

[0085]

[0086] In the above formula (7), F(u i ) represents the word frequency feature of the word corresponding to any one of the nodes, and f(u i ) represents the occurrence frequency of the word corresponding to any one of the nodes in the user's consultation text.

[0087] Similarly, for the word position feature, it is determined whether the word corresponding to any one of the nodes is located in the title or the first position of the user's consultation text. If so, the word position feature of the word corresponding to any one of the nodes is set to 0.5; otherwise, it is set to 0.

[0088] For the part-of-speech factor feature, if the word corresponding to any one of the nodes is a noun, the part-of-speech factor feature of the word corresponding to any one of the nodes is set to 0.5; if the word corresponding to any one of the nodes is a verb, the part-of-speech factor feature of the word corresponding to any one of the nodes is set to 0.4; and if the part-of-speech of the word corresponding to any one of the nodes is a part-of-speech other than a verb and a noun, then the part-of-speech factor feature of the word corresponding to any one of the nodes is set to 0.1.

[0089] Finally, for the word span feature, for example, but not limited to, the following formula (3) can be used to calculate and obtain it.

[0090]

[0091] In the above formula (3), p(u i ) represents the word span feature of the word corresponding to any one of the nodes, last(u i ) represents the last occurrence position of the word corresponding to any one of the nodes in the user's consultation text, first(u i ) represents the first occurrence position of the word corresponding to any one of the nodes in the user's consultation text, and M is the total number of nodes.

[0092] Thus, through the foregoing description, the word feature information corresponding to each node can be calculated; then, in combination with the restart probability and edge weight value of each foregoing node, the weight optimization processing of each node can be performed.

[0093] Specifically, the detailed process of the consultation server performing weight optimization processing is as follows:

[0094] The consultation server is first used to obtain the target node set corresponding to each node. Among them, the target node set of any one node includes all the nodes connected to the any one node in the word graph model; in this embodiment, the determination process of the target node set of each node has been described above and will not be elaborated here.

[0095] After obtaining the set of target nodes corresponding to each node, consult the server, which is used to initialize the iteration count t to 1 and obtain the weights of each target node in the set of target nodes corresponding to each node at the (t - 1)-th iteration; in specific applications, it is equivalent to using the weights of each node in the previous iteration to calculate the weights in the current iteration. Among them, when t is 1, the weight of any target node at the (t - 1)-th iteration is the initial weight of the any target node; at the same time, the initial weights of each node can be, but are not limited to, preset settings, such as being set to 1 or 0.5.

[0096] After obtaining the weights of each target node corresponding to each node at the previous iteration, the weight update can be performed by combining the foregoing word feature information, restart probability, and edge weights, that is: consult the server, which is used to calculate the weights of each node at the t-th iteration according to the word feature information of the words corresponding to each node, the restart probability of each node, the edge weights of each node, and the weights of each target node in the set of target nodes corresponding to each node at the (t - 1)-th iteration.

[0097] In specific implementation, taking any node as an example, the specific process of weight update is described as follows:

[0098] For any node, the server is first used to generate a weight control factor for the any node according to the word frequency feature, part-of-speech factor feature, word position feature, and word span feature in the word feature information corresponding to the any node (in this embodiment, the weighted sum of the foregoing four features is used to calculate the weight control factor of the any node); then, the following formula (4) is used to calculate the weight of the any node at the t-th iteration.

[0099]

[0100] In the above formula (4), Q t (u i ) represents the weight of the any node at the t-th iteration, OP(u i ) represents the weight control factor corresponding to the any node, β represents the damping coefficient, C(u i ) represents the restart probability of the any node, e(u i , u k ) represents the any node and the k-th target node u k in the set of target nodes corresponding to the any node, G represents the set of target nodes corresponding to the any node, out(u k ) represents the edge weight of the k-th target node, Q t-1 (u k) represents the weight of the k-th target node at the (t - 1)-th iteration; in this embodiment, the edge weight value of the k-th target node is the sum of the edge weight values between the k-th target node and all the nodes connected to the k-th target node. The calculation process of the edge weight value between two nodes can be referred to the foregoing formula (2), which will not be elaborated herein.

[0101] Thus, after calculating the weights of each node at the t-th iteration based on the foregoing formula (4), it can be determined whether to end the iteration, that is, the consulting server is used to determine whether the iteration stop condition is satisfied based on the weights of each node at the t-th iteration. If not, the consulting server is used to increment t by 1 and re-obtain the weights of each target node in the target node set corresponding to each node at the (t - 1)-th iteration until the iteration stop condition is satisfied, and the optimal weights of each node are obtained; in this embodiment, the iteration stop condition is that the weight change values of all nodes are less than a preset threshold.

[0102] Thus, through the foregoing elaboration, the semantic information of the user consulting text and words can be introduced into the weight update process of each node in the word graph model, so as to combine the word graph model with the semantic features of the text and words to extract keywords; based on this, the accuracy of keyword extraction can be improved, thereby ensuring the accuracy of subsequent information consulting retrieval matching based on keywords.

[0103] In a possible design, this embodiment provides an information retrieval method, which can increase the matching degree between the retrieval result and the user consulting text, that is, increase the relevance between the result and the input. The information retrieval process is as follows:

[0104] In specific implementation, the consulting server is first used to perform information matching processing in the information technology service knowledge base according to the consulting keyword set, so as to obtain several initial retrieval texts after the information matching processing; then, the optimal weights corresponding to each consulting keyword in the consulting keyword set are sorted in descending order to generate a user consulting text feature vector after the descending order arrangement.

[0105] Next, the consulting server is used to generate a retrieval text feature vector corresponding to each initial retrieval text. The retrieval text feature vector of any initial retrieval text includes the optimal weights of each keyword corresponding to the any initial retrieval text, and the arrangement of the optimal weights of each keyword corresponding to the any initial retrieval text is in descending order; in this embodiment, the keyword extraction process corresponding to any initial retrieval text is the same as that of the foregoing user consulting text. Therefore, when extracting keywords, the optimal weights of each keyword can also be obtained, and then the optimal weights of each keyword are sorted in descending order, and the retrieval text feature vector corresponding to the any initial retrieval text can be obtained.

[0106] Then, the consulting server is used to calculate the information matching degree between each initial retrieval text and the user consulting text based on the retrieval text feature vector corresponding to each initial retrieval text and the user consulting text feature vector. Among them, the higher the information matching degree, the more relevant the retrieval result is to the user input user consulting text. Therefore, by constructing the feature vector of the user consulting text and the feature vector of the retrieval text to calculate the matching degree of the retrieval result, more matching retrieval results can be screened out, and the accuracy of the retrieval can be improved.

[0107] In specific implementation, the following provides a calculation method for the information matching degree. Taking any initial retrieval text as an example, the specific calculation process is as follows:

[0108] For any initial retrieval text, the consulting server is used to calculate the first matching degree between the any initial retrieval text and the user consulting text by using the following formula (5).

[0109]

[0110] In the above formula (5), D1 represents the first matching degree between the any initial retrieval text and the user consulting text, w z represents the z-th optimal weight in the user consulting text feature vector, w hz represents the z-th optimal weight in the retrieval text feature vector corresponding to the any initial retrieval text, and Z represents the length of the user consulting text feature vector.

[0111] Then, the consulting server is used to calculate the second matching degree between the any initial retrieval text and the user consulting text by using the following formula (6).

[0112]

[0113] In the above formula (6), D2 represents the second matching degree between the any initial retrieval text and the user consulting text.

[0114] After calculating the matching degree between the any initial retrieval text and the user consulting text based on the foregoing formula (5) and formula (6), the information matching degree between the any initial retrieval text and the user consulting text can be obtained based on this, that is: it is also used to calculate the information matching degree between the any initial retrieval text and the user consulting text according to the first matching degree and the second matching degree. In this embodiment, for example, but not limited to, the weighted summation method can be used to calculate the information matching degree of the two based on the first matching degree and the second matching degree.

[0115] Thus, after calculating the information matching degrees between each initial retrieval text and the user consultation text, the consultation server can be used to determine multiple retrieval texts from several initial retrieval texts according to the information matching degrees between each initial retrieval text and the user consultation text, and send the multiple retrieval texts to the consultation client; in specific applications, the initial retrieval texts can be sorted in descending order of the information matching degrees, and then the top 3, 5 or 8 initial retrieval texts before sorting are selected as the final retrieval texts and visualized.

[0116] Through the above description, after using the consultation keywords to match the initial retrieval results, this embodiment also constructs a retrieval text feature vector through the weights of the keywords in the initial retrieval results; then, calculates the information matching degrees between each retrieval text feature vector and the feature vector of the user consultation text to perform a secondary screening of the initial retrieval results. Based on this, the output retrieval results can be made more in line with the user's consultation intention, thereby further increasing the accuracy of information consultation.

[0117] Through the above detailed description of the information technology service consultation platform based on intelligent software, when extracting the keywords of the user consultation text, this invention introduces the semantic information of the user consultation text and each word in the text into the weight iterative optimization process of each node in the word graph model, that is, uses the semantic information of the user consultation text and each word in the text to assign the restart probability and edge weight of each node. In this way, the word graph model can be combined with the semantic features of the text and words to extract keywords; based on this, the accuracy of keyword extraction can be improved, thereby ensuring the accuracy of subsequent information consultation retrieval matching based on keywords; thus, this invention improves the information consultation efficiency compared with manual services, and significantly improves the accuracy and reliability of information consultation results compared with traditional intelligent consultation systems; at the same time, constructs a feature vector through the weights of the keywords of the user consultation text and the initial retrieval text, and calculates the matching degree between the result and the input text based on the feature vector; then, performs a secondary screening of the initial retrieval results through the matching degree to obtain the final retrieval text. Based on this, the retrieval output can be made more in line with the user's consultation intention, and therefore, the accuracy of information consultation is further improved, and thus it is very suitable for large-scale application and promotion in the field of information consultation services.

[0118] In a possible design, the second aspect of this embodiment provides a working method for the information technology service consultation platform described in the first aspect of the embodiment, and its execution process can be but is not limited to the following steps S1 to S6.

[0119] The consultation client obtains the user consultation text and sends the user consultation text to the consultation server;

[0120] The consultation server constructs a word graph model based on the user's consultation text. In the word graph model, each node is used to represent a word in the user's consultation text, and the nodes with co-occurrence relationships are connected by edges;

[0121] The consultation server calculates the restart probability and edge weight of each node in the word graph model according to the user's consultation text and the semantic features of the words corresponding to each node, and performs weight optimization processing on each node in the word graph model based on the restart probability and edge weight of each node, so as to obtain the optimal weight of each node after the optimization processing;

[0122] The consultation server determines a consultation keyword set corresponding to the user's consultation text according to the optimal weight of each node;

[0123] The consultation server performs information matching in the information technology service knowledge base according to the consultation keyword set, obtains a number of retrieval texts, and sends the number of retrieval texts to the consultation client;

[0124] The consultation client visually displays the number of retrieval texts.

[0125] For the working process, working details and technical effects of this embodiment, reference can be made to the first aspect of the embodiment, which will not be elaborated here.

[0126] The third aspect of this embodiment provides an electronic device, including: a memory, a processor and a transceiver that are communicatively connected in sequence. Among them, the memory is used to store computer programs, the transceiver is used to send and receive messages, and the processor is used to read the computer programs and execute the working method of the information technology service consultation platform as described in the second aspect of the embodiment.

[0127] Specifically, the memory may include, but is not limited to, random access memory (RAM), read only memory (ROM), flash memory, first input first output (FIFO), and / or first in last out (FILO), etc.; specifically, the processor may include one or more processing cores, such as a 4-core processor, an 8-core processor, etc. The processor may be implemented in at least one hardware form of DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), or PLA (Programmable Logic Array). At the same time, the processor may also include a main processor and a coprocessor. The main processor is a processor used to process data in the wake state, also known as the CPU (Central Processing Unit); the coprocessor is a low-power processor used to process data in the standby state.

[0128] In some embodiments, the processor may be integrated with a GPU (Graphics Processing Unit), and the GPU is responsible for rendering and drawing the content to be displayed on the display screen. For example, the processor may be, but is not limited to, a microprocessor of the STM32F105 series, a reduced instruction set computer (RISC) microprocessor, a processor with an X86 architecture, or a processor integrated with an embedded neural-network processing unit (NPU); the transceiver may be, but is not limited to, a Wi-Fi wireless transceiver, a Bluetooth wireless transceiver, a General Packet Radio Service (GPRS) wireless transceiver, a ZigBee (low-power local area network protocol based on the IEEE802.15.4 standard) wireless transceiver, a 3G transceiver, a 4G transceiver, and / or a 5G transceiver, etc. In addition, the device may also include, but is not limited to, a power module, a display screen, and other necessary components.

[0129] For the working process, working details, and technical effects of the electronic device provided in this embodiment, reference may be made to the first aspect of the embodiment, which will not be elaborated herein.

[0130] In the fourth aspect of this embodiment, there is provided a storage medium storing instructions including the working method of the information technology service consulting platform described in the second aspect of the embodiment, that is, instructions are stored on the storage medium, and when the instructions run on a computer, they execute the working method of the information technology service consulting platform described in the second aspect of the embodiment.

[0131] Among them, the storage medium refers to a carrier for storing data, and may include, but is not limited to, floppy disks, optical discs, hard disks, flash memories, USB flash drives, and / or memory sticks, etc. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices.

[0132] For the working process, working details, and technical effects of the storage medium provided in this embodiment, reference may be made to the first aspect of the embodiment, which will not be elaborated herein.

[0133] In the fifth aspect of this embodiment, there is provided a computer program product containing instructions, which, when running on a computer, causes the computer to execute the working method of the information technology service consulting platform described in the second aspect of the embodiment. Among them, the computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices.

[0134] Finally, it should be noted that the above are only the preferred embodiments of the present invention and are not used to limit the protection scope of the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. An information technology service consulting platform based on intelligent software, characterized in that: include: A consultation client is used to obtain a user consultation text and send the user consultation text to a consultation server; The consultation server is used to construct a word graph model based on the user consultation text, wherein each node in the word graph model is used to represent a word in the user consultation text, and the nodes with co-occurrence relationship are connected by edges; The consulting server is used to calculate the restart probability and edge weight of each node in the word graph model according to the user consulting text and the semantic features of the words corresponding to each node, and perform weight optimization processing on each node in the word graph model based on the restart probability and edge weight of each node, so as to obtain the optimal weight of each node after the optimization processing; The consulting server is used to determine the consulting keyword set corresponding to the user's consulting text according to the optimal weight of each node; The consulting server is further used to match information in the information technology service knowledge base according to the consulting keyword set, obtain a number of search texts, and send the number of search texts to the consulting client; The consulting client is also used to visually display the plurality of search texts.

2. The information technology service consulting platform based on intelligent software according to claim 1, characterized in that: The consultation server is used to perform word segmentation processing on the user consultation text to obtain a word segmentation word set, and construct the word graph model based on the word segmentation word set, wherein each node in the word graph model is used to represent a word in the word segmentation word set; The consulting server is used to generate a word semantic vector corresponding to each word in the word segmentation word set by using a Word2vec model, and is used to generate a text semantic vector corresponding to the user consulting text by using a Doc2vec model; The consulting server is used to calculate the similarity between the word semantic vector corresponding to each word and the text semantic vector, and determine the restart probability and edge weight of each node based on the similarity between the word semantic vector corresponding to each word and the text semantic vector; The consulting server is used to calculate the word feature information of the words corresponding to each node; The consulting server is also used to perform weight optimization processing on each node in the word graph model according to the restart probability, edge weight and word feature information of each node, so as to obtain the optimal weight of each node after the optimization processing.

3. The information technology service consulting platform based on intelligent software according to claim 2 is characterized in that: The consulting server is used to calculate the restart probability of any node using the following formula (1); In the above formula (1), C(u i ) represents the restart probability of any node, s(u i ,T) is used to represent the similarity between the word semantic vector of the word corresponding to any node and the text semantic vector, s(u m ,T) is used to represent the similarity between the word semantic vector of the word corresponding to the mth node and the text semantic vector, and M is the total number of nodes.

4. The information technology service consulting platform based on intelligent software according to claim 2 is characterized in that: For any node, the consulting server is used to filter out all nodes connected to the any node based on the word graph model, so as to form a target node set using all the filtered nodes; The consulting server is further used to calculate the edge weight between the any node and each target node in the target node set by using the following formula (2), so as to use the edge weight between the any node and each target node as the edge weight of the any node; e(u i ,u k )=s(u i ,T)+s(u k ,T) (2) In the above formula (2), e(u i ,u k ) represents the edge weight between any node and the kth target node in the target node set, s(u i ,T) is used to represent the similarity between the word semantic vector of the word corresponding to any node and the text semantic vector, s(u k ,T) is used to represent the similarity between the word semantic vector of the word corresponding to the k-th target node and the text semantic vector, where k=1,2,3,...,K, and K is the total number of target nodes.

5. The information technology service consulting platform based on intelligent software according to claim 2 is characterized in that: The consulting server is used to calculate the word frequency characteristics, part-of-speech factor characteristics, word position characteristics and word span characteristics of the words corresponding to each node, and use the word frequency characteristics, part-of-speech factor characteristics, word position characteristics and word span characteristics of the words corresponding to each node to form the word feature information of the words corresponding to each node.

6. The information technology service consulting platform based on intelligent software according to claim 5, characterized in that: The consulting server is used to calculate the word span feature of the word corresponding to any node according to the following formula (3); In the above formula (3), p(u i ) represents the word span feature of any node corresponding to the word, last(u i ) indicates the last position of the word corresponding to any node in the user's consultation text, first(u i ) represents the position where the word corresponding to any node first appears in the user consultation text, and M is the total number of nodes.

7. The information technology service consulting platform based on intelligent software according to claim 1, characterized in that: A consulting server is used to obtain a target node set corresponding to each node, wherein the target node set of any node includes all nodes connected to the any node in the word graph model; The consulting server is used to initialize the number of iterations t to 1, and obtain the weight of each target node in the target node set corresponding to each node at the t-1th iteration, wherein when t is 1, the weight of any target node at the t-1th iteration is the initial weight of any target node; The consulting server is used to calculate the weight of each node at the tth iteration according to the word feature information of the words corresponding to each node, the restart probability of each node, the edge weight of each node, and the weight of each target node in the target node set corresponding to each node at the t-1th iteration; The consulting server is used to determine whether the iteration stop condition is met based on the weight of each node at the tth iteration; If not, the consulting server is used to add 1 to t and re-obtain the weight of each target node in the target node set corresponding to each node at the t-1th iteration until the iteration stop condition is met to obtain the optimal weight of each node.

8. The information technology service consulting platform based on intelligent software according to claim 7 is characterized in that: The word feature information of a word corresponding to any node includes: word frequency feature, part-of-speech factor feature, word position feature and word span feature of the word corresponding to any node; Wherein, for any node, the consulting server is used to generate a weight control factor of any node according to the word frequency feature, part-of-speech factor feature, word position feature and word span feature in the word feature information corresponding to any node; The consulting server is used to calculate the weight of any node at the tth iteration by using the following formula (4); In the above formula (4), Q t (u i ) represents the weight of any node at the tth iteration, OP(u i ) represents the weight control factor corresponding to any node, β represents the damping coefficient, C(u i ) represents the restart probability of any node, e(u i ,u k ) represents any node, and the kth target node u in the target node set corresponding to any node k The edge weight between them, G represents the target node set corresponding to any node, out(u k ) represents the edge weight of the kth target node, Q t-1 (u k ) represents the weight of the k-th target node at the t-1th iteration.

9. The information technology service consulting platform based on intelligent software according to claim 1, characterized in that: A consulting server, used for performing information matching processing in an information technology service knowledge base according to the consulting keyword set, so as to obtain a plurality of initial search texts after the information matching processing; The consultation server is used to arrange the optimal weights corresponding to the consultation keywords in the consultation keyword set in descending order, so as to generate a user consultation text feature vector after arranging them in descending order; The consulting server is used to generate a search text feature vector corresponding to each initial search text, wherein the search text feature vector of any initial search text contains the optimal weight of each keyword corresponding to the initial search text, and the optimal weight of each keyword corresponding to the initial search text is arranged in descending order; The consulting server is used to calculate the information matching degree between each initial search text and the user consulting text based on the search text feature vector corresponding to each initial search text and the user consulting text feature vector; The consulting server is further used to determine multiple search texts from a number of initial search texts according to the information matching degree between each initial search text and the user consultation text, and send the multiple search texts to the consulting client.

10. The information technology service consulting platform based on intelligent software according to claim 9, characterized in that: For any initial search text, the consulting server is used to calculate a first matching degree between the any initial search text and the user consulting text by using the following formula (5); In the above formula (5), D1 represents the first matching degree between any of the initial search texts and the user consultation text, w z represents the zth optimal weight in the user consultation text feature vector, w hz represents the zth optimal weight in the retrieval text feature vector corresponding to any of the initial retrieval texts, and Z represents the length of the user consultation text feature vector; The consulting server is used to calculate a second matching degree between any one of the initial search texts and the user consulting text using the following formula (6); In the above formula (6), D2 represents the second matching degree between any of the initial search texts and the user consultation text; The consulting server is further used to calculate the information matching degree between any one of the initial search texts and the user consulting text according to the first matching degree and the second matching degree.

Images