Information technology consultation service method and system based on correlation analysis

Abstract

The invention discloses an information technology consultation service method and system based on correlation analysis, and relates to the technical field of information retrieval consultation, and the method comprises the steps: extracting keywords in a user consultation request, and recognizing user demands in the keywords through a user demand model; processing the user demand, executing parameterized retrieval, performing correlation analysis on a retrieval result of the parameterized retrieval, and generating a correlation list based on a result of the correlation analysis; analyzing each entry in the correlation list, and outputting an answer set meeting the user demand; and integrating the answer set to form an interactive question and answer interface for a user to view. According to the method, accurate understanding of complex user query is realized, the correlation and accuracy of retrieval results are improved, and meanwhile, the adaptability and knowledge coverage range of the system are enhanced through dynamic weight adjustment and knowledge graph integration.

Description

Technical Field

[0001] The present invention relates to the technical field of information retrieval and consultation, and in particular to an information technology consultation service method and system based on correlation analysis. Background Art

[0002] The field of information technology consulting services has experienced significant technological advances in recent years, especially in natural language processing and information retrieval. Early consulting services mainly relied on simple keyword matching and basic information retrieval techniques, such as Boolean search and vector space model. As technology developed, more complex algorithms were introduced, such as latent semantic analysis (LSA) and probabilistic topic models (such as LDA). These methods have improved the accuracy and relevance of retrieval to a certain extent. In recent years, the application of deep learning techniques, such as recurrent neural networks (RNN) and transformer models (such as BERT), has further promoted the development of natural language understanding and information retrieval, enabling the system to better understand the semantics and context of user queries.

[0003] However, despite these advances, existing information technology consulting service systems still have some significant limitations. First, most systems perform poorly when handling complex, multimodal user queries and have difficulty integrating text and voice input. Second, existing systems often use static retrieval models, which are difficult to adapt to users' personalized needs and professional terminology in different fields. Third, when building user demand models, most systems fail to make full use of linguistic features and semantic structure information, resulting in an insufficient understanding of user intent. In addition, existing correlation analysis methods usually only consider limited dimensions, such as TF-IDF or cosine similarity, ignoring important factors such as the timeliness of documents and user interaction history. Finally, in the answer generation stage, many systems still rely on simple extractive methods and cannot provide sufficiently detailed and customized answers. Summary of the invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides an information technology consulting service method and system based on correlation analysis to solve the problems that the existing information technology consulting service system has a shallow understanding of user needs, lacks contextual awareness during information retrieval, and has difficulties in integrating text and voice input.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In the first aspect, an embodiment of the present invention provides an information technology consulting service method based on correlation analysis, which includes extracting keywords in user consulting requests, identifying user needs in the keywords through a user need model; processing the user needs, performing parameterized retrieval, performing correlation analysis on the retrieval results of the parameterized retrieval, and generating a correlation list based on the results of the correlation analysis; analyzing each entry in the correlation list and outputting a set of answers that meet the user needs; integrating the answer set to form an interactive question-and-answer interface for users to view.

[0008] As a preferred solution of the information technology consulting service method based on correlation analysis described in the present invention, the types of user consulting requests include text form, voice form and text-speech fusion form.

[0009] As a preferred solution of the information technology consulting service method based on correlation analysis described in the present invention, wherein: extracting keywords in user consultation requests includes the following steps: preprocessing the user consultation requests; the preprocessing includes: if the user consultation request is in voice form, converting it into text form; if the user consultation request is in text-speech fusion form, separating it into a text part and a voice part, converting the voice part into text form, adding voice tags, and merging it with the text part; all types of user consultation requests are uniformly converted into a standardized text format, and retaining the voice tags.

[0010] The standardized text is subjected to linguistic preprocessing to obtain a word list; each word in the word list comprises a root form and a corresponding part-of-speech tag.

[0011] Calculating the importance of each term in the term list specifically includes the following steps: calculating the standard TF-IDF value of all terms; wherein, for terms with speech tags, performing secondary processing based on the standard TF-IDF value:

[0012] M(t,d)=TF-IDF(t,d)×(1+log(1+F(t)));

[0013] Among them, M(t, d) represents the fused importance score of term t with speech tags to document d; TF-IDF(t, d) represents the standard TF-IDF value of term t to document d; F(t) is a comprehensive function of speech features, which is used to evaluate the importance of term t in the speech part, which is determined by speech salience, semantic relevance and contextual importance.

[0014] M(t, d) is used as the importance of the terms with speech tags; for the terms without speech tags, the standard TF-IDF value is used as the importance.

[0015] All terms are sorted in descending order according to the importance, the sorted term list is traversed, and terms whose importance is greater than the screening threshold are selected as keywords; the screening threshold is set according to the frequency of occurrence of all terms.

[0016] As a preferred solution of the information technology consulting service method based on correlation analysis of the present invention, the speech feature comprehensive function is expressed as:

[0017]

[0018] The speech saliency is expressed as:

[0019]

[0020] The semantic relevance is expressed as:

[0021]

[0022] The context importance is expressed as:

[0023]

[0024] Among them, A(t) represents phonetic prominence; B(t) represents semantic relevance; C(t) represents contextual importance; D(t) represents the syllable duration ratio of term t; E represents the average syllable duration; G(t) represents the relative volume of the term; H represents the average volume; J(t) represents the word vector similarity; K(t) represents the normalized frequency ranking score; L(t) represents contextual relevance; N(t) represents the position score of term t in the sentence.

[0025] The process of constructing the user demand model includes the following steps: performing entity recognition on the keywords, marking and classifying the recognized entities to form entity information; analyzing the original text of the user consultation request containing the keywords, and constructing a dependency tree containing the keywords; embedding the entity information into the dependency tree, marking the corresponding entity information at the corresponding nodes of the dependency tree, forming a structured dependency tree containing entity information, and serving as a user demand model.

[0026] As a preferred solution of the information technology consulting service method based on correlation analysis of the present invention, processing the user demand and performing parameterized retrieval include the following steps:

[0027] A multi-level parameterized query tree is constructed based on the user's needs, and query parameters are mapped to nodes of different levels of the multi-level parameterized query tree; wherein each node contains a query parameter type, a value range and a weight; parameterized retrieval is performed based on the multi-level parameterized query tree; the parameterized retrieval includes starting from the root node of the multi-level parameterized query tree, accessing each node in a preset traversal order, constructing query conditions according to the query parameter type, value range and weight of the node, combining into a complete query statement and executing until a preset retrieval depth is reached or all nodes are covered, and finally merging the retrieval results of all nodes to obtain the retrieval result of the parameterized retrieval.

[0028] As a preferred solution of the information technology consulting service method based on correlation analysis described in the present invention, wherein: correlation analysis is performed on the retrieval results of the parameterized retrieval, and a correlation list is generated based on the results of the correlation analysis, including the following steps: using the BM25 algorithm to perform correlation analysis on the retrieval results of the parameterized retrieval, and outputting a correlation score as the result of the correlation analysis; sorting the retrieval results of the parameterized retrieval in descending order according to the correlation score, and filtering out the retrieval results whose correlation score is higher than the correlation score threshold; deduplicating the filtered retrieval results, and generating a correlation list based on the deduplicated retrieval results.

[0029] As a preferred solution of the information technology consulting service method based on correlation analysis described in the present invention, wherein: analyzing each entry in the correlation list and outputting a set of answers that meet user needs, the method includes the following steps: dividing the document of each entry in the correlation list into separate sentences, performing word segmentation on each sentence, and adding a part-of-speech tag to each vocabulary unit; based on the part-of-speech tag, identifying the actions and participants in the sentence, and constructing semantic role annotations; using the semantic role annotations, calculating the overlap between the keywords contained in the sentence and the keywords in the user demand model; combining the semantic role annotations and the overlap, evaluating the similarity between the dependency structure of each sentence and the structured dependency tree in the user demand model; calculating the score of each sentence, and accumulating them to obtain a document score; based on the document score, selecting the top M documents as answer candidates, where M is a preset positive integer; extracting the sentences with the highest scores from the answer candidates to form answers, and integrating them to form an answer set.

[0030] In the second aspect, the present invention provides an information technology consulting service system based on correlation analysis, including: an extraction and identification module, used to extract keywords in user consultation requests, and identify user needs in the keywords through a user demand model; a processing and generation module, used to process the user needs, perform parameterized retrieval, perform correlation analysis on the retrieval results of the parameterized retrieval, and generate a correlation list based on the results of the correlation analysis; an answer generation module, used to analyze each entry in the correlation list, and output a set of answers that meet the user needs; an interactive question and answer interface module, used to integrate the answer set to form an interactive question and answer interface for users to view.

[0031] In a third aspect, an embodiment of the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, any step of the information technology consulting service method based on correlation analysis as described in the first aspect of the present invention is implemented.

[0032] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the information technology consulting service method based on correlation analysis as described in the first aspect of the present invention.

[0033] The beneficial effects of the present invention are as follows: the present invention improves the performance and user experience of information technology consulting services by integrating multimodal input processing, deep semantic analysis, parameterized retrieval, and multi-dimensional correlation calculation. The present invention achieves accurate understanding of complex user queries, improves the relevance and accuracy of search results, and enhances the adaptability and knowledge coverage of the system through dynamic weight adjustment and knowledge graph integration. In addition, the answer generation mechanism of the present invention combines semantic role labeling and structured dependency analysis, which can provide more accurate and targeted answers, thereby greatly improving the quality and efficiency of consulting services. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0035] Figure 1 The present invention is an overall flow chart of the information technology consulting service method and system based on correlation analysis.

[0036] Figure 2This is a computer device diagram of the information technology consulting service method and system based on correlation analysis in the present invention. DETAILED DESCRIPTION

[0037] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0038] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0039] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0040] Example 1, reference Figure 1 , which is the first embodiment of the present invention, and which provides an information technology consulting service method based on correlation analysis, comprising:

[0041] In the related technologies, firstly, most systems perform poorly in processing complex, multimodal user queries, especially in integrating text and voice input. Secondly, existing systems often adopt static retrieval models, which are difficult to adapt to the personalized needs of users and professional terms in different fields. Thirdly, when building user demand models, most systems fail to make full use of linguistic features and semantic structure information, resulting in an insufficient understanding of user intent. In addition, existing correlation analysis methods usually only consider limited dimensions, such as TF-IDF or cosine similarity, ignoring important factors such as the timeliness of documents and user interaction history. Finally, in the answer generation stage, many systems still rely on simple extractive methods and cannot provide sufficiently detailed and customized answers.

[0042] This application provides an effective solution to the above-mentioned problems. Next, multiple embodiments will be combined to explain in detail how to implement the information technology consulting service based on correlation analysis;

[0043] Figure 1 The overall flow chart of an information technology consulting service method and system based on correlation analysis is shown, including:

[0044] S1: Extract keywords from user consultation requests and identify user needs in the keywords through the user demand model;

[0045] S2: Process user requirements, perform parameterized search, apply multi-dimensional relevance calculation to the search results of parameterized search, and generate a relevance list;

[0046] S3: Analyze each item in the relevance list and output a set of answers that meet the user's needs;

[0047] S4: Integrate the answer set to form an interactive question-and-answer interface for users to view.

[0048] In this embodiment, each step in S1 to S4 will be described one by one:

[0049] S1: extracting keywords from a user's consultation request, and identifying user needs in the keywords through a user need model.

[0050] Specifically, the types of user consultation requests include text form, voice form and text-speech fusion form. Among them, the text-speech fusion form refers to the combination of text form and voice form, and has a variety of application scenarios, such as: 1. Multimodal input: users use text and voice at the same time in the same interaction (users can start voice input and then switch to text to supplement or correct information). 2. Real-time display of speech-to-text: the system displays the recognized text in real time when the user is performing voice input; the user directly edits the displayed text to correct errors or add additional information while or after voice input. 3. Application of assistive technology: For some special user groups with hearing impairments or visual impairments, a hybrid input mode in the form of text-speech fusion is needed to improve the accessibility of the system.

[0051] S1.1: Extract keywords from user consultation requests, specifically using natural language processing technology to capture keywords from collected user consultation requests.

[0052] First, the collected user consultation requests are preprocessed, specifically: if the user consultation request is in voice form, use speech recognition technology (such as Google Speech-to-Text API: Google Speech-to-Text API) to convert it into text form; if the user consultation request is in the form of text-speech fusion, use multimodal separation technology to separate it into text and voice parts, use speech recognition technology to convert the voice part into text form, add speech tags, and merge it with the text part; all types of user consultation requests are uniformly converted into standardized text format, and the speech tags are retained.

[0053] Secondly, the standardized text is linguistically preprocessed to obtain a list of terms. The process of linguistic preprocessing is as follows: 1. Using stemming and lemma reduction tools in natural language processing technology (such as Porter Stemmer or lemma reducer in the Natural Language Toolkit NLTK library), the words in the text are converted into root forms, that is, the most basic form after all deformations are removed; 2. Stop words are removed and part-of-speech tags are performed. Therefore, each term in the list of terms contains the root form and the corresponding part-of-speech tag (the speech tags will not be deleted during the linguistic preprocessing process).

[0054] Then, the TF-IDF (Term Frequency-Inverse Document Frequency) algorithm is used to calculate the importance of each term in the term list, which includes the following steps:

[0055] Calculate the standard TF-IDF value of all terms, expressed as:

[0056]

[0057] Where TF-IDF(t, d) represents the standard TF-IDF value of term t for document d; t，d N represents the number of times term t appears in document d; d Represents the total number of words in document d; N represents the total number of documents; n t represents the number of documents containing term t; Represents the frequency of term t in document d. The higher the term frequency, the more important the term is in the document. Represents the inverse document frequency of term t, reflecting the rarity of the term in the entire document collection. The larger the value, the more important the term.

[0058] Among them, for words with speech tags, secondary processing is performed based on the standard TF-IDF value:

[0059] M(t,d)=TF-IDF(t,d)×(1+log(1+F(t)));

[0060]

[0061]

[0062] Among them, M(t, d) represents the fused importance score of term t with speech tags to document d; F(t) is a comprehensive function of speech features, which is used to evaluate the importance of term t in the speech part, determined by speech salience, semantic relevance and contextual importance; A(t) represents speech salience; B(t) represents semantic relevance; C(t) represents contextual importance; D(t) represents the syllable duration ratio of term t, which measures the ratio of the pronunciation duration of term t in actual speech to its expected duration; E represents the average syllable duration; G(t) represents the relative volume of the term; H represents the average volume; J(t) represents the word vector similarity; K(t) represents the normalized frequency ranking score; L(t) represents the contextual relevance; N(t) represents the position score of term t in the sentence.

[0063] It should be noted that the speech feature comprehensive function F(t) consists of a numerator and a denominator. In the numerator: First, the speech salience A(t) reflects the emphasis in the speech input; second, sin(B(t)) introduces periodic changes to simulate the modulation effect of the semantic association B(t) on the speech salience A(t), so as to capture the interactive effect between speech emphasis and semantic association; finally, the context importance C(t) is directly added to the numerator to retain the direct influence of context importance. The denominator: |A(t)-C(t)| reflects the difference between speech salience and context importance; (1-e -B(t) ) ranges from [0, 1) and increases with the increase of semantic relevance B(t); the denominator makes the value of the speech feature synthesis function F(t) relatively lower when the difference between speech saliency and context importance is large and the semantic relevance is high, thereby giving a more balanced consideration in the final score.

[0064] Preferably, the speech comprehensive feature function F(t) of the present invention significantly improves the performance and accuracy of the multimodal information retrieval system in practical applications. The comprehensive feature function F(t) combines speech saliency, semantic relevance and contextual importance to effectively capture the emphasis, semantic changes and contextual relationships in speech input, thereby more accurately understanding the user's intention. When processing query requests in the form of text-speech fusion, the comprehensive feature function F(t) can dynamically balance the influence of speech features and text semantics, so that the present invention can maintain a high degree of adaptability when facing different user expression habits and query complexity. This improvement makes the present invention have obvious advantages over the prior art in application scenarios that require precise understanding and rapid response, such as intelligent customer service, professional consulting and cross-language information retrieval, which not only improves the relevance and accuracy of the retrieval results, but also better handles the complex problems of ambiguous queries and context dependence, thereby significantly improving the user experience and the effectiveness of the present invention.

[0065] Table 1 Performance comparison of multimodal information retrieval systems

[0066] index F(t) of the present invention Traditional TF-IDF method BERT-based methods Query understanding accuracy (%) 92.5 85.3 87.1 Average response time (ms) 78 95 89 Relevance score (0-1) 0.876 0.792 0.815 User satisfaction (1-5) 4.62 4.13 4.25 Multimodal fusion efficiency (%) 89.7 76.4 79.2 Success rate of ambiguous query processing (%) 86.3 74.8 77.5 Contextual relevance improvement (%) 23.6 12.7 15.4 Computational complexity (relative value) 1.15 1.00 1.08

[0067] As can be seen from Table 1, the speech comprehensive feature function F(t) proposed in the present invention is significantly superior to the existing traditional TF-IDF method and BERT-based method in multiple key performance indicators. In terms of query understanding accuracy, F(t) reached 92.5%, which is 7.2 percentage points higher than the traditional TF-IDF method and 5.4 percentage points higher than the BERT-based method, reflecting its advantage in understanding complex query intent. Although the computational complexity of (t) is slightly higher than that of the traditional method (1.15 vs 1.00), its average response time (78ms) is lower than that of the TF-IDF method (95ms) and the BERT method (89ms), indicating that F(t)'s more accurate query understanding reduces subsequent processing steps. In terms of multimodal fusion efficiency and ambiguous query processing, F(t) achieved high performance of 89.7% and 86.3% respectively, far exceeding traditional and BERT methods, highlighting its excellent ability in processing complex multimodal inputs. In addition, F(t)'s performance in improving contextual relevance (23.6%) is almost twice that of traditional methods (12.7%), and is significantly better than the BERT method (15.4%), demonstrating its effective use of contextual information. These technical advantages ultimately translate into higher user satisfaction (4.62 vs 4.13 / 4.25), fully verifying the value of this invention in practical applications. In summary, these data strongly support the significant contribution of the speech comprehensive feature function F(t) in improving the performance of multimodal information retrieval systems.

[0068] Furthermore, the average syllable duration E is expressed as:

[0069]

[0070] The average volume H is expressed as:

[0071]

[0072] The syllable duration ratio D(t) of term t is expressed as:

[0073]

[0074] This ratio reflects whether the pronunciation time of word t in speech is different from expected. If this ratio is greater than 1, it means that the pronunciation time of the word is longer than expected; if it is less than 1, it means that the pronunciation time is shorter than expected, which may indicate that the speaker emphasizes or downplays certain words.

[0075] The relative volume G(t) of a term is expressed as:

[0076]

[0077] The word vector similarity J(t) is expressed as:

[0078]

[0079] in, is the word vector of term t; It is the word vector of the core concept of the query.

[0080] The frequency ranking normalized score K(t) is expressed as:

[0081]

[0082] The context relevance L(t) is expressed as:

[0083]

[0084] Among them, PMI(t, i) is the pointwise mutual information; context(t) is the context window of term t.

[0085] The position score N(t) of term t in a sentence is expressed as:

[0086]

[0087] Here, pos(t) is the position of term t in the sentence (counting starts from 1); len(s) is the length of the sentence (number of terms).

[0088] M(t, d) is taken as the importance of terms with speech tags; for terms without speech tags, the standard TF-I F value is taken as the importance of the terms.

[0089] Finally, all terms are sorted in descending order according to the importance, the sorted term list is traversed, and terms with importance greater than the screening threshold are selected as keywords.

[0090] Among them, the screening threshold is set according to the frequency of occurrence of all terms.

[0091] It should be noted that this embodiment uses Google Speech-to-Text API to convert the consultation request in voice form into text form, thus realizing seamless conversion from voice to text, enabling the system to process inputs in different forms, and improving the applicability and user-friendliness of the system. Using Porter Stemmer or WordNetLemmatizer in the NLTK library for stemming and lemmatization, the basic form of words in the text is extracted, which helps to reduce the diversity of words and simplifies the subsequent processing steps.

[0092] Preferably, by supporting multiple input forms (text, voice and text-speech fusion forms), the present invention can better adapt to modern interactive environments, such as multimodal input scenarios, real-time speech-to-text applications, and auxiliary technology applications for special user groups. By using multimodal separation technology and speech recognition technology, the present invention can accurately process these complex input forms and uniformly convert them into standardized text formats. This unified processing not only simplifies the subsequent analysis steps, but also ensures the consistency and comparability of keyword extraction under different input forms. In addition, in the linguistic preprocessing stage, the present invention uses stemming and word form restoration technology, which helps to reduce the interference caused by word form changes and improve the accuracy and generalization ability of keyword recognition.

[0093] The present invention introduces an innovative secondary processing mechanism when using the TF-IDF algorithm to calculate the importance of terms, mainly for terms with speech tags. This secondary processing mechanism significantly improves the accuracy and relevance of keyword extraction by fusing speech features and semantic information. The present invention introduces three indicators, namely, speech salience, semantic relevance and contextual importance, and comprehensively considers the prominence of terms in speech, the semantic similarity with the core concept of the query, and the importance in the context. This multi-dimensional evaluation method enables the present invention to more accurately capture the key points emphasized by the user in the speech input, while considering the semantic importance and contextual relevance of the terms. In order to accurately understand the user's intention, the present invention identifies the words that the user particularly emphasizes or weakens in the speech by considering the syllable duration ratio and relative volume of the terms. At the same time, by calculating the word vector similarity and the frequency ranking normalized score, the present invention can better identify the keywords related to the core concept of the user's query. In addition, the introduction of contextual relevance and sentence position score enables the present invention to consider the importance of the terms in the entire sentence structure, thereby extracting more context-related keywords. This method, which comprehensively considers speech features, semantic associations and contextual information, not only improves the accuracy of keyword extraction, but also better adapts to the expression habits of different users and professional terminology in specific fields, thereby providing more accurate and relevant information retrieval and analysis services in various complex consulting scenarios, such as customer service, professional consulting, medical diagnosis and other fields.

[0094] S1.2: Identify user needs in keywords through user need models. Specifically, use named entity recognition technology and dependency syntax analysis tools to build a user need model, and identify user needs in keywords through the user need model.

[0095] Specifically, the process of building a user demand model includes the following steps:

[0096] First, use named entity recognition technology (such as the Spacy (natural language processing library) model or the Stanford NER (Stanford named entity recognizer) model) to identify keywords, and mark and classify the identified entities to form entity information. Assuming that the keywords are artificial, intelligent, development, and process, the NER model may identify artificial intelligence as a composite entity;

[0097] Secondly, use dependency syntactic analysis tools (such as Spacy's dependency syntactic analysis module or StanfordParser) to analyze the original text of the user consultation request containing keywords, identify the dependency relationship between words (for example, analyze the sentence "I want to know the relationship between words in the development history of artificial intelligence"), and build a dependency relationship tree containing keywords based on the dependency syntactic analysis results (for example, I am the subject, want is the predicate verb, know is the object complement, and the development history of artificial intelligence is the object).

[0098] Finally, the entity information is embedded in the dependency tree, and the corresponding entity information is marked at the corresponding nodes of the dependency tree to form a structured dependency tree containing entity information, which is used as the user demand model.

[0099] It should be noted that the use of named entity recognition technology to perform entity recognition on keywords realizes the identification and classification of key entities in user queries, which helps to build a more accurate user demand model.

[0100] By using dependency syntactic analysis tools to build a dependency tree, we can achieve a deep understanding of the relationship between words in the user's information request text, which helps to grasp the user's query intent more accurately.

[0101] By embedding entity information into the dependency tree and creating a structured user demand model, it can fully reflect the user's query intention and provide precise guidance for subsequent information retrieval.

[0102] S2: Process user needs, perform parameterized search, perform relevance analysis on the search results of the parameterized search, and generate a relevance list based on the results of the relevance analysis.

[0103] S2.1: Build a multi-level parameterized query tree based on user needs, and map query parameters to different level nodes of the multi-level parameterized query tree, where each node contains the query parameter type, value range, and weight.

[0104] S2.2: Perform parameterized retrieval based on a multi-level parameterized query tree.

[0105] Specifically, parameterized retrieval includes starting from the root node of the multi-level parameterized query tree, visiting each node in a preset traversal order, building query conditions according to the query parameter type, value range and weight of the node, combining them into a complete query statement and executing it until the preset retrieval depth is reached or all nodes are covered, and finally merging the retrieval results of all nodes to obtain the retrieval results of the parameterized retrieval.

[0106] S2.3: Use the BM25 algorithm to perform correlation analysis on the retrieval results of the parameterized retrieval, and output the correlation score as the result of the correlation analysis.

[0107] S2.4: Sort the search results of the parameterized search in descending order according to the relevance score, set a relevance score threshold, and filter out the search results with relevance scores higher than the relevance score threshold; perform deduplication processing on the filtered search results, delete the search results with content duplication exceeding a preset ratio, and generate a relevance list based on the deduplication-processed search results.

[0108] S3: Analyze each entry in the relevance list and output a set of answers that meet the user's needs.

[0109] S3.1: Use a natural language processing tool (such as NLTK or Spacy) to split the document of each entry in the relevance list into separate sentences, perform tokenization on each sentence, that is, break the sentence into individual lexical units, and add a part-of-speech tag to each lexical unit to determine its grammatical function in the sentence (noun, verb, adjective, etc.).

[0110] It should be noted that the part-of-speech tags of S3.1 are not the same as the part-of-speech tags of the terms in S1. The former part-of-speech tags are performed on the retrieved documents in order to prepare for the subsequent semantic analysis and answer generation. The latter part-of-speech tagging belongs to the step: linguistic preprocessing of the standardized text to obtain a list of terms, which is performed on the original user input in order to prepare for the subsequent keyword extraction and user demand model construction. The processing objects are also different: the former processes user input, and the latter processes search results; and the processing granularity is also different: the former processes to the term level, and the latter processes to the sentence level first, and then to the word level.

[0111] S3.2: Apply semantic analysis technology to identify actions (predicates) and participants (arguments) in sentences. For example, in the sentence "I want to know the development history of artificial intelligence", "want" is an action, "I" is the executor of the action, "know" is another action, and "the development history of artificial intelligence" is the object of "knowing"; construct semantic role annotations.

[0112] S3.3: Using semantic role labeling, calculate the overlap between the keywords contained in the sentence and the keywords in the user demand model, expressed as,

[0113]

[0114] Among them, J(A, B) represents the overlap between the keywords contained in the sentence and the keywords in the user demand model, A represents the keyword set in the sentence, and B represents the keyword set in the user demand model.

[0115] S3.4: Combine semantic role labeling and overlap to evaluate the similarity between the dependency structure of each sentence and the structured dependency tree in the user demand model.

[0116] S3.5: Based on the overlap obtained from S3.3 and the similarity obtained from S3.4, the score of each sentence is calculated and accumulated to obtain the document score.

[0117] S3.6: Based on the document scores, select the top M documents as answer candidates, where M is a preset positive integer.

[0118] S3.7: Extract the sentences with the highest scores from the answer candidates, form the answers, and integrate them to form an answer set.

[0119] S4: Integrate the answer set to form an interactive question-and-answer interface for users to view.

[0120] Design a user-friendly front-end interface that displays a collection of answers and allows users to click through the answers. The interface should be interactive so that users can select an answer to view further details or ask new related questions.

[0121] Implement a feedback mechanism where users rate the quality of answers. This feedback can be used to improve future answer generation algorithms. The interface can also provide a search box where users can directly enter new queries and the system will re-execute all steps to generate a new set of answers.

[0122] Ultimately, by integrating all the filtered answers, users can easily and quickly find the information they need and further explore related information as needed.

[0123] It should be noted that designing a user-friendly front-end interface and implementing a feedback mechanism enables effective interaction between users and the system, improves user experience, and provides a basis for continuous improvement of the system.

[0124] All screened answers are integrated to form an interactive question-and-answer interface, allowing users to easily and quickly find the information they need and further explore relevant information as needed, ultimately achieving an efficient, accurate, and user-friendly question-and-answer system.

[0125] Through these steps, the invention realizes an intelligent question-answering system that can accurately understand user needs, efficiently retrieve relevant information, and provide high-quality answers, greatly improving user experience and information retrieval efficiency.

[0126] In summary, the present invention improves the performance and user experience of information technology consulting services by integrating multimodal input processing, deep semantic analysis, parameterized retrieval, and multi-dimensional correlation calculation. The present invention achieves accurate understanding of complex user queries, improves the relevance and accuracy of search results, and enhances the adaptability and knowledge coverage of the system through dynamic weight adjustment and knowledge graph integration. In addition, the answer generation mechanism of the present invention combines semantic role labeling and structured dependency analysis, which can provide more accurate and targeted answers, thereby greatly improving the quality and efficiency of consulting services.

[0127] Example 2, reference Figure 1 , which is a second embodiment of the present invention, provides an information technology consulting service system based on correlation analysis, including: a keyword capturing module, responsible for collecting consulting requests from users, and using natural language processing technology to capture keywords in the consulting requests;

[0128] The user demand identification module is responsible for building a user demand model based on keywords and using named entity recognition technology and dependency syntax analysis tools to identify user needs;

[0129] The context-aware retrieval module is responsible for activating the context-aware mechanism based on the identified user needs, retrieving background information related to the user needs, and forming a relevance list;

[0130] The answer generation module is responsible for further analyzing each item in the relevance list using advanced natural language processing algorithms and outputting a set of answers that meet user needs;

[0131] The interactive question-and-answer interface module is responsible for integrating all screened answer sets to form an interactive question-and-answer interface for users to view.

[0132] Example 3, reference Figure 2, which is the third embodiment of the present invention, which is different from the previous embodiment in that: if the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0133] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.

[0134] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering or, if necessary, processing in another suitable manner, and then stored in a computer memory.

[0135] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0136] Example 4 is the fourth example of the present invention. In order to further verify the technical solution of the present invention, experimental simulation data of an information technology consulting service method based on correlation analysis is provided.

[0137] The experimental environment is a high-performance workstation equipped with an Intel Core i9-11900K processor, 128GB RAM, and an NVIDIA RTX 3090 graphics card. The experimental dataset contains 100,000 diverse user queries, covering text, speech, and text-speech fusion, as well as 1,000,000 corresponding documents covering multiple technical fields. To ensure the comprehensiveness and fairness of the experiment, three existing technologies were selected for comparison: the traditional TF-IDF retrieval method, the BERT-based semantic retrieval method, and a commercial intelligent question-answering system.

[0138] The experimental process strictly follows the following steps: First, all methods are parameter optimized to ensure that they operate in their respective optimal states. Then, each method is tested using the same dataset, focusing on the performance of key links such as query understanding, parameterized retrieval, relevance analysis, and answer generation. In the query understanding stage, the present invention first converts the multimodal input into a standardized text format, and then extracts keywords using the improved TF-IDF algorithm and speech feature comprehensive function. Next, the user demand model is constructed through named entity recognition and dependency syntax analysis. In the parameterized retrieval stage, a multi-level parameterized query tree is constructed based on the user demand model, and a depth-first search is performed to obtain preliminary retrieval results. In the relevance analysis stage, the optimized BM25 algorithm is used, combined with dynamic weight adjustment and knowledge graph integration, to accurately sort the retrieval results. Finally, in the answer generation stage, semantic role labeling and structured dependency analysis are used to extract and combine the most relevant information fragments from high-relevance documents to form the final answer. Throughout the process, the system continuously records various performance indicators, including processing time, accuracy, relevance score, etc. In order to evaluate the performance of the system under different loads, the experiments were conducted under low load (10,000 queries / hour), medium load (50,000 queries / hour) and high load (100,000 queries / hour) conditions. Under each load condition, the experiment was repeated 50 times and the average value was taken as the final result to eliminate the influence of random factors and ensure the reliability and representativeness of the data.

[0139] Table 2: Query processing performance comparison

[0140]

[0141] As shown in Table 2, the present invention has achieved an excellent performance of 98.7% in the success rate of multimodal input processing, far exceeding other methods. This fully demonstrates the powerful ability of the present invention in processing complex and diverse user input. It is worth noting that the present invention only takes 38ms to build a user demand model, which is much faster than other methods, while maintaining a high keyword extraction accuracy (94.6%) and entity recognition accuracy (93.8%). Table 2 shows that the present invention can not only quickly process user input, but also accurately understand user intent, laying a solid foundation for subsequent retrieval and answer generation.

[0142] Table 3. Comparison of retrieval and correlation analysis performance

[0143]

[0144]

[0145] As shown in Table 3, the construction time of the parameterized query tree of the present invention is only 41ms, which is 41.4% faster than the commercial intelligent question answering system, while the traditional TF-IDF method and the BERT-based method do not have this advanced feature. The retrieval accuracy (P@10) reached 0.942, 2.4 percentage points higher than the second place, and the average retrieval time was only 73ms, which is the fastest among all methods. The present invention has excellent performance in both dynamic weight adjustment effect (23.6%) and knowledge graph integration efficiency (96.2%). The combination of these two technologies greatly improves the system's processing ability for complex queries and the relevance of the results. The optimized BM25 algorithm has achieved a relevance score of 0.924, which is significantly higher than other methods. These data fully demonstrate the outstanding contribution of the present invention in improving retrieval efficiency, accuracy and relevance.

[0146] Table 4 Answer generation and system performance comparison

[0147]

[0148] As shown in Table 4, the present invention has excellent performance in answer generation and overall system performance. The answer generation accuracy rate reached 94.3%, and the answer relevance score was as high as 0.931, both of which were significantly ahead of other methods. In terms of semantic role labeling accuracy, the present invention reached 93.1%, providing key support for generating more accurate answers that better meet user needs. The user satisfaction score was as high as 4.82 points (out of 5 points), which fully reflects the superiority of the present invention in practical applications. From the perspective of overall system performance, the total system response time of the present invention is only 205ms, which is 28.6% faster than traditional methods. Under high load conditions, the throughput of the present invention reached 486 queries / second, which is 10% higher than the second place, while maintaining a high resource utilization rate of 75.3%. These data show that the present invention not only improves the response speed and processing power of the system, but also optimizes the efficiency of computing resource utilization.

[0149] In summary, the present invention shows significant performance advantages in various aspects such as query understanding, parameterized retrieval, relevance analysis and answer generation, which are mainly reflected in: (1) efficient processing of multimodal input; (2) fast and accurate construction of user demand model; (3) fast construction and accurate retrieval of parameterized query tree; (4) efficient relevance analysis combining dynamic weight adjustment and knowledge graph; (5) accurate answer generation based on semantic role labeling; (6) excellent system load handling capacity. These advantages not only overcome the shortcomings of the existing technology in processing complex queries and providing accurate answers, but also greatly improve the overall quality and efficiency of information technology consulting services. The experimental data in Tables 2 to 4 strongly prove that the present invention can provide users with a faster, more accurate and more satisfactory consulting service experience, and has important value and broad prospects in practical applications.

[0150] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. An information technology consulting service method based on correlation analysis, characterized in that: include, Extracting keywords from the user's consultation request, and identifying the user's needs in the keywords through a user needs model; Processing the user demand, executing parameterized search, performing relevance analysis on the search results of the parameterized search, and generating a relevance list based on the results of the relevance analysis; Analyze each entry in the relevance list and output a set of answers that meet the user's needs; The answer set is integrated to form an interactive question-and-answer interface for users to view.

2. The information technology consulting service method based on correlation analysis according to claim 1, characterized in that: The types of the user consultation request include text form, voice form and text-speech fusion form.

3. The information technology consulting service method based on correlation analysis according to claim 2, characterized in that: Extracting keywords from user consultation requests includes the following steps: Preprocessing the user consultation request; the preprocessing includes: If the user consultation request is in voice form, it is converted into text form; if the user consultation request is in text-speech fusion form, it is separated into a text part and a voice part, the voice part is converted into text form, and voice tags are added, and then merged with the text part; all types of user consultation requests are uniformly converted into a standardized text format, and the voice tags are retained; Performing linguistic preprocessing on the standardized text to obtain a word list; each word in the word list includes a root form and a corresponding part-of-speech tag; Calculating the importance of each term in the term list specifically includes the following steps: Calculate the standard TF-IDF value of all terms; for terms with speech tags, perform secondary processing based on the standard TF-IDF value: M(t,d)=TF-IDF(t,d)×(1+log(1+F(t))); Where M(t, d) represents the fused importance score of term t with speech tags to document d; TF-IDF(t, d) represents the standard TF-IDF value of term t to document d; F(t) is a speech feature comprehensive function, which is used to evaluate the importance of term t in the speech part, determined by speech salience, semantic relevance and contextual importance; Let M(t, d) be the importance of the word with the phonetic mark; For terms without speech tags, the standard TF-IDF value is used as the importance; All terms are sorted in descending order according to the importance, the sorted term list is traversed, and terms whose importance is greater than a screening threshold are selected as keywords; the screening threshold is set according to the frequency of occurrence of all terms.

4. The information technology consulting service method based on correlation analysis according to claim 3, characterized in that: The speech feature comprehensive function is expressed as: The speech saliency is expressed as: The semantic relevance is expressed as: The context importance is expressed as: Among them, A(t) represents the phonetic prominence; B(t) represents the semantic relevance; C(t) represents the contextual importance; D(t) represents the syllable duration ratio of term t; E represents the average syllable duration; G(t) represents the relative volume of the term; H represents the average volume; J(t) represents the word vector similarity; K(t) represents the frequency ranking normalized score; L(t) represents the contextual relevance; N(t) represents the position score of term t in the sentence; The process of constructing the user demand model includes the following steps: Performing entity recognition on the keywords, and marking and classifying the recognized entities to form entity information; Analyzing the original text of the user consultation request containing the keyword, and constructing a dependency tree containing the keyword; The entity information is embedded in the dependency tree, and the corresponding entity information is marked at the corresponding nodes of the dependency tree to form a structured dependency tree containing the entity information, which is used as a user demand model.

5. The information technology consulting service method based on correlation analysis according to claim 4, characterized in that: Processing the user demand and performing parameterized search includes the following steps: Constructing a multi-level parameterized query tree based on the user's needs, and mapping query parameters to nodes at different levels of the multi-level parameterized query tree; wherein each node includes a query parameter type, a value range, and a weight; Parameterized retrieval is performed based on the multi-level parameterized query tree; the parameterized retrieval includes starting from the root node of the multi-level parameterized query tree, accessing each node in a preset traversal order, constructing query conditions according to the query parameter type, value range and weight of the node, combining them into a complete query statement and executing it until a preset retrieval depth is reached or all nodes are covered, and finally merging the retrieval results of all nodes to obtain the retrieval result of the parameterized retrieval.

6. The information technology consulting service method based on correlation analysis according to claim 5, characterized in that: Performing a correlation analysis on the retrieval results of the parameterized retrieval and generating a correlation list based on the results of the correlation analysis includes the following steps: Using the BM25 algorithm to perform correlation analysis on the retrieval results of the parameterized retrieval, and outputting a correlation score as a result of the correlation analysis; Sorting the retrieval results of the parameterized retrieval in descending order according to the relevance score, and screening out the retrieval results whose relevance score is higher than a relevance score threshold; The filtered search results are deduplicated, and a relevance list is generated based on the deduplicated search results.

7. The information technology consulting service method based on correlation analysis according to claim 6, characterized in that: Analyzing each entry in the relevance list and outputting a set of answers that meet the user's needs includes the following steps: Segment the document of each entry in the relevance list into separate sentences, perform word segmentation on each sentence, and add a part-of-speech tag to each vocabulary unit; Based on the part-of-speech tags, identifying actions and participants in the sentence, and constructing semantic role annotations; Utilizing the semantic role labeling, calculating the overlap between the keywords contained in the sentence and the keywords in the user demand model; In combination with the semantic role labeling and the overlap degree, evaluating the similarity between the dependency structure of each sentence and the structured dependency tree in the user demand model; Calculate the score of each sentence and add them up to get the document score; Based on the document scores, select the top M documents as answer candidates, where M is a preset positive integer; The sentences with the highest scores are extracted from the answer candidates to form answers, and the answers are integrated to form an answer set.

8. An information technology consulting service system based on correlation analysis, based on the information technology consulting service method based on correlation analysis according to any one of claims 1 to 7, characterized in that: include, An extraction and identification module is used to extract keywords from user consultation requests and identify user needs in the keywords through a user need model; A processing and generating module, used for processing the user demand, executing parameterized search, performing correlation analysis on the search results of the parameterized search, and generating a correlation list based on the results of the correlation analysis; An answer generation module, used for analyzing each entry in the correlation list and outputting an answer set that meets the user's needs; The interactive question-and-answer interface module is used to integrate the answer set to form an interactive question-and-answer interface for users to view.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the information technology consulting service method based on correlation analysis described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the information technology consulting service method based on correlation analysis described in any one of claims 1 to 7 are implemented.

Images