An AI consultation-based enterprise service system and method

Abstract

This invention relates to the field of enterprise consulting service technology, specifically to an AI-based enterprise service system and method, comprising a context construction module, a knowledge organization module, a matching and recognition module, a path organization module, and an output generation module. It establishes a correspondence between service request phrases and operational verbs, constructs structural groups based on constraints, reorganizes semantic content by matching enterprise knowledge fragments, filters execution text to generate case comparisons, identifies temporal logic to establish a sequential linked list, and pairs task information to generate consulting response text. This invention clarifies consulting intent by constructing phrase correspondences and binding semantic boundaries, extends knowledge fragments through overlap comparison and semantic reorganization to improve question adaptability, optimizes case matching through keyword and service object dual filtering, and establishes a sequential linked list to map task execution order, enhancing the coherence and logical flow of responses.

Description

Technical Field

[0001] This invention relates to the field of enterprise consulting service technology, and in particular to an enterprise service system and method based on AI consulting. Background Technology

[0002] The field of enterprise consulting services encompasses methods and tools for providing systematic analysis and decision support for internal management and external operations. Its core content lies in providing management with consulting solutions through the collection, organization, and analysis of enterprise information, including system design, strategic planning, business process optimization, and human and financial management support. It typically covers the establishment of information acquisition channels, standardized management methods for data and knowledge, and an analytical system that matches decision-making needs, thus constituting a comprehensive technical field encompassing information technology, enterprise management theory, and consulting methodologies.

[0003] Among them, the enterprise service system and method based on AI consulting refers to using natural language processing and knowledge reasoning techniques in artificial intelligence to transform management or business problems raised by enterprises into a processable data form, and then generating analysis and suggestions through preset consulting rules and industry knowledge bases. The technical matters covered include enterprise policy interpretation, process standardization suggestions, business risk warnings and resource allocation guidance. Specifically, it achieves problem understanding and solutions through language semantic recognition, problem matching, knowledge retrieval and rule mapping, thus forming the technical realization of the enterprise consulting service process.

[0004] Existing technologies primarily extract static fields when processing consultation input, lacking the expression of dynamic structural relationships between service requests and operational verbs. When the input contains complex semantic combinations or multi-layered logical goals, semantic confusion is easily generated, leading to unclear interpretation of consultation intent. In the knowledge matching process, fragment extraction mainly focuses on title keywords, ignoring the inherent semantic dependencies of the context, which can easily cause information combination breaks. In addition, the response generation stage does not identify and map the temporal relationships in the user input, resulting in problems such as unbalanced content order and jumpy expression in the output results, affecting the executability and logical coherence of consultation suggestions. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of existing technologies by proposing an AI-based enterprise service system and method.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: an enterprise service system based on AI consulting, the system comprising:

[0007] The context construction module obtains service request phrases, operational verbs, and limiting words from enterprise consultation input, matches operational verbs and request phrases in word order, sets limiting words as boundaries to limit semantic scope, and combines them in order to form context structure content, generating a consultation context structure table.

[0008] The knowledge organization module calls up the request phrases in the consultation context structure table, compares them with the case titles, the first sentences of the viewpoint paragraphs and the first sentences of the industry entries in the consultation materials for overlapping words, marks the texts that meet the overlap requirements, and combines them in structural order to obtain a set of enterprise knowledge fragments.

[0009] The matching and recognition module extracts user needs, service objects and execution content based on the enterprise knowledge fragment set. It performs keyword matching by combining the target words in the user input with the verb combinations in the execution content, and filters the content that meets the requirements of overlap ratio and text correspondence to generate an AI consulting case comparison set.

[0010] The path organization module calls the execution content from the AI ​​consultation case comparison set, extracts time expression words to form a time chain, combines sequence words and connecting words to form an execution chain, compares the task order in the user input with the number of execution chain offsets, constructs a task order mapping, and generates a consultation service order linked list.

[0011] As a further embodiment of the present invention, the consultation context structure table includes service request phrase location mapping items, key operation verb semantic annotation items, and constraint condition boundary binding items; the enterprise knowledge fragment set includes a set of high-frequency overlapping terms, semantically similar content fragments, and contextually coherent reorganized content; the AI ​​consultation case comparison set includes keyword overlap results, service object matching items, and filtered execution text fragments; and the consultation service sequence list includes time sequence nodes, logical connection chain groups, and sequence mapping relationship pairs.

[0012] As a further aspect of the present invention, the context construction module includes a semantic extraction submodule, a position mapping submodule, a semantic binding submodule, and a structure organization submodule;

[0013] The semantic extraction submodule obtains all phrases in the enterprise consultation input text, filters out expressions with verbal properties and service items with noun properties, extracts all service request phrases and key operational verbs based on the syntactic position of the verb expression and the collocation structure of adjacent words, and generates a semantic phrase set after removing modifiers and tone expressions that have no semantic carrying function.

[0014] The position mapping submodule, based on the syntactic position of each service request phrase and key operation verb in the semantic phrase set, calls the position index value of the semantic phrase in the original text, extracts its hierarchical structure in the sentence, calculates the position mapping value of the phrase pair based on the word order relationship between the verb and the noun, and obtains the word order position mapping matrix.

[0015] The semantic binding submodule, based on the determined word pair structure in the word order position mapping matrix, obtains the prepositional structures and time-related adverb phrases representing restrictive conditions in the text. By identifying their contextual adjacency and main clause hierarchy in the text, it filters and maps the restrictive condition phrases that share subject-predicate or modifying structures, constructs the contextual boundary relationship between them and the operating verb, and obtains the restrictive condition boundary binding sequence.

[0016] The structure organization submodule calls all word pairs in the constraint boundary binding sequence and word order position mapping matrix, and combines key operation verbs with their bound constraints according to the order of appearance of service request phrases in the original text. After splicing the three structures, the sequence is numbered to generate a consultation context structure table.

[0017] As a further aspect of the present invention, the knowledge organization module includes a phrase comparison submodule, a content tagging submodule, and a semantic linking submodule;

[0018] The phrase comparison submodule retrieves all service request phrases from the consultation context structure table, collects past case title sentences, expert opinion paragraph opening sentences, and industry entry first line content from enterprise consultation materials, compares them one by one based on the word sequence in the phrase unit, counts the number of phrases in each content segment that are literally consistent with the service request phrases, and filters out content with an overlap count greater than or equal to the overlap benchmark value to obtain the number of phrase overlap segments.

[0019] The content tagging submodule re-splits the content structure of each segment in the number of overlapping segments based on the original source type of each segment, extracts the main words and subordinate components of the sentence in which the segment is located, calls the type value of the key operation verbs in the context structure table, filters out segments that do not overlap with the operation intention, and performs sequence tagging on the remaining part to generate context matching tag quantity.

[0020] The semantic concatenation submodule calls all the valid segments from the context matching markers, constructs a segment sorting index according to the order of appearance of service request phrases in the original text, determines the sorting order based on the co-occurrence of main words and verb order relationships in each group of segments, and splices all segments in their logical order to establish a set of enterprise knowledge segments.

[0021] As a further embodiment of the present invention, the matching and recognition module includes a content extraction submodule, a keyword comparison submodule, and a result filtering submodule;

[0022] The content extraction submodule acquires all text content in the enterprise knowledge fragment set, identifies the service scenario expression structure in each fragment, extracts phrase segments containing user demand descriptions, phrase units describing service objects, and verb structure groups containing execution links, removes non-instructional fragments and content with missing verb structures, and establishes the execution content phrase quantity;

[0023] The keyword comparison submodule, based on each set of verb phrases in the execution content phrase quantity, calls the target phrase in the user input, counts the word overlap between the keywords in the target phrase and the keywords in the verb phrase, calculates the overlap number divided by the number of words in the target phrase to obtain the overlap ratio, and generates the phrase overlap ratio.

[0024] The result filtering submodule, based on each ratio in the phrase overlap ratio, calls the service object description text in the corresponding segment, detects the semantic co-occurrence value between it and the object-type phrase in the user input statement, and judges based on the overlap ratio and co-occurrence value with the set keyword overlap threshold and semantic relevance threshold, respectively, and filters the segments that meet the dual threshold conditions to obtain the AI ​​consultation case comparison set.

[0025] As a further embodiment of the present invention, the path organization module includes a time extraction submodule, a sequence construction submodule, and a sequence mapping submodule;

[0026] The time extraction submodule obtains all execution process content from the AI ​​consultation case comparison set, extracts words with time meaning to form a sequence, calls the grammatical structure relationship between verbs in the sentence and their corresponding time expression words, filters out time words with sequential expression properties, removes repeated or non-time modifiers, and generates time expression sequence values.

[0027] The sequential construction submodule collects logical connectors that co-occur with each time term in the time expression sequence value, and connects the time terms and logical connectors according to the original text arrangement order to construct a chain structure of continuous expression. After removing node pairs with overlapping semantic conflicts, the sequential connection list value is obtained.

[0028] The sequence mapping submodule calls the content of each linked list node in the sequential connection linked list value, collects the verb groups in the task submission order in the enterprise user input text, judges the correspondence between the linked list node and the task verb based on the word meaning matching result, counts and summarizes the position difference between the two in their respective sequences, calculates the average magnitude of all differences, establishes the sequence offset relationship value, and generates the consultation service sequence linked list after performing sequence mapping based on the sequence offset relationship value.

[0029] As a further aspect of the present invention, the process of determining the correspondence between the linked list node and the task verb based on the word meaning matching result specifically involves: calculating the word vectors of the content of each linked list node in the sequentially connected linked list value and the verb group in the task submission order in the enterprise user input text, calculating the cosine similarity between the word vectors, and determining that there is a correspondence between the linked list node and the task verb when the cosine similarity is greater than a preset matching threshold.

[0030] The process of sequence mapping based on the order offset relationship value is as follows: if the order offset relationship value is positive, the execution order of the nodes in the consultation service order chain is shifted forward by the number of steps corresponding to the average magnitude; if the order offset relationship value is negative, the execution order of the nodes in the consultation service order chain is shifted backward by the number of steps corresponding to the average magnitude.

[0031] As a further aspect of the present invention, the system further includes:

[0032] The output generation module extracts the user-input question phrases and matches them with the task descriptions in the chain list according to the execution structure in the consultation service sequence chain list. It then organizes the corresponding content into complete expression fragments and combines them into continuous response segments according to the chain list order to generate enterprise consultation response text.

[0033] The enterprise consultation response text includes structured response statements, sequentially paired expression fragments, and task execution semantic combination sentences.

[0034] As a further aspect of the present invention, the output generation module includes a task matching submodule, a fragment generation submodule, and a text assembly submodule;

[0035] The task matching submodule retrieves all task description fields from the consultation service sequence list, extracts keyword content from the user input, compares the position of each keyword with the verb phrases in the task description field according to the position of the keyword in the text, and determines the matching based on whether the two form a word overlap structure in the main components of the sentence, and generates a keyword matching quantity value.

[0036] The fragment generation submodule calls the word pairs that have been matched in the keyword pairing quantity value, extracts the phrase structure of the verb in the task description field and its matching keyword, and integrates them into a complete expression structure according to the subject-verb-object combination rule. It then filters out fragment combinations with semantic conflicts or syntactic breaks to obtain the semantic expression combination quantity.

[0037] The text assembly submodule arranges each expression fragment in the semantic expression combination according to its original order in the consultation service sequence chain, sequentially placing it in the overall sentence, splicing all semantic expression fragments to form a coherent text, and then filling in any missing conjunctions and sequence words at the splicing nodes to create the enterprise consultation response text.

[0038] An AI-based enterprise service method, executed based on the aforementioned AI-based enterprise service system, includes the following steps:

[0039] S1: Obtain enterprise consultation texts in resource allocation planning scenarios, identify service request phrases, key operational verbs and limiting words, establish the positional correspondence between phrases and verbs, extract the boundary semantic relationship between limiting words and target verbs, complete the structural group organization and sequence arrangement, and obtain the consultation context structure table.

[0040] S2: Call the service request phrases in the consultation context structure table, select the case title short sentences, the first line of industry entries and the first sentence of expert opinions in the enterprise knowledge base in the performance optimization management scenario, perform word segmentation and overlap frequency statistics, set the word overlap limit value to filter effective semantic fragments, and perform semantic recombination of effective fragments according to the original order to obtain the enterprise knowledge fragment set;

[0041] S3: Based on the enterprise knowledge fragment set, extract user demand descriptions, service object descriptions and execution process content in the job responsibility design scenario, establish a set of input target phrases and execution verb phrases, calculate the keyword overlap ratio, and perform conditional filtering based on the relevance of the service object to obtain an AI consulting case comparison set;

[0042] S4: Call the execution text in the AI ​​consultation case comparison set, identify time expression words and sequence expression words in the schedule scenario, construct a sequence node linked list by combining logical connector words, compare the task order in the user input with the word positions in the node linked list, calculate the offset and establish a mapping, and generate a consultation service sequence linked list;

[0043] S5: Based on the structure of the consultation service sequential chain list, extract the user's input question keywords and task fields in the system implementation scenario, perform position matching and sentence recombination, and splice the sentence fragments into complete text according to the sequential chain structure to obtain the enterprise consultation response text.

[0044] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0045] In this invention, when processing enterprise consultation texts, a precise correspondence between service requests and operational verbs is constructed, and semantic boundaries are used to bind constraints, thereby improving the structural clarity of problem intent extraction. The semantic extension of knowledge fragments is achieved through term overlap and semantic reorganization mechanisms, enhancing the logical continuity and problem adaptability between texts. The accuracy of case identification is improved by using keyword overlap ratio and service object relevance as dual screening methods. A sequential linked list is constructed by combining time expression words and logical connector words, and the mapping between task execution order and user input is completed accordingly, ensuring the structural order, logical consistency, and expression continuity of the response content. Attached Figure Description

[0046] Figure 1 This is a system flowchart of the present invention;

[0047] Figure 2 This is a flowchart illustrating the process of obtaining the context construction module of this invention.

[0048] Figure 3 This is a flowchart illustrating the knowledge acquisition process of the knowledge organization module in this invention.

[0049] Figure 4 This is a flowchart illustrating the acquisition process of the matching and recognition module in this invention.

[0050] Figure 5 This is a flowchart illustrating the acquisition process of the path organization module in this invention.

[0051] Figure 6 This is a flowchart illustrating the process of obtaining the output generation module of this invention. Detailed Implementation

[0052] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0053] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0054] In the embodiments of this invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning. Similarly, the terms "of," "corresponding (relevant)," and "corresponding" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, they convey the same meaning.

[0055] In this embodiment of the invention, sometimes a subscript such as W1 may be written in a non-subscript form such as W1. When the difference is not emphasized, the meaning they express is the same.

[0056] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0057] Please see Figure 1 This invention provides a technical solution: an enterprise service system based on AI consulting, the system comprising:

[0058] The context construction module obtains service request phrases, key operation verbs, and limiting condition words from the enterprise's consultation input text. It matches the service request phrases with the key operation verbs in terms of position, binds the limiting condition words with the target action words according to the semantic order of the context, and then organizes the binding results into structural groups and arranges them in order to generate a consultation context structure table.

[0059] The knowledge organization module calls the service request phrases in the consultation context structure table and compares them with the title sentences of past cases, the first sentences of expert opinion paragraphs, and the first line of industry entries in the enterprise's consultation materials. Content that meets the limit of overlap is marked as a valid fragment. Then, the valid fragments are reorganized according to the semantic connection order to obtain the enterprise knowledge fragment set.

[0060] The matching and recognition module, based on the enterprise knowledge fragment set, extracts the user demand description, service object description and execution process content. It calculates the keyword overlap between the target phrase in the user input and the action verb phrase in the execution process, and performs two-level filtering based on the overlap ratio and the relevance of the service object text. It retains the text that meets the overlap requirements and generates an AI consulting case comparison set.

[0061] The path organization module calls the execution process content from the AI ​​consulting case comparison set, extracts time expression words from the text to construct a continuous sequence, combines sequence expression words with logical connector words to form a sequential linked list, compares the task submission order in the enterprise user input with the number of order offsets in the sequential linked list, establishes a sequence mapping relationship, and generates a consulting service sequential linked list.

[0062] The output generation module extracts the user-input keywords from the sequential structure of the consultation service sequence list and pairs them with the task description field in the sequence list. The matching results are then reassembled into coherent sentence fragments, and all the fragments are concatenated in the execution order to form the response text, generating the enterprise consultation response text.

[0063] The consultation context structure table includes service request phrase location mapping items, key operation verb semantic annotation items, and constraint condition boundary binding items. The enterprise knowledge fragment set includes a set of high-frequency overlapping terms, semantically similar content fragments, and contextually coherent reorganized content. The AI ​​consultation case comparison set includes keyword overlap results, service object matching items, and filtered execution text fragments. The consultation service sequence list includes time sequence nodes, logical connection chain groups, and sequence mapping relationship pairs. The enterprise consultation response text includes structured response statements, sequentially paired expression fragments, and task execution semantic combination sentences.

[0064] Please see Figure 2 The context construction module includes a semantic extraction submodule, a position mapping submodule, a semantic binding submodule, and a structure organization submodule;

[0065] The semantic extraction submodule obtains all phrases in the enterprise consultation input text, filters out expressions with verbal properties and service items with noun properties, extracts all service request phrases and key operational verbs based on the syntactic position of the verb expression and the collocation structure of adjacent words, and generates a semantic phrase set after removing modifiers and tone expressions that have no semantic carrying function.

[0066] The text "Our company hopes to optimize its existing supply chain management process, specifically to improve order processing efficiency and reduce logistics and warehousing costs. A preliminary plan is required to be completed within three months and must be compatible with the existing ERP system" is retrieved. Dependency analysis identifies verbal expressions such as "optimize," "improve," "reduce," "complete," and "compatible," and identifies corresponding noun service items: "supply chain management process," "order processing efficiency," "logistics and warehousing costs," "preliminary plan," and "ERP system." Based on the syntactic relationship between the verb "optimize" and the following noun "supply chain management process," and the absence of other core verbs separating them, the service request phrase "supply chain management process" and the key operational verb "optimize" are extracted. Using the same syntactic position and collocation analysis method, the service request phrases "order processing efficiency" and "optimize" are extracted sequentially. The key operation verbs are "improve", "logistics and warehousing costs" and "reduce", "preliminary plan" and "complete", and "ERP system" and "compatible". Next, the system identifies modifying and moderating expressions in the text, such as "our company hopes", "the specific goal is", "requirement", and "and need". These phrases are determined to be modifying or auxiliary components in the syntactic analysis tree and do not carry core service instructions. Therefore, they are removed from the extracted phrase set, generating a semantic phrase set containing five phrase pairs: {service request phrase: "supply chain management process", key operation verb: "optimize"}, {service request phrase: "order processing efficiency", key operation verb: "improve"}, {service request phrase: "logistics and warehousing costs", key operation verb: "reduce"}, {service request phrase: "preliminary plan", key operation verb: "complete"}, and {service request phrase: "ERP system", key operation verb: "compatible"}.

[0067] The position mapping submodule, based on the syntactic position of each service request phrase and key operation verb in the semantic phrase set, calls the position index value of the semantic phrase in the original text, extracts its hierarchical structure in the sentence, calculates the position mapping value of the phrase pair based on the word order relationship between the verb and the noun, and obtains the word order position mapping matrix;

[0068] Based on the first service request phrase "supply chain management process" and the key operational verb "optimize" in the semantic phrase set, we retrieve their position index values ​​in the original text "Our company hopes to optimize the existing supply chain management process...". After word segmentation, the starting index value of "optimize" is 5, and the starting index value of "supply chain management process" is 7. Both are extracted to belong to the first main clause level in the syntactic analysis tree. Based on the word order relationship between the verb "optimize" and the noun "supply chain management process", we calculate the position mapping value of the phrase pair. The calculation method is the noun's starting index minus the verb's starting index, i.e., 7 minus 5 equals 2. This positive value indicates that the noun follows the verb. Similarly, for "improve order processing efficiency", its position mapping value is 8; for "reduce logistics and warehousing costs", its position mapping value is 8; for "complete the preliminary plan", its position mapping value is 10; and for "compatible with the existing ERP system", its position mapping value is 6. Combining these five phrase pairs and their corresponding position mapping values ​​yields the word order position mapping matrix.

[0069] The semantic binding submodule, based on the word pair structure determined in the word order position mapping matrix, obtains the prepositional structures and time-related adverb phrases representing restrictive conditions in the text. By identifying their contextual adjacency and main clause hierarchy in the text, it filters and maps restrictive condition phrases that share subject-predicate or modifying structures, constructs the contextual boundary relationship between them and the operating verb, and obtains the restrictive condition boundary binding sequence.

[0070] Based on the established word pair structure {"complete", "preliminary plan"} in the word order mapping matrix, phrases expressing restrictive conditions in the original text are extracted. Part-of-speech tagging identifies the prepositional phrase "within three months" and time-related adverbial phrases. Analysis of their contextual adjacency reveals that "within three months" is immediately preceding "complete the preliminary plan." Syntactic analysis confirms that it shares the same main clause level with "complete the preliminary plan" and there is no subordinate clause separation. Therefore, it is determined that "within three months" is a direct modification of the operational verb "complete". The modifiers are used to construct a contextual boundary relationship between the modifier and the verb "complete", forming a binding pair ("complete", "within three months"). For the phrase pair {"compatible", "ERP system"}, the adjacent modifier "existing" is identified, and a binding pair ("compatible", "existing") is constructed. For other phrase pairs such as {"optimize", "supply chain management process"}, no prepositional or adverbial structures with time, place, manner, or conditional restrictions are identified within 5 word distances in their context, so no binding is performed, resulting in a constraint condition boundary binding sequence.

[0071] The structure organization submodule calls all word pairs in the constraint boundary binding sequence and word order position mapping matrix. Following the order of appearance of service request words in the original text as the main line, it combines key operation verbs with their bound constraints, and after splicing the three structures, it assigns sequence numbers to generate a consultation context structure table.

[0072] The algorithm calls upon all word pairs in the constraint boundary binding sequence and word order position mapping matrix. Following the order of appearance of service request words in the original text, namely "supply chain management process", "order processing efficiency", "logistics and warehousing cost", "preliminary plan", and "ERP system" as the main thread, it combines key operational verbs with their bound constraints. For the first word pair {"optimize", "supply chain management process"}, since it has no binding conditions, it is directly concatenated. For the fourth word pair {"complete", "preliminary plan"}, it is concatenated with the binding condition "within three months" to form the structure group ("complete", "preliminary plan", "within three months"). After completing the combination and concatenation of all word pairs, the five generated structure groups are sequentially numbered, starting from 1 and increasing sequentially, to generate the consultation context structure table.

[0073] Table 1: Structure of Consultation Context

[0074]

[0075] As shown in Table 1, this table fully records the structured consultation context parsed from user input.

[0076] Please see Figure 3 The knowledge organization module includes a phrase comparison submodule, a content tagging submodule, and a semantic linking submodule;

[0077] The phrase comparison submodule retrieves all service request phrases from the consultation context structure table, collects past case title sentences, expert opinion paragraph opening sentences, and industry entry first line content from enterprise consultation materials, compares them one by one based on the word sequence in the phrase unit, counts the number of phrases in each content segment that are literally consistent with the service request phrases, and filters out content with an overlap count greater than or equal to the overlap benchmark value to obtain the number of phrase overlap segments.

[0078] All service request phrases, such as "supply chain management process," were retrieved from the consulting context structure table. Additionally, the titles of 1500 past case studies, the opening sentences of 800 expert opinion paragraphs, and the first lines of 2000 industry entries stored in the company's internal knowledge base were collected. Each phrase was compared one by one based on its word sequence within the term unit. For example, comparing "supply chain management process" with the case title "XX Group's Supply Chain Management Process Reengineering Practice" revealed a complete overlap of the terms "supply chain," "management," and "process," with a term overlap count of 3. The number of terms in each content segment that literally matched "supply chain management process" was counted. This overlap count was then compared to a benchmark value. The benchmark value was set based on data analysis of 1000 successful consulting cases. Statistics showed that when the number of overlapping terms between a service request phrase and a knowledge segment is greater than or equal to 50% of the total number of terms in that segment, its reference value significantly increases. In this example, "supply chain management process" contains the three core terms "supply chain," "management," and "process," so the benchmark value was set to 3. 50% is rounded down to 2. Since the number of overlapping titles in the aforementioned case is 3, which is greater than or equal to the baseline value of 2, the short title phrase "XX Group's Supply Chain Management Process Reengineering Practice" is selected. After a thorough comparison of all the data, the number of overlapping segments of all terms that meet the conditions is obtained.

[0079] The content tagging submodule re-splits the content structure of each segment in the number of overlapping segments based on the original source type, extracts the main words and subordinate components of the sentence in which the segment is located, calls the type value of the key operation verbs in the context structure table, filters out segments that do not overlap with the operation intention, and performs sequence tagging on the remaining part to generate context matching tag quantity.

[0080] Based on the original source type of each fragment in the overlapping fragment quantity, for example, the fragment "XX Group's Supply Chain Management Process Reengineering Practice" originates from past cases, its content structure is re-splittered, and the main words of the sentence containing the fragment are extracted, namely the subject "Group", the predicate "reengineering", the object "process", and the subordinate components "XX", "supply chain management", and "practice". The system then calls the key operation verb "optimize" corresponding to the service demand "supply chain management process" that matches the fragment in the context structure table. Its type value is defined as "improvement type". The system compares the semantics of "reengineering" with the semantics of "optimize". Both belong to the "improvement type" operation intent and have strong overlap, so the fragment is retained. If another fragment is "the definition and scope of supply chain management process", its operation intent is "descriptive type" and has no overlap with the "improvement type" intent of "optimize", then it is filtered out. All the remaining parts that pass the screening are sequence-marked to generate context matching mark quantity.

[0081] The semantic concatenation submodule calls all valid fragments from the context matching markers, constructs a fragment sorting index according to the order of service request phrases in the original text, determines the sorting order based on the co-occurrence of main words and verb order relationships in each group of fragments, and splices all fragments in their logical order to build a set of enterprise knowledge fragments.

[0082] The system calls upon all valid fragments from the context matching markers. Following the order of service request phrases in the original text (prioritizing "supply chain management process" and then "order processing efficiency"), a fragment sorting index is constructed. For the three valid fragments related to "supply chain management process": A: "Process status diagnosis is the first step in optimization," B: "Optimize processing nodes by introducing automated equipment," and C: "The optimized process needs stress testing," the system identifies "diagnosis" as a prerequisite for "optimization" and "optimization" as a prerequisite for "testing" based on the co-occurrence of key words and verb order. This determines their logical order as ABC. All fragments are then assembled according to this logical order to create a set of enterprise knowledge fragments containing multiple themes and possessing internal order.

[0083] Please see Figure 4 The matching and recognition module includes a content extraction submodule, a keyword comparison submodule, and a result filtering submodule;

[0084] The content extraction submodule acquires all text content in the enterprise knowledge fragment set, identifies the service scenario expression structure in each fragment, extracts phrase segments containing user demand descriptions, phrase units describing service objects, and verb structure groups containing execution content, removes non-instructional fragments and content missing verb structures, and establishes the execution content phrase quantity;

[0085] The system acquires all text content from the enterprise knowledge fragment set and identifies the service scenario expression structure in each fragment. For example, from the fragment "Providing services to manufacturing enterprises with annual sales of over 500 million, firstly analyzing inventory turnover rate, secondly evaluating existing logistics routes, and finally outputting optimization solutions," the system extracts the phrase segment describing user needs, "Providing services to manufacturing enterprises with annual sales of over 500 million," the phrase unit describing the service target, "manufacturing enterprises with annual sales of over 500 million," and the verb structure group {"analyze inventory turnover rate," "evaluate logistics routes," "output optimization solutions"} for the execution steps. At the same time, the system detects the fragment "Supply chain finance is a future development trend," which is judged as a non-instructional fragment and removed because it does not contain an imperative verb structure. The fragment "order system" is also removed because it lacks a verb structure for specific operations. Finally, a set of execution content phrases containing only executable steps is established.

[0086] The keyword comparison submodule, based on the verb phrases of each execution stage in the execution content phrase quantity, calls the target phrases in the user input, counts the word overlap between the keywords in the target phrases and the keywords in the verb phrases, calculates the overlap ratio by dividing the number of overlaps by the number of words in the target phrases, and generates the phrase overlap ratio.

[0087] Based on a set of execution verb phrases {"automated order review", "improve sorting efficiency"} from the execution content phrase set, the target phrase "improve order processing efficiency" from the user input is called. The target phrase set is divided into four keywords: {"improve", "order", "process", "efficiency"}. The overlap between the keywords in the target phrase set and the keywords in the verb phrase set {"automation", "order", "review", "improve", "sorting", "efficiency"} is counted. It is found that the words "order" and "efficiency" overlap by 2. The overlap is calculated by dividing the overlap by the number of words in the target phrase set, which gives the overlap ratio, i.e., 2 divided by 4 equals 0.5. This process is repeated for all verb phrases in the execution content phrase set to generate a series of phrase overlap ratios, such as 0.5, 0.75, 0.25, etc.

[0088] The results filtering submodule, based on each ratio in the phrase overlap ratio, calls the service object description text in the corresponding segment, detects the semantic co-occurrence value between it and the object-type phrase in the user input statement, and judges based on the overlap ratio and co-occurrence value with the set keyword overlap threshold and semantic relevance threshold respectively, filters the segments that meet the dual threshold conditions, and obtains the AI ​​consultation case comparison set.

[0089] Based on a phrase overlap ratio of 0.75, the service object description text "suitable for e-commerce warehouses with a daily order volume exceeding 5000" in the corresponding fragment is retrieved. The semantic co-occurrence value between this text and the implicit object-type phrase "our company" in the user input (assuming a daily order volume of 8000, as determined by a database query) is detected. The co-occurrence value is calculated to be 0.9 based on the similarity in business scale descriptions. A judgment is then made based on preset keyword overlap and semantic relevance thresholds. The keyword overlap threshold is set at the minimum value (0.6) that filters out 70% of successful cases, based on the analysis of 1000 historical cases. The semantic relevance threshold is also set at 0.7, based on historical data. Since the overlap ratio of 0.75 is greater than 0.6 and the semantic co-occurrence value of 0.9 is greater than 0.7, this fragment meets the dual threshold conditions and is selected and retained. Finally, an AI consulting case comparison set consisting of multiple selected fragments is obtained.

[0090] Please see Figure 5 The path organization module includes a time extraction submodule, a sequence construction submodule, and a sequence mapping submodule;

[0091] The time extraction submodule obtains all execution process content from the AI ​​consultation case comparison set, extracts words with time meaning to form a sequence, calls the grammatical structure relationship between verbs in the sentence and their corresponding time expression words, filters out time words with sequential expression properties, removes repeated or non-time modifiers, and generates time expression sequence values.

[0092] The code retrieves a specific execution phase from an AI consulting case study: "The first phase of the project, lasting one week, involves data research and requirements analysis. Next, from the second to the fourth week, solution design and prototype development are conducted. Finally, system testing is performed in the second month." It extracts time-related words to form a sequence ["first phase", "one week", "second to fourth week", "finally", "second month"]. The code then examines the grammatical relationship between the verb "complete" and its corresponding time expression "one week." It determines that "one week" represents the execution duration, while "first phase" is a sequential time term. Similarly, "second to fourth week" and "second month" are also sequential time terms, while "finally" is purely a logical sequence word. The code removes "one week" (representing duration) and non-time modifiers, and merges duplicate concepts to generate a time sequence value ["first phase", "second to fourth week", "second month"].

[0093] The sequential construction submodule collects logical connectives that co-occur with each time term in the time expression sequence value, and connects the time terms and logical connectives according to the original text arrangement order to construct a chain structure of continuous expression. After removing node pairs with overlapping semantic conflicts, the sequential connection list value is obtained.

[0094] Based on the chronological order of time terms in the time sequence value (i.e., "Phase 1" precedes "Week 2 to Week 4"), logical connectors that co-occur with them, such as "next" and "finally," are collected. The time terms and logical connectors are then connected according to the original text order to construct a node "Phase 1: Complete Data Survey and Requirements Analysis." The logical connector "next" is connected to the node "Week 2 to Week 4: Conduct Solution Design and Prototype Development" through directed edges, forming a chain structure of continuous expression. The system checks whether there are any semantic conflicts in this chain structure, such as whether the node "Week 5" appears after "Week 2." After confirming that there are no conflicts, the sequential connection list value is obtained.

[0095] The sequence mapping submodule calls the content of each linked list node in the sequential connection linked list value, collects the verb groups in the task submission order in the input text of enterprise users, judges the correspondence between the linked list nodes and the task verbs based on the word meaning matching results, counts and summarizes the position difference between the two in their respective sequences, calculates the average magnitude of all differences, establishes the sequence offset relationship value, and generates the consultation service sequence linked list after performing sequence mapping based on the sequence offset relationship value.

[0096] The process involves calling the content of each linked list node in the sequential connection list value. For example, node 1, "Complete data research and demand analysis," collects the task submission order implicit in the user's input text, i.e., the order of statement is 1-Optimize process, 2-Improve efficiency, 3-Reduce cost. Extract the verb phrases {"optimize", "improve", "reduce"}. Based on the semantic matching results, it is determined that the verbs "analyze" and "research" in linked list node 1 correspond most strongly to the user task verb "optimize". Linked list node 2, "Solution design," corresponds to the user task "Improve efficiency". The positions of the two in their respective sequences are 1 and 2, with a position difference of 0. The position differences of all corresponding nodes are counted and summarized. The average magnitude of all differences is calculated to establish a sequence offset relationship value. Based on this relationship value of 0, sequence mapping is directly performed to generate a consulting service sequence linked list.

[0097] Table 2: Sequential List of Consulting Services

[0098]

[0099] See Table 2, which clearly lists the service execution steps and timeline.

[0100] Please see Figure 6 The output generation module includes a task matching submodule, a fragment generation submodule, and a text assembly submodule;

[0101] The task matching submodule retrieves all task description fields from the consultation service sequence list, extracts keyword content from user input, compares the position of each keyword with the verb phrases in the task description field according to the position of the keyword in the text, and determines the matching based on whether the two form a word overlap structure in the main components of the sentence, and generates the keyword matching quantity value.

[0102] Retrieve all task description fields from the consultation service sequence list, such as the task description field "data research and demand analysis" in sequence 1, and extract the keyword content "supply chain", "order processing", and "logistics cost" from the user input. Based on the position of the keywords in the original text, compare them one by one with the verb phrases in the task description fields. Match the keyword "supply chain" with the task description "data research and demand analysis" to determine whether the two have word overlap or strong semantic relationship in the main components of the sentence. In this example, "data research" and "supply chain" are highly related to the steps before optimization, so they are judged as a match. Perform this judgment on all task descriptions and all keywords to generate the keyword pairing count.

[0103] The fragment generation submodule calls the word pairs that have been matched in the keyword pairing quantity value, extracts the phrase structure of the verbs in the task description field and their matching keywords, and integrates them into a complete expression structure according to the subject-verb-object combination rules. It then filters out fragment combinations with semantic conflicts or syntactic breaks to obtain the semantic expression combination quantity.

[0104] The system retrieves matched phrase pairs from the keyword pairing count, such as the keyword "supply chain" and the verbs "research" and "analysis" in the task description field. It extracts their phrase structures and integrates them according to the subject-verb-object combination rules to form "We will conduct data research and demand analysis on your supply chain status." This ensures the sentence is complete. Then, the system filters out possible semantic conflicts or syntactic breaks in the combination of fragments. For example, if a fragment such as "We will reduce your system compatibility" is generated, which contradicts the user's intention of "being compatible with existing ERP," it is filtered out. Finally, the system obtains a semantic expression combination consisting of multiple syntactically correct and semantically coherent expression fragments.

[0105] The text assembly submodule arranges each expression fragment in the semantic expression combination according to its original order in the consultation service sequence chain, and splices all semantic expression fragments to form a coherent text. It then checks whether there are any missing conjunctions and sequence words in the splicing nodes and fills them in to create the enterprise consultation response text.

[0106] Based on the original order of each expression fragment in the semantic expression combination in the consultation service sequence chain (i.e., fragment with priority 1 first, fragment with priority 2 last), their positions in the overall sentence are arranged sequentially. All semantic expression fragments are then spliced ​​together to form the initial draft: "We will conduct data research and demand analysis on your supply chain status, then design targeted solutions and develop prototypes, then conduct system testing and deployment, and finally complete a cost-benefit assessment." The splicing nodes are checked, and logical connectors and temporal words are found to be missing between nodes. Therefore, "according to your needs" is added at the beginning of the sentence, and "firstly," "secondly," "subsequently," "finally," etc. are added between the fragments. The time plan is confirmed based on the constraint "within three months." After completing the text, the final enterprise consultation response text is created.

[0107] An AI-based consulting approach to enterprise services includes the following steps:

[0108] S1: Obtain enterprise consultation texts in resource allocation planning scenarios, identify service request phrases, key operational verbs and limiting words, establish the positional correspondence between phrases and verbs, extract the boundary semantic relationship between limiting words and target verbs, complete the structural group organization and sequence arrangement, and obtain the consultation context structure table.

[0109] S2: Call the service request phrases in the consultation context structure table, select the case title short sentences, the first line of industry entries and the first sentence of expert opinions in the enterprise knowledge base in the performance optimization management scenario, perform word segmentation and overlap frequency statistics, set the word overlap limit value to filter effective semantic fragments, and perform semantic recombination of effective fragments according to the original order to obtain the enterprise knowledge fragment set;

[0110] S3: Based on the enterprise knowledge fragment set, extract user demand descriptions, service object descriptions and execution process content in the job responsibility design scenario, establish a set of input target phrases and execution verb phrases, calculate the keyword overlap ratio, and perform conditional filtering based on the relevance of the service object to obtain an AI consulting case comparison set;

[0111] S4: Call the execution text in the AI ​​consulting case comparison set, identify time expression words and sequence expression words in the schedule scenario, construct a sequential node linked list by combining logical connector words, compare the task order in the user input with the word positions in the node linked list, calculate the offset and establish a mapping, and generate a consulting service sequential linked list;

[0112] S5: Based on the structure of the consultation service sequential list, extract the user's input question keywords and task fields in the system implementation scenario, perform position matching and sentence reorganization, and splice the sentence fragments into complete text according to the sequential chain structure to obtain the enterprise consultation response text.

[0113] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. An AI consultation-based enterprise service system, characterized by, The system comprises: A context construction module acquires service appeal phrases, operation verbs and limiting condition words in enterprise consultation input, matches operation verbs with appeal phrases according to the order of words, sets limiting condition words as boundary to limit semantic range, sequentially combines to form context structure content, and generates consultation context structure table; The context construction module comprises a semantic extraction submodule, a position mapping submodule, a semantic binding submodule and a structure arrangement submodule; The semantic extraction submodule acquires all phrases in enterprise consultation input text, filters expression content with verb properties and noun service items, extracts all service appeal phrases and key operation verbs based on the syntactic position of verb expression and the collocation structure of adjacent words, filters out modifying components and mood expressions without semantic bearing, and generates a semantic phrase set; The position mapping submodule calls the position index value of the semantic phrase in the original text according to the syntactic position of each service appeal phrase and key operation verb in the semantic phrase set, extracts the hierarchical structure in the sentence, calculates the position mapping value of the phrase pair according to the order relationship between the verb and the noun, and obtains the syntactic position mapping matrix; The semantic binding submodule acquires preposition structures and time adverb phrases representing limiting conditions in the text according to the phrase pair structure determined in the syntactic position mapping matrix, filters limiting condition phrases that share subject-predicate structure or modifying structure with the mapping phrase pair by identifying their context adjacency and main sentence hierarchical relationship, constructs the context boundary relationship between the operation verb and the limiting condition, and obtains the limiting condition boundary binding sequence; The structure arrangement submodule calls the limiting condition boundary binding sequence and all phrase pairs in the syntactic position mapping matrix, combines the key operation verb and its bound limiting condition according to the order of appearance of the service appeal phrase in the original text as the main line, performs three structure splicing and sequence numbering, and generates the consultation context structure table; The knowledge arrangement module calls the appeal phrases in the consultation context structure table, compares them with the case titles, viewpoint paragraph first words and industry item first sentences in the consultation materials, marks the text that meets the coincidence requirement, combines them according to the structure order, and obtains the enterprise knowledge fragment set; The matching recognition module extracts user demand, service object and execution content based on the enterprise knowledge fragment set, combines target words in user input with verbs in execution content for keyword matching, filters content that meets the coincidence proportion and text corresponding requirements, and generates an AI consultation case comparison set; The path arrangement module calls the execution content in the AI consultation case comparison set, extracts time expression phrases into a time chain, combines sequential words and connecting words into an execution chain, compares the task order in user input with the execution chain offset number, constructs a task order mapping, and generates a consultation service order chain table. 2.The AI-advisory based enterprise service system of claim 1, wherein: The consultation context structure table comprises service appeal word group position mapping items, key operation verb semantic annotation items and restriction condition boundary binding items, the enterprise knowledge fragment set comprises a high-frequency coincident term collection, semantically similar content fragments and context coherent reorganized content, the AI consultation case contrast set comprises key word coincidence results, service object matching items and screened execution text fragments, and the consultation service order linked list comprises time sequence nodes, logical connection chain groups and order mapping relationship pairs. 3.The AI-advisory based enterprise service system of claim 1, wherein, The knowledge arrangement module comprises a word group comparison sub-module, a content marking sub-module and a semantic concatenation sub-module. The word group comparison sub-module acquires all service appeal word groups in the consultation context structure table, collects past case title short sentences, expert viewpoint paragraph first sentences and industry item first line content in enterprise consultation data, compares the word groups one by one based on word sequence in the word item unit, counts the number of word items identical to the service appeal word groups in each content fragment, screens out content with a coincidence number greater than or equal to a coincidence reference value, and obtains a word item coincidence fragment quantity. The content marking sub-module re-divides the content structure according to the original source type of each fragment in the word item coincidence fragment quantity, extracts the main word and dependent component of the sentence where the fragment is located, calls the type value of the key operation verb in the context structure table, screens out fragments with no intersection with the operation intention, performs sequence marking on the remaining part, and generates context matching marking quantity. The semantic concatenation sub-module calls all fragments marked as valid in the context matching marking quantity, constructs a fragment ordering index according to the appearance order of the service appeal word groups in the original text, determines the arrangement order according to the co-occurrence and verb order relationship between the main words in each group of fragments, splices all fragments in their logical order, and establishes an enterprise knowledge fragment set. 4.The AI-advisory-based enterprise service system of claim 1, wherein, The matching recognition module comprises a content extraction sub-module, a key word comparison sub-module and a result screening sub-module. The content extraction sub-module acquires all text content in the enterprise knowledge fragment set, recognizes the service scene expression structure in each fragment, extracts a word group section containing user demand description, a phrase unit containing service object description and a verb structure group containing execution link content, eliminates non-instructional fragments and content missing verb structures, and establishes an execution content word group quantity. The key word comparison sub-module calls target word groups in user input according to each execution link verb group in the execution content word group quantity, counts the word surface coincidence value of key words in the target word groups and key words in the verb groups, calculates the coincidence proportion value obtained by dividing the coincidence number by the number of target word groups, and generates a word group coincidence proportion value. The result screening sub-module calls service object description text in the corresponding fragment based on each value in the word group coincidence proportion value, detects the semantic co-occurrence degree value between the object class phrase in the user input sentence and the service object description text, judges the coincidence proportion value and the co-occurrence degree value according to the set key word coincidence threshold value and semantic correlation threshold value respectively, screens fragments meeting the double threshold conditions, and acquires an AI consultation case contrast set. 5.The AI-advisory based enterprise service system of claim 1, wherein, The path arrangement module comprises a time extraction sub-module, an order construction sub-module and an order mapping sub-module. The time extraction submodule obtains all execution link contents in the AI consultation case control set, extracts a sequence of words with time implications, calls a grammatical structure relationship between a verb in a sentence and a corresponding time expression word, filters out time words with sequential expression properties, and removes repeated or non-time modifiers to generate a time expression sequence value; The sequential construction submodule collects logical connection words co-occurring with each time word in the time expression sequence value based on the order of the time word, and connects the time word and the logical connection word in accordance with the original arrangement order of the text to construct a chain structure with sequential expression. After removing node pairs with cross semantic conflicts, a sequential connection list value is obtained. The sequential mapping submodule calls the content of each list node in the sequential connection list value, collects verb groups in the task submission order in the enterprise user input text, determines the correspondence between the list node and the task verb based on the word meaning matching result, counts the position difference between the two in their respective sequences and aggregates them, calculates the average magnitude of all difference values, establishes a sequential offset relationship value, and generates a consultation service sequential list after sequence mapping based on the sequential offset relationship value. 6.The AI-advisory-based enterprise service system of claim 5, wherein, The process of determining the correspondence between the list node and the task verb based on the word meaning matching result is as follows: the word vectors of each list node content in the sequential connection list value and the verb groups in the task submission order in the enterprise user input text are calculated, the cosine similarity between the word vectors is calculated, and when the cosine similarity is greater than a predetermined matching threshold, it is determined that the list node and the task verb have correspondence. The process of sequence mapping based on the sequential offset relationship value is as follows: if the sequential offset relationship value is positive, the execution order of the nodes in the consultation service sequential list is moved forward by the number of steps corresponding to the average magnitude; if the sequential offset relationship value is negative, the execution order of the nodes in the consultation service sequential list is moved backward by the number of steps corresponding to the average magnitude. 7.The AI-advisory-based enterprise service system of claim 1, wherein, The system further comprises: The output generation module extracts the user input question word group and the list task description according to the execution structure in the consultation service sequential list, organizes the corresponding content into a complete expression segment, combines it into a continuous reply segment in the list order, and generates an enterprise consultation reply text. The enterprise consultation reply text includes structured answer sentences, sequential paired expression segments, and task execution semantic combination sentences. 8.The AI-advisory-based enterprise service system of claim 7, wherein, The output generation module includes a task pairing submodule, a segment generation submodule, and a text assembly submodule. The task pairing submodule obtains all task description fields in the consultation service sequential list, extracts the keyword content in the user input, and performs position correspondence comparison with the verb phrases in the task description fields one by one according to the position of the keyword in the text. The matching determination is made based on whether the two form a word surface overlapping structure in the sentence trunk component to generate a keyword pairing quantity value. The fragment generation submodule calls the completed matching keyword pair in the keyword pair number value, extracts the phrase structure of the verb in the task description field and its matching keyword, integrates it into a complete expression structure according to the subject-predicate-object combination rule, screens out fragment combinations with semantic conflicts or syntactic interruptions, and obtains a semantic expression combination quantity; The text assembly submodule arranges the front and back positions of each expression fragment in the overall sentence according to the original order of the expression fragment in the consultation service order linked list, splices all semantic expression fragments to form a coherent text, detects whether the conjunction words and time sequence words of the splicing node are missing, and fills in the gaps, and establishes an enterprise consultation reply text. 9.A method for AI consultation-based enterprise service, characterized by, The method is used in the AI consultation-based enterprise service system of any one of claims 1-8, comprising the following steps: S1: Obtain the enterprise consultation text in the resource configuration planning scenario, identify the service demand word group, key operation verb and restriction condition word, establish the position correspondence relationship between the word group and the verb, extract the boundary semantic relationship between the restriction condition word and the target verb, complete the structure group arrangement and sequence arrangement, and obtain a consultation context structure table; S2: Call the service demand word group in the consultation context structure table, select the case title short sentence, industry item first line content and expert opinion first sentence in the enterprise knowledge base in the performance optimization management scenario, perform word segmentation and coincidence frequency statistics, set the word item coincidence limit value to filter effective semantic fragments, perform semantic reorganization of the effective fragments according to the original order, and obtain an enterprise knowledge fragment set; S3: Based on the enterprise knowledge fragment set, extract the user demand description, service object description and execution link content in the post responsibility design scenario, establish the input target word group and the execution verb group word item set, calculate the key word coincidence ratio, combine the service object related degree for condition screening, and obtain an AI consultation case comparison set; S4: Call the execution text in the AI consultation case comparison set, identify the time expression word and the order expression word in the progress arrangement scenario, construct an order node linked list combined with the logical connection word, compare the task order in the user input with the word group position in the node linked list, calculate the offset and establish the mapping, and generate a consultation service order linked list; S5: According to the structure content of the consultation service order linked list, extract the problem key word and task field of the user input in the system implementation scenario, perform position pairing and sentence reorganization, splice the sentence fragments into a complete text according to the order chain structure, and obtain an enterprise consultation reply text.

Images