AI large model-based power customer service semantic analysis method and device and medium

Abstract

The application provides an AI large model-based power customer service semantic analysis method and device and a medium, the method comprising: slicing a received power consultation text based on natural language to obtain multiple consultation text segments; performing semantic relationship analysis on each consultation text segment based on the semantic understanding capability of an AI large model, and constructing a semantic relationship graph based on the semantic relationship between words in each consultation text segment; determining a target power business scenario associated with each consultation text segment based on the semantic relationship graph and a power business knowledge graph; performing cross-segment semantic integration on each consultation text segment based on the logical connection between each consultation text segment to obtain a cross-segment semantic integration text; and outputting a reply text based on the cross-segment semantic integration text and the target power business scenario. The application can comprehensively and accurately handle customer consultations, significantly improving the efficiency and accuracy of power consultation services, thereby improving the user service experience.

Description

Technical Field

[0001] This invention relates to the field of computer technology, and in particular to a method, apparatus, and medium for semantic analysis of power customer service based on a large AI model. Background Technology

[0002] In the current electricity customer service operations, semantic analysis methods based on rule matching and simple keyword extraction are widely used. First, a knowledge base is established containing a large number of electricity business-related keywords and corresponding business scenarios. When the customer service system receives a customer's inquiry text, it performs word segmentation, breaking the text down into individual words. Next, keywords are extracted from these words and matched against keywords in the knowledge base. If a matching keyword is found, the system determines the customer's inquiry intent based on pre-defined rules in the knowledge base, thereby linking it to the corresponding business processing flow.

[0003] While semantic analysis methods for power customer service based on rule matching and simple keyword extraction can handle common customer inquiries to some extent, their ability to understand complex semantics is insufficient. They struggle to accurately handle customer questions with ambiguous meanings, diverse expressions, or those involving multiple business scenarios. Furthermore, existing methods may only identify a subset of keywords, failing to accurately understand the multiple complex intentions simultaneously involved by the customer. This leads to inaccurate and incomplete problem handling, impacting customer experience and service efficiency. With the continuous expansion of the power business and the increasing diversification of customer needs, this method is no longer sufficient to meet actual business requirements. Summary of the Invention

[0004] This invention provides a method, device, and medium for semantic analysis of power customer service based on an AI large model, in order to improve the efficiency and accuracy of power consultation services and enhance the user service experience.

[0005] In a first aspect, this invention provides a semantic analysis method for power customer service based on a large AI model, including:

[0006] The received electricity consultation text is sliced ​​based on natural language processing to obtain multiple consultation text fragments;

[0007] Based on the semantic understanding capabilities of the AI ​​large model, semantic relationships are analyzed for each consultation text fragment, and a semantic relationship graph is constructed based on the semantic relationships between words in each consultation text fragment;

[0008] Based on the semantic relationship graph and the power business knowledge graph, the target power business scenario associated with each consultation text fragment is determined.

[0009] Based on the logical connections between each consultation text fragment, cross-fragment semantic integration is performed on each consultation text fragment to obtain cross-fragment semantic integrated text;

[0010] The response text is output based on the cross-fragment semantic integration text and the target power business scenario.

[0011] Secondly, the present invention also provides an AI-based large-scale model-based semantic analysis device for power customer service, applied to the AI-based large-scale model-based semantic analysis method for power customer service as described in the first aspect; the AI-based large-scale model-based semantic analysis device for power customer service includes:

[0012] The text slicing module is used to slice the received electricity consultation text based on natural language to obtain multiple consultation text fragments;

[0013] The semantic relationship parsing module is used to perform semantic relationship parsing on each consultation text fragment based on the semantic understanding capabilities of the AI ​​large model, and to construct a semantic relationship graph based on the semantic relationships between words in each consultation text fragment;

[0014] The business scenario association module is used to determine the target power business scenario associated with each consultation text fragment based on the semantic relationship graph and the power business knowledge graph.

[0015] The text integration module is used to perform cross-fragment semantic integration of each consultation text fragment based on the logical connections between each consultation text fragment, to obtain cross-fragment semantically integrated text.

[0016] The intent recognition output module is used to output a response text based on the cross-fragment semantic integration text and the target power business scenario.

[0017] Thirdly, the present invention also provides an electronic device, comprising: a memory for storing computer software programs; and a processor for reading and executing the computer software programs, thereby realizing the power customer service semantic analysis method based on the AI ​​large model as described above.

[0018] Fourthly, the present invention also provides a non-transitory computer-readable storage medium storing a computer software program, which, when executed by a processor, implements the above-described AI-based large-scale model-based semantic analysis method for power customer service.

[0019] Fifthly, the present invention also provides a computer program product, including a computer program that, when executed by a processor, implements any of the above-mentioned AI-based large-scale model-based semantic analysis methods for power customer service.

[0020] The AI-based big data model-based semantic analysis method for power customer service provided in this invention leverages the powerful semantic understanding capabilities of the AI ​​big data model to adapt to various expression methods. As long as the semantics are the same or similar, it can accurately understand the intent. Furthermore, through text segmentation, semantic relationship parsing, and cross-segment semantic integration, it can deeply understand the complex semantics in power consultation texts. At the same time, through the identification of related business scenarios and cross-segment semantic integration, it can clearly sort out the relationships between different business scenarios, thereby accurately identifying multiple complex consultation intents in user consultation texts. This enables comprehensive and accurate handling of customer consultations, significantly improving the efficiency and accuracy of power consultation services, and thus enhancing the user service experience. Attached Figure Description

[0021] Figure 1 This is a flowchart illustrating the semantic analysis method for power customer service based on an AI large model provided in an embodiment of the present invention.

[0022] Figure 2 This is a schematic diagram of the structure of the power customer service semantic analysis device based on an AI large model provided in an embodiment of the present invention;

[0023] Figure 3 An embodiment diagram of the electronic device provided in this invention;

[0024] Figure 4 An embodiment diagram of a computer-readable storage medium provided in accordance with the present invention. Detailed Implementation

[0025] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0026] In the description of this invention, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0027] In the description of this invention, the term "for example" is used to mean "used as an example, illustration, or description." Any embodiment described as "for example" in this invention is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use the invention. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that the invention can be made without using these specific details. In other instances, well-known structures and processes will not be described in detail to avoid obscuring the description of the invention with unnecessary detail. Therefore, the invention is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed herein.

[0028] See Figure 1 , Figure 1 This is a flowchart illustrating the semantic analysis method for power customer service based on an AI large-scale model provided by the present invention. In this embodiment of the invention, the executing entity of the semantic analysis method for power customer service based on an AI large-scale model is a power customer service device. Therefore, the semantic analysis method for power customer service based on an AI large-scale model includes:

[0029] Step 10: Slice the received electricity consultation text into multiple consultation text fragments based on natural language.

[0030] Optionally, when users inquire about electricity-related issues, they need to input the electricity inquiry text into the electricity customer service device. Therefore, after receiving the electricity inquiry text, the electricity customer service device segments the text according to specific rules based on natural language processing technology, resulting in multiple inquiry text fragments, as described in steps 101 to 105. The specific rules can be based on punctuation marks (such as periods, question marks, exclamation marks, etc.), specific keywords (such as words indicating content transitions like "in addition" and "also"), and semantic coherence to determine the segment positions. In one embodiment, the electricity customer service device receives the inquiry text "My electricity bill suddenly increased this month, what's the reason? Also, there have been frequent power outages in the community recently, what's going on?". The electricity customer service device first identifies the period and the keyword "in addition", and segments the text into two fragments: Fragment 1 is "My electricity bill suddenly increased this month, what's the reason?"; Fragment 2 is "There have been frequent power outages in the community recently, what's going on?".

[0031] Step 20: Based on the semantic understanding capabilities of the AI ​​big data model, perform semantic relationship analysis on each consultation text fragment, and construct a semantic relationship graph based on the semantic relationships between words in each consultation text fragment.

[0032] Furthermore, for each consultation text fragment, the power customer service device utilizes the semantic understanding capabilities of an AI big data model to analyze the semantic relationships between words in the fragment, such as subject-predicate, verb-object, and modifier relationships, and presents these relationships in the form of a semantic relationship graph, as described in steps 201 to 204. The AI ​​big data model, learned from a large amount of text data, can accurately determine the semantic connections between words, providing a foundation for subsequently determining the business scenario.

[0033] Continuing with the above example, for segment 1, "My electricity bill suddenly increased this month, what's the reason?", the AI ​​model analyzes that "my family" is the subject, "electricity bill increased" is the predicate, "this month" is the time adverbial, and "reason" is the object being inquired about. In the constructed semantic relationship graph, "my family" and "electricity bill increased" have a subject-predicate association edge, "this month" and "electricity bill increased" have a time modification association edge, and "reason" and "electricity bill increased" have a causal relationship query-directing edge.

[0034] Step 30: Determine the target power business scenario associated with each consultation text fragment based on the semantic relationship graph and the power business knowledge graph.

[0035] Furthermore, the power customer service device matches the constructed semantic relationship graph with the power business knowledge graph. The power business knowledge graph contains various power business concepts, attributes, and the relationships between them. Therefore, by comparing the elements in the semantic relationship graph with the nodes and relationships in the business knowledge graph, the target power business scenario that best matches each consultation text fragment is obtained, as described in steps 301 to 304.

[0036] Continuing with the above examples, for fragment 1, "My electricity bill suddenly increased this month, what could be the reason?", the semantic relationship graph highlights the issue of abnormal electricity bills. The power business knowledge graph contains a "Power Bill Anomaly Analysis" business scenario node, which is related to sub-nodes such as "Electricity Usage," "Electricity Price Adjustment," and "Appliance Failure." Through matching, the target power business scenario associated with this fragment is determined to be the "Power Bill Anomaly Analysis" scenario. For fragment 2, "Recently, there have been frequent power outages in my neighborhood, what's going on?", the semantic relationship graph reflects the power outage issue, and it matches the "Power Outage Fault Troubleshooting" business scenario node in the power business knowledge graph, determining its target power business scenario to be "Power Outage Fault Troubleshooting."

[0037] Step 40: Based on the logical connections between each consultation text fragment, perform cross-fragment semantic integration on each consultation text fragment to obtain cross-fragment semantically integrated text.

[0038] Furthermore, the power customer service device analyzes the logical connections between each consultation text fragment, such as causal relationships, parallel relationships, and progressive relationships, and integrates the various consultation text fragments according to these logical connections to form a cross-fragment semantic integrated text with coherent semantics, so as to understand the user's consultation intent as a whole, as described in steps 401 to 404.

[0039] Continuing with the above embodiment, fragment 1 and fragment 2 are parallel, both being user inquiries about electricity issues. The electricity customer service device integrates them into "The user inquired about the reasons for the sudden increase in electricity bills and the frequent power outages in the community." If the text contains "Because of the power outage in the community, the items in my refrigerator spoiled, and the electricity bill seems to have increased," then there is a causal relationship between the fragments, and they are integrated into "The user stated that the power outage in the community caused damage to the items in the refrigerator, and there is a suspected increase in electricity bills, so they inquired about the relevant reasons."

[0040] Step 50: Output the response text based on cross-fragment semantic integration text and the target power business scenario.

[0041] Furthermore, the power customer service device inputs each consultation text fragment from the cross-fragment semantic integration text into its corresponding target power business scenario response module. The business scenario response module searches the log database for historical consultation and response records similar to the consultation text fragments and generates a response result for each consultation text fragment based on the matching results. Further, the power customer service device combines the response results into the final response text according to the priority of each consultation text fragment in the cross-fragment semantic integration text (which can be determined based on the order of user questions, the urgency of the question, etc.).

[0042] Continuing with the above examples, for segment 1, "My electricity bill suddenly increased this month, what could be the reason?", in the "Electricity Bill Anomaly Analysis" business scenario response module, the log database matches a similar inquiry, "Last month's electricity bill suddenly increased," and the response record is, "The increased electricity bill may be because you added a high-power appliance to your home this month, or the meter may be malfunctioning. You can check the usage of the appliance and contact us to arrange for someone to check the meter." For segment 2, "Recently, there have been frequent power outages in the community, what's going on?", in the "Power Outage Troubleshooting" business scenario response module, the log database matches an inquiry about "frequent power outages in the community," and the response record is, "Power outages in the community may be due to line maintenance or a sudden power failure. We are already investigating, please pay attention to the community announcements for the specific power restoration time." Following the order of Segment 1 and Segment 2 (i.e., the order of user questions), the combined response text is: "The increased electricity bill may be due to the addition of high-power appliances to your home this month, or a malfunction in the electricity meter. You can check the usage of your appliances and contact us to arrange for someone to check the meter. The power outage in the community may be due to line maintenance or a sudden power failure. We are already investigating, and please pay attention to the community announcements for the specific power restoration time."

[0043] This invention, through the powerful semantic understanding capabilities of a large AI model, can adapt to various expression methods. As long as the semantics are the same or similar, it can accurately understand the intent. Furthermore, through text segmentation, semantic relationship parsing, and cross-segment semantic integration, it can deeply understand the complex semantics in power consultation texts. At the same time, through the identification of related business scenarios and cross-segment semantic integration, it can clearly sort out the relationships between different business scenarios, thereby accurately identifying multiple complex consultation intents in user consultation texts. This enables comprehensive and accurate handling of customer consultations, significantly improving the efficiency and accuracy of power consultation services, and thus enhancing the user service experience.

[0044] In one embodiment, steps 101 to 105 are described as follows:

[0045] Step 101: Segment the power consultation text according to punctuation marks to obtain initial text fragments.

[0046] Optionally, after receiving an electricity inquiry text, the electricity customer service device first segments the text into relatively independent initial text fragments based on common punctuation marks, such as periods, question marks, exclamation marks, and semicolons. Punctuation marks are commonly used in natural language to distinguish complete semantic units. Continuing with the above embodiment, taking the electricity inquiry text "My electricity bill suddenly increased this month, what's the reason? Also, there have been frequent power outages in the community recently, what's going on?" as an example, the electricity customer service device identifies the question mark and comma in the text and segments it into two initial text fragments. Fragment 1 is "My electricity bill suddenly increased this month, what's the reason?", and fragment 2 is "Also, there have been frequent power outages in the community recently, what's going on?".

[0047] Step 102: Merge adjacent text segments in the initial text segment whose semantic coherence is greater than or equal to a preset coherence threshold to obtain a merged text segment.

[0048] Furthermore, for the initial text fragments, the electricity customer service device assesses the semantic coherence between adjacent text fragments, where semantic coherence is determined based on the text topic and context. If the semantic coherence of adjacent text fragments is greater than or equal to a preset coherence threshold, the electricity customer service device merges the adjacent text fragments into a new text fragment. For example, if two adjacent fragments both revolve around electricity bill-related issues and have a contextual relationship, the merging condition is met.

[0049] In one embodiment, a more complex electricity consultation text contains two adjacent segments: segment A is "My electricity bill this month is ridiculously high," and segment B is "Is it because I recently replaced my air conditioner with a high-powered one?" The electricity customer service device analyzes the text's theme (both revolve around the reasons for the high electricity bill) and context (segment B immediately follows segment A, speculating on the reasons for the high electricity bill). It determines that segments A and B have a high degree of semantic coherence, exceeding a preset coherence threshold. Therefore, it merges segments A and B into "My electricity bill this month is ridiculously high; is it because I recently replaced my air conditioner with a high-powered one?", resulting in the merged text segment of segments A and B.

[0050] Step 103: Identify the power keywords of each merged text segment based on the preset power keyword dictionary, and perform topic clustering based on the power keywords and semantic information of each merged text segment to obtain the topic clustering result of each merged text segment.

[0051] Furthermore, the power customer service device has a built-in preset power keyword dictionary. For each merged text segment, the power customer service device identifies the power keywords based on the power keyword dictionary.

[0052] Furthermore, the electricity customer service device performs topic clustering on merged text fragments with similar electricity keywords and semantic information based on the electricity keywords in each merged text fragment and the semantic information of the combined text. For example, if multiple merged text fragments contain keywords related to electricity bills such as "electricity bill," "electricity price," and "electricity consumption," and their semantics all revolve around electricity bill-related issues, they are clustered into one category.

[0053] In one embodiment, there are three merged text fragments: fragment C is "My electricity bill suddenly increased this month, I don't know why," fragment D is "The electricity price was recently adjusted, will it affect my electricity bill?", and fragment E is "After a power outage in the community, the electricity meter showed abnormal electricity consumption when the power was restored." The power customer service device, based on an electricity keyword dictionary, identifies that fragments C and D both contain the electricity keyword "electricity bill," and their semantics both revolve around electricity bill issues. Fragments C and D are clustered under the theme of "electricity bill issues." Fragment E, although related to electricity, has keywords and semantics that lean more towards abnormal electricity meter consumption after a power outage, and is clustered under the theme of "power outage and abnormal meter readings."

[0054] Step 104: Based on the preset related word dictionary, identify the related words associated with each merged text fragment in the power consultation text, and divide the power consultation text into structural levels based on the related words and topic clustering results of each merged text fragment to obtain the text structure level of each merged text fragment.

[0055] Furthermore, the power customer service device has a built-in preset related word dictionary. Therefore, the power customer service device identifies the related words associated with each merged text fragment in the original power consultation text, such as "in addition," "moreover," "because," "therefore," etc., based on the preset related word dictionary. These related words can reflect the logical relationship between the texts.

[0056] Furthermore, the electricity customer service device uses the related words of each merged text segment and the topic clustering results of each merged text segment to perform structural hierarchical division of the electricity consultation text, clarifying the structural position of each merged text segment in the entire consultation content and determining its text structural hierarchy. For example, two segments connected by "in addition" usually belong to a parallel structural hierarchy.

[0057] Returning to the initial example, in fragment 1, "My electricity bill suddenly increased this month, what's the reason?" and fragment 2, "Also, there have been frequent power outages in the community recently, what's going on?", the power customer service device identified the related word "also" in fragment 2. Combining this with the topic clustering results (fragment 1 is on the topic of "electricity bill problem", and fragment 2 is on the topic of "power outage problem"), it determined that these two fragments belong to a parallel structure hierarchy. The text structure hierarchy of fragment 1 is the first level in the parallel structure, and fragment 2 is the second level in the parallel structure.

[0058] Step 105: The text fragments after merging the clustering results of the same topic and the text structure hierarchy are divided into the same slice to obtain multiple consultation text fragments.

[0059] Furthermore, the power customer service device divides text fragments into the same slice after merging the clustering results of the same topic and the text structure hierarchy, thus obtaining multiple consultation text fragments. This ensures that the semantics within each consultation text fragment are closely related and that they have a clear location within the overall consultation text structure, facilitating subsequent targeted semantic understanding and processing. Based on the analysis of the previous embodiment, fragment 1, "My electricity bill suddenly increased this month, what's the reason?" and fragment 2, "Also, there have been frequent power outages in the community recently, what's going on?", have different topic clustering results ("electricity bill problem" and "power outage problem," respectively), but their text structure hierarchy is parallel. Therefore, two consultation text fragments are ultimately obtained: fragment 1 is a slice related to "electricity bill problem," and fragment 2 is a slice related to "power outage problem."

[0060] This invention can systematically decompose and classify complex power consultation texts, resulting in multiple logically clear, semantically coherent, and thematically specific consultation text fragments. These fragments can more accurately reflect different aspects of user consultations, facilitating subsequent semantic relationship analysis and identification of related power business scenarios based on AI large-scale models. This greatly improves the efficiency and accuracy of subsequent processing, enabling power customer service devices to more accurately understand user intent, thereby comprehensively and accurately handling customer consultations, significantly improving the efficiency and accuracy of power consultation services, and ultimately enhancing the user service experience.

[0061] In one embodiment, steps 201 to 205 are described as follows:

[0062] Step 201: For each consultation text fragment, perform grammatical parsing on the consultation text fragment based on the AI ​​big data model to identify words of various parts of speech in the consultation text fragment.

[0063] Optionally, for each consultation text fragment, the power customer service device uses an AI big data model to perform grammatical parsing on each consultation text fragment. In this process, the AI ​​big data model will identify words of various parts of speech in the consultation text fragment, such as nouns, verbs, adjectives, adverbs, etc.

[0064] Taking the consultation text fragment "My electricity bill suddenly increased this month" as an example, the power customer service device uses an AI model for grammatical analysis. It identifies "my home" as a noun, serving as the subject of the sentence and indicating the doer or object of the action; "this month" as a noun phrase, acting as an adverb of time to limit the time frame of the event "the electricity bill increased"; "electricity bill" as a noun, the core object of the sentence; "suddenly" as an adverb modifying the verb "increased", describing the state of the action "increased"; and "increased" as a verb, serving as the predicate of the sentence, expressing the change in the subject "my home's electricity bill".

[0065] Step 202: Based on the third semantic feature vector of each word, perform semantic relationship analysis on the consultation text fragment to identify the preliminary semantic relationship between each word in the consultation text fragment.

[0066] Furthermore, each word has a specific position in the semantic space corresponding to the consultation text fragment. This position is represented by the semantic feature vector of the word, which is determined based on the contribution degree and semantic feature value of each semantic feature in the word. Therefore, the power customer service device uses the semantic feature vector of each word to perform semantic relationship analysis on the consultation text fragment. By calculating the similarity, direction, and other relationships between vectors, it identifies the preliminary semantic relationships between each word in the consultation text fragment. For example, if the semantic feature vectors of two words are similar, they may be closely related semantically.

[0067] Continuing with the above embodiments, the semantic feature vector of "My Home" includes features related to the home, such as the place of residence and family members; the semantic feature vector of "Electricity Bill" includes features related to electricity consumption and cost calculation. Calculations reveal that the semantic feature vectors of "My Home" and "Electricity Bill" have a high degree of overlap in certain dimensions, indicating a relationship between them, namely, "My Home's" "Electricity Bill". Simultaneously, the semantic feature vectors of "Suddenly" and "Increased" are closely related in the dimension representing state change, preliminarily establishing a semantic relationship where "Suddenly" modifies "Increased".

[0068] Step 203: Adjust the preliminary semantic relationship based on the context information of each word to obtain the enhanced semantic relationship between each word in the consultation text fragment.

[0069] Furthermore, while preliminary semantic relationships provide basic connections between words, relying solely on semantic feature vectors may lead to inaccuracies. Therefore, the power customer service device further considers the contextual information of each word to adjust the preliminary semantic relationships. This contextual information includes other words before and after the word, the overall context of the sentence, and the overall theme of the consultation text in which the text segment is located. By combining this information, the semantic relationships between words can be determined more accurately, resulting in enhanced semantic relationships.

[0070] Continuing with the above example, in the text fragment "My electricity bill suddenly increased this month, what's the reason?", from a preliminary semantic perspective, "increased" and "reason" don't seem to be directly related. However, considering the context, the entire sentence is asking why the electricity bill increased. Therefore, by considering the contextual information, the power customer service device strengthens the causal semantic relationship between "increased" and "reason," that is, "increased electricity bill" is the result, and "reason" is an inquiry into the factors that led to that result.

[0071] Step 204: Based on the AI ​​big model, determine the semantic relationship direction between each word according to the vector direction of the third semantic feature vector between each word, and determine the semantic relationship level of the enhanced semantic relationship between each word according to the semantic relationship direction between each word and the enhanced semantic relationship.

[0072] Furthermore, the power customer service device again leverages a large AI model to determine the semantic relationship direction between each word based on the direction of the semantic feature vectors between each word. For example, in a subject-verb relationship, the vector direction can point from the subject to the verb, indicating that the subject performs the action of the verb. Simultaneously, by combining the enhanced semantic relationships, the semantic relationship hierarchy between each word is determined. This hierarchy reflects the primary and secondary, subordinate, and other relationships between different semantic relationships. For example, core semantic relationships such as subject-verb-object relationships are at a higher level, while modification relationships are relatively at a lower level.

[0073] Continuing with the above example, in the sentence "My family's electricity bill suddenly increased this month," from the perspective of semantic feature vector direction, the vector direction from "my family" to "increased" reflects the direction of the subject-predicate relationship, that is, "my family" is the agent of the action "increased." The vector direction from "suddenly" to "increased" reflects the direction of the modification relationship, that is, "suddenly" modifies "increased." From the perspective of semantic relationship hierarchy, the subject-predicate relationship between "my family" and "increased" is at a higher level because it constitutes the core structure of the sentence; while the modification relationship between "suddenly" and "increased" is at a lower level, which is a further description of the predicate verb in the core structure.

[0074] Step 205: Using words in each consultation text segment as graph nodes, construct edges between nodes based on the enhanced semantic relationships, semantic relationship directions, and semantic relationship levels between each word in each consultation text segment to obtain a semantic relationship graph.

[0075] Furthermore, the electricity customer service device uses words in each consultation text fragment as graph nodes, and uses the enhanced semantic relationships, semantic relationship directions, and semantic relationship levels between each word as the basis for constructing edges between nodes. Specifically, the type of edge is determined according to the semantic relationship (such as subject-predicate edge, modifier edge, etc.), the direction of the edge is determined according to the semantic relationship direction, and the weight or level of the edge can be determined according to the semantic relationship level. In this way, a semantic relationship graph that can intuitively display the semantic relationships of words in the consultation text fragment is constructed. Continuing with the above embodiment, for the text fragment "My electricity bill suddenly increased this month," "my home," "this month," "electricity bill," "suddenly," and "increased" are each a graph node. There is a subject-predicate relationship edge between “my home” and “gets taller”, pointing from “my home” to “gets taller”, and this edge is at a higher level; there is a time-modifying relationship edge between “this month” and “gets taller”, pointing from “this month” to “gets taller”, and this edge is at a lower level; there is a state-modifying relationship edge between “suddenly” and “gets taller”, pointing from “suddenly” to “gets taller”, and this edge is also at a lower level. These edges and nodes together constitute a semantic relationship graph that clearly shows the text fragment.

[0076] This invention can transform consultation text fragments into structured semantic relationship graphs. The semantic relationship graphs comprehensively and meticulously display the semantic connections between words in the text fragments, including relationship types, directions, and levels. This provides a solid data foundation for subsequently determining the target power business scenarios associated with the consultation text fragments. Therefore, the power customer service device can quickly and accurately match the corresponding power business scenarios by analyzing the key nodes and relationships in the graph, greatly improving the accuracy and efficiency of understanding consultation texts. This enables comprehensive and accurate handling of customer inquiries, significantly improving the efficiency and accuracy of power consultation services, and thus enhancing the user service experience.

[0077] In one embodiment, steps 301 to 304 are described as follows:

[0078] Step 301: Combine the first graph node in the semantic relationship graph and the second graph node in the power business knowledge graph to obtain the target graph node pair.

[0079] Optionally, based on the above embodiments, the semantic relationship graph is constructed based on consultation text fragments, displaying the semantic connections between words in the text; the power business knowledge graph contains various concepts, attributes, and relationships in the power field. Therefore, the power customer service device traverses each first graph node in the semantic relationship graph and simultaneously traverses each second graph node in the power business knowledge graph. For each node pair, the semantic similarity between the first graph node and the second graph node in the node pair is calculated. The semantic similarity calculation can be based on various methods, such as cosine similarity based on word vectors. If the semantic similarity of a node pair is greater than a preset similarity threshold, then this node pair meets the preset condition, and the first graph node and the second graph node in the node pair are combined into a target graph node pair. The preset similarity threshold is set according to actual conditions.

[0080] In one embodiment, the preset similarity threshold is 0.7. The semantic relationship graph contains a node "electricity fee," which is connected to other words through various semantic relationships, such as "increased" and "reason." The electricity business knowledge graph contains a node "electricity fee calculation." The electricity customer service device uses a word vector model to calculate the semantic similarity between "electricity fee" and "electricity fee calculation." Assuming that the cosine similarity calculation yields a similarity value of 0.8, since 0.8 > 0.7, "electricity fee" (first graph node) and "electricity fee calculation" (second graph node) meet the preset condition and form a target graph node pair.

[0081] Step 302: For each target graph node pair, perform path traversal based on the first adjacent node corresponding to the first graph node and the second adjacent node corresponding to the second graph node to obtain the initial association path.

[0082] Furthermore, for each target graph node pair, the power customer service device obtains the first adjacent node corresponding to the first graph node and the second adjacent node corresponding to the second graph node, where an adjacent node refers to a node that is directly connected to the current node through some relationship.

[0083] Furthermore, the power customer service device performs path traversal on the first and second adjacent nodes. During the traversal, only when the semantic similarity between the first and second adjacent nodes is greater than a preset similarity threshold (i.e., the preset condition is met) are these two adjacent nodes included in the initial association path. By continuously finding the next-level adjacent nodes that meet the conditions from the current adjacent node, an initial association path is gradually constructed, starting from the first graph node, passing through a series of adjacent nodes that meet the conditions, and finally reaching the second graph node. Therefore, it can be understood that the initial association path is composed of the first and second adjacent nodes that meet the preset conditions.

[0084] In one embodiment, for the target graph node pair "Electricity Fee" (first graph node) and "Electricity Fee Calculation" (second graph node) obtained earlier, in the semantic relationship graph, the first adjacent node of "Electricity Fee" is "Increased," and in the power business knowledge graph, the second adjacent node of "Electricity Fee Calculation" is "Electricity Consumption." The power customer service device calculates the semantic similarity between "Increased" and "Electricity Consumption," assuming a similarity value of 0.75 (greater than the preset similarity threshold of 0.7). Continuing from "Increased" and "Electricity Consumption," the adjacent node of "Increased" is "Cause," and the adjacent node of "Electricity Consumption" is "Abnormal Increase," and the semantic similarity between "Cause" and "Abnormal Increase" is calculated, assuming a value of 0.8. Thus, by continuously traversing adjacent nodes that meet the conditions, an initial association path is obtained: Electricity Fee - Increased - Cause (semantic relationship graph) and Electricity Fee Calculation - Electricity Consumption - Abnormal Increase (power business knowledge graph).

[0085] Step 303: Determine the target association path based on the semantic consistency between the first semantic relation of the initial association path in the semantic relation graph and its second semantic relation in the power business knowledge graph.

[0086] Furthermore, for each initial association path, the power customer service device compares its first semantic relationship in the semantic relationship graph with its second semantic relationship in the power business knowledge graph. Semantic relationships include causal relationships, parallel relationships, and subordinate relationships. If the two semantic relationships are semantically consistent, then this initial association path may become the target association path. For example, in the semantic relationship graph, "electricity cost" and "increased" have a subject-predicate relationship; in the power business knowledge graph, "electricity cost calculation" and "increased electricity consumption" also have a causal relationship, meaning that increased electricity consumption affects electricity cost calculation. Here, the semantic relationships are semantically related and consistent. The device performs this consistency judgment on each pair of corresponding semantic relationships along the entire initial association path. Only when all corresponding semantic relationships meet the consistency requirement is the initial association path determined as the target association path.

[0087] In one embodiment, the initial association path obtained above is: Electricity Bill - Increase - Cause (semantic relationship graph) and Electricity Bill Calculation - Electricity Consumption - Abnormal Increase (power business knowledge graph). In the semantic relationship graph, "Electricity Bill" and "Increase" have a subject-predicate relationship, and "Increase" and "Cause" have a causal relationship. In the power business knowledge graph, "Electricity Bill Calculation" and "Electricity Consumption" have a calculation basis relationship, and "Electricity Consumption" and "Abnormal Increase" have a state description relationship, and "Abnormal Increase in Electricity Consumption" and "Change in Electricity Bill Calculation" have a causal relationship. Overall, the semantic relationships in the semantic relationship graph and the power business knowledge graph are semantically consistent, so this initial association path is determined as the target association path.

[0088] Step 304: Based on the position of each node in the target association path in the power business knowledge graph, determine the target power business scenario for each consultation text fragment.

[0089] Furthermore, the electricity customer service device examines the position of each node in the target association path within the electricity business knowledge graph. Nodes in the electricity business knowledge graph typically correspond to different electricity business concepts, processes, or scenarios. Therefore, by analyzing the positions of nodes in the target association path and the relationships between them, the target electricity business scenario associated with the consultation text fragment can be determined. For example, if the key nodes in the target association path are concentrated under the "electricity bill anomaly handling" branch in the electricity business knowledge graph, then the target electricity business scenario can be determined to be the electricity bill anomaly handling scenario, as described in steps 3041 to 3044.

[0090] This invention enables deep association between a semantic relationship graph constructed based on consultation text and a power business knowledge graph. Therefore, it can accurately determine the corresponding target power business scenario for each consultation text fragment, moving beyond simple keyword matching on the surface of the text. Thus, when faced with a large number of complex and diverse power consultation texts, it can quickly and accurately classify different consultations into the corresponding business scenarios, laying a solid foundation for providing targeted responses. This significantly improves the efficiency and accuracy of power customer service in handling consultations, enabling comprehensive and accurate processing of customer inquiries and significantly enhancing the efficiency and accuracy of power consultation services, thereby improving the user service experience.

[0091] In one embodiment, steps 3041 to 3044 are described as follows:

[0092] Step 3041: Obtain the first target node in the target association path.

[0093] Optionally, the power customer service device begins searching for nodes that meet specific conditions within the target association path. The power business knowledge graph has a certain structure, containing central nodes that typically represent core power business concepts or processes. Therefore, the device determines the position of each node in the target association path within the power business knowledge graph, checking if it falls within a preset range of the central nodes. This preset range can be defined based on the graph's topology, such as the range of nodes reachable from the central node after a certain number of hops (e.g., 2-3 hops). Further, the power customer service device identifies nodes in the target association path that meet this condition as the first target node.

[0094] In one embodiment, the central node of the power business knowledge graph is "Power Service," and the preset range is two hops. In the previously determined target association path, there is a node "Electricity Bill Calculation." Starting from "Power Service," and passing through the intermediate node "Electricity Cost Management," "Electricity Bill Calculation" can be reached within two hops. Therefore, "Electricity Bill Calculation" is identified as the first target node by the power customer service device.

[0095] Step 3042: Determine the business scenarios in the power business knowledge graph that have a relationship link with the first target node as the initial power business scenarios.

[0096] Furthermore, the power customer service device searches the power business knowledge graph for nodes that are linked to the first target node. These links can be of various types, such as causal relationships, subordinate relationships, and upstream / downstream process relationships. Therefore, it can be understood that the nodes connected to the first target node through these links often represent specific power business scenarios. Consequently, the power customer service device identifies these business scenarios as the initial power business scenarios.

[0097] Continuing with the above embodiments, for the previously determined first target node "Electricity Bill Calculation," in the power business knowledge graph, the nodes linked to "Electricity Bill Calculation" are "Electricity Bill Anomaly Handling" (because the results of electricity bill calculation may be abnormal, there is a causal relationship) and "Electricity Bill Generation" (Electricity Bill Calculation is a prerequisite process for Electricity Bill Generation, there is an upstream and downstream relationship). The power customer service device identifies "Electricity Bill Anomaly Handling" and "Electricity Bill Generation" as the initial power business scenario.

[0098] Step 3043: Sort the initial power business scenarios based on the semantic fit between the first semantic feature vector of the consultation text fragment and the second semantic feature vector of each initial power business scenario to obtain the sorted power business scenarios.

[0099] Furthermore, to more accurately identify the target business scenario from the initial power business scenarios, each initial power business scenario will be analyzed in depth. The power customer service device extracts the first semantic feature vector from the consultation text fragment, which can be obtained through natural language processing techniques such as word vector models and text embedding models. This vector can comprehensively reflect the semantic information of the consultation text fragment. At the same time, the power customer service device extracts the second semantic feature vector for each initial power business scenario. This vector is obtained based on the processing of information such as text descriptions and knowledge definitions related to the business scenario, and represents the semantic characteristics of the business scenario.

[0100] Furthermore, the power customer service device calculates the semantic fit between the first semantic feature vector of the consultation text fragment and the second semantic feature vector of each initial power business scenario. The semantic fit can be calculated using various methods, such as cosine similarity and Euclidean distance. Based on the calculated semantic fit, the power customer service device ranks the initial power business scenarios, with higher semantic fit scenarios having higher priority in the ranking.

[0101] Continuing with the above embodiments, for the inquiry text fragment "My electricity bill suddenly increased this month, what's the reason?", the power customer service device obtains its first semantic feature vector through a natural language processing model. For the initial power business scenario "electricity bill anomaly handling", based on the knowledge description of this business scenario (such as including analysis of the reasons for the electricity bill anomaly, the anomaly handling process, etc.), its second semantic feature vector is obtained through text processing. The semantic fit between the two is calculated using cosine similarity, resulting in a score of 0.85. For another initial power business scenario "electricity bill generation", the same processing yields a semantic fit of 0.4 with the inquiry text fragment. Sorted from highest to lowest semantic fit, the order is: "electricity bill anomaly handling" first, followed by "electricity bill generation".

[0102] Step 3044: The highest priority power business scenario among the sorted power business scenarios is determined as the target power business scenario for each consultation text fragment.

[0103] Furthermore, the power customer service device identifies the highest-priority power business scenario among the sorted power business scenarios as the target power business scenario for each inquiry text fragment. This highest-priority scenario, being the most semantically compatible with the inquiry text fragment, is most likely the actual business scenario involved in the user's inquiry. In this way, the device accurately matches the inquiry text fragment with the corresponding power business scenario, preparing for providing a targeted response.

[0104] In one embodiment, based on the preceding ranking results, "Electricity Bill Anomaly Handling" has the highest priority among the ranked electricity service scenarios. Therefore, the electricity customer service device determines the target electricity service scenario of the inquiry text fragment "My electricity bill suddenly increased this month, what is the reason?" as the "Electricity Bill Anomaly Handling" scenario.

[0105] This invention, starting from the target association path, accurately determines the target power business scenario corresponding to each consultation text fragment through multiple steps of screening and analysis. This avoids the limitations of matching business scenarios based solely on surface keywords or simple relationships. By combining the semantic features of the consultation text, it achieves more intelligent and accurate business scenario matching, greatly improving the accuracy of power customer service in handling consultations. This allows customer service representatives to quickly locate the business area to which the user's problem belongs, thereby enabling comprehensive and accurate handling of customer consultations. This significantly improves the efficiency and accuracy of power consultation services, thereby enhancing the user service experience.

[0106] In one embodiment, steps 401 to 404 are described as follows:

[0107] Step 401: Classify the logical connections between each consultation text fragment to obtain the logical connection category between each consultation text fragment, and construct a logical connection graph with each consultation text fragment as a node and the logical connection category between each consultation text fragment as a node edge.

[0108] Optionally, the power customer service device analyzes multiple consultation text fragments, identifies the logical connections between each fragment, and categorizes each identified logical connection into a corresponding logical connection category. These logical connections may include causal relationships, parallel relationships, progressive relationships, and adversative relationships. For example, if one consultation text fragment describes a power outage event and another fragment describes the damage to electrical appliances caused by the power outage, then these two fragments have a causal relationship.

[0109] Furthermore, the power customer service device uses each consultation text fragment as a node and the logical relationship categories between each consultation text fragment as node edges to construct a logical relationship graph.

[0110] In one embodiment, the power customer service device receives three inquiry text fragments: Fragment A, "My electricity bill suddenly increased this month, what's the reason?"; Fragment B, "Recently, there have been frequent power outages in the community, causing a lot of inconvenience to our lives"; and Fragment C, "The food in my refrigerator spoiled after the power outage." Analysis reveals a causal relationship between fragments B and C, as the power outage caused the food spoilage. Fragment A has no direct logical relationship with fragments B and C, but is considered a parallel relationship. Therefore, in the constructed logical connection graph, fragments A, B, and C are nodes, the edge between fragment B and fragment C is labeled as a "causal relationship," and the edge between fragment A and fragments B and C is labeled as a "parallel relationship."

[0111] Step 402: For each second target node in the logical connection graph, perform path traversal in the logical connection graph based on the edges of each node to determine the number of paths that start from the second target node and / or end at the second target node.

[0112] Furthermore, for each node in the logical connection graph (i.e. the second target node), the power customer service device performs path traversal in the logical connection graph according to the edges of each node. Path traversal is to find all possible paths from one node to other nodes in the graph along the edges of the nodes.

[0113] Furthermore, for each second target node, the power customer service device determines the number of paths originating from and / or ending at that node. The number of paths reflects the importance and connectivity of that node within the overall logical connection graph. If a node has a large number of paths connecting it, it indicates that it plays a crucial role in the logical connections and its logical relationships with other segments are more complex and diverse.

[0114] In one embodiment, for segment B in the above logical connection diagram ("Recently, the community has experienced frequent power outages, causing a lot of inconvenience to our lives"), starting from segment B, there is a path leading to segment C via a "causal relationship" edge; ending at segment B, there is no path (assuming reverse paths are not considered in the current analysis). Therefore, the number of paths for segment B is 1. For segment C ("After the power outage, the food in my refrigerator spoiled"), ending at segment C, there is a path from segment B via a "causal relationship" edge; starting from segment C, there is no path. Therefore, the number of paths for segment C is also 1. For segment A ("My electricity bill suddenly increased this month, what is the reason?"), starting from segment A, there are two paths leading to segments B and C respectively via "parallel relationship" edges; ending at segment A, there are also two paths. Therefore, the number of paths for segment A is 4.

[0115] Step 403: Identify consultation text fragments with a path number greater than a preset number as key text fragments.

[0116] Furthermore, a preset number is set as the judgment criterion. The power customer service device identifies consultation text fragments with a path number greater than the preset number as key text fragments. Among them, key text fragments occupy an important position in the logical structure of the entire consultation content. Their logical connections with other fragments are richer and closer, and they play a key role in understanding the user's overall consultation intent.

[0117] Continuing with the above embodiment, the preset number is 2. Based on the previously calculated number of paths, the number of paths for segment A is 4, which is greater than the preset number of 2; the number of paths for segments B and C is 1, which is less than the preset number of 2. Therefore, the power customer service device determines that segment A is the key text segment.

[0118] Step 404: Based on the key text fragments and the textual semantic topics of each consultation text fragment, perform cross-fragment semantic integration on each consultation text fragment to obtain cross-fragment semantic integrated text.

[0119] Furthermore, the power customer service device uses identified key text fragments as its core and performs cross-fragment semantic integration based on the semantic theme of each consultation text fragment. Specifically, the device analyzes the semantic theme of each consultation text fragment; for example, fragment A's semantic theme revolves around abnormal electricity bills, fragment B's semantic theme revolves around power outages in the community, and fragment C's semantic theme revolves around food damage in the refrigerator caused by power outages. Further, guided by the semantics of the key text fragments, the device integrates the semantics of other related fragments according to logical connections. During the integration process, the text undergoes appropriate linguistic organization and adjustments to form a coherent and complete cross-fragment semantically integrated text, thereby clearly presenting the user's overall consultation intent, as described in steps 4041 to 4043.

[0120] This invention constructs a logical connection diagram, analyzes the number of paths to determine key text fragments, and then integrates them with semantic themes. This avoids simple text splicing and enables accurate identification of multiple complex consultation intentions in user consultation texts. As a result, customer consultations can be processed comprehensively and accurately, significantly improving the efficiency and accuracy of power consultation services and thus enhancing the user service experience.

[0121] In one embodiment, steps 4041 to 4043 are described as follows:

[0122] Step 4041: For any two target consultation text fragments, determine the semantic theme relevance of the text semantic theme of the target consultation text fragments under the logical connection between the target consultation text fragments.

[0123] Optionally, when performing cross-segment semantic integration, the power customer service device analyzes any two target consultation text segments. For each consultation text segment, natural language processing techniques and semantic analysis models are used to determine its semantic theme. For example, through keyword extraction, LDA topic modeling, and other methods, the main power-related themes of each segment are identified, such as electricity billing issues, power outages, and equipment maintenance.

[0124] Furthermore, the power customer service device calculates the semantic topic relevance between two target consultation text fragments based on the established logical connections (such as causal, parallel, or progressive relationships) between their text semantic topics. The calculation of semantic topic relevance can be based on a semantic similarity algorithm, with weighted adjustments made according to the characteristics of the logical connections. For example, for two fragments with a causal relationship, if one fragment's topic is "power outage" and the other's is "appliance malfunction," the causal relationship will be given a higher weight when calculating the relevance, resulting in a relatively high relevance between them.

[0125] In one embodiment, there are two consultation text fragments: fragment D "Power outage caused by transformer failure in the community" and fragment E "My computer data was lost after the power outage." First, the semantic topic of fragment D is determined to be "cause of power outage (transformer failure)," and the semantic topic of fragment E is "consequences of power outage (loss of computer data)." Since these two fragments have a causal logical connection, a semantic similarity algorithm (such as cosine similarity) is used to calculate the similarity between the two topics, and a weight is assigned based on the causal relationship. For example, the unweighted cosine similarity is 0.6; after considering the causal relationship weight (set to 1.5), the semantic topic relevance is 0.6 * 1.5 = 0.9.

[0126] Step 4042: Based on the position of the key text fragment in the logical connection diagram and the semantic topic relevance between the key text fragment and other consultation text fragments, construct a cross-fragment semantic integration framework; other consultation text fragments are consultation text fragments other than the key text fragment.

[0127] Furthermore, the power customer service device identifies the location of key text fragments in the logical connection diagram. These key text fragments often occupy a central position or connect multiple important paths, playing a crucial guiding role in the overall semantic integration. Therefore, the power customer service device analyzes the semantic thematic relevance between key text fragments and other consultation text fragments (i.e., fragments other than the key text fragments).

[0128] Furthermore, the power customer service device constructs a cross-segment semantic integration framework based on the semantic thematic relevance between key text fragments and other consultation text fragments. This framework centers on the key text fragment and determines the connection methods and hierarchical structure between other text fragments and the key text fragment based on semantic thematic relevance and logical connections. For example, if a non-key text fragment has a high semantic thematic relevance and close logical connection with a key text fragment, it will be placed closer to the key text fragment in the framework and connected through corresponding logical relationship edges. The cross-segment semantic integration framework can be represented using a tree structure or a graph structure, clearly displaying the semantic and logical relationships between the various text fragments.

[0129] In one embodiment, the key text fragment F is "My electricity bill has increased abnormally this month," and other inquiry text fragments include fragment G "I added a high-power appliance this month" and fragment H "My neighbor also reported that their electricity bill has increased." Key text fragment F is centrally located in the logical connection diagram because it represents the core issue of the user's inquiry. The semantic topic correlation between fragment F and fragment G is calculated to be 0.8 (due to the possibility that adding a high-power appliance might lead to abnormal electricity bills, a causal relationship exists), and the semantic topic correlation between fragment F and fragment H is 0.6 (both revolve around the issue of increased electricity bills, indicating a parallel relationship). Therefore, in the reconstructed cross-fragment semantic integration framework, fragment F is central, fragment G is closely connected to fragment F via a "causal relationship" edge, and fragment H is connected to fragment F via a "parallel relationship" edge. Furthermore, fragment G is structurally closer to fragment F because of its higher correlation.

[0130] Step 4043: Based on the cross-fragment semantic integration framework, each consultation text fragment is subjected to cross-fragment semantic integration to obtain cross-fragment semantic integrated text.

[0131] Furthermore, based on the constructed cross-segment semantic integration framework, the power customer service device begins to perform cross-segment semantic integration on each consultation text segment. Specifically, starting from the key text segment, and following the connection methods and hierarchical structure determined in the framework, the semantics of other related text segments are sequentially integrated. During the integration process, the power customer service device organizes and adjusts the language of the text to ensure that the integrated text is semantically coherent and logically sound. For example, appropriate conjunctions such as "because...therefore...", "at the same time," and "in addition" are added according to logical relationships. In this way, the various independent consultation text segments are integrated into a complete and clear cross-segment semantically integrated text that reflects the user's consultation intent.

[0132] Continuing with the above embodiments, based on the cross-segment semantic integration framework constructed earlier, the power customer service device starts with the key text segment F, "My electricity bill has increased abnormally this month." Since segment G, "A high-power appliance was added this month," is closely linked to F through a causal relationship, it is integrated into "My electricity bill has increased abnormally this month because a high-power appliance was added this month." Then, segment H, "My neighbor also reported that their electricity bill has increased," is added through a parallel relationship, finally resulting in the cross-segment semantically integrated text, "My electricity bill has increased abnormally this month because a high-power appliance was added this month, and my neighbor also reported that their electricity bill has increased."

[0133] This invention constructs an integration framework by quantifying the semantic topic relevance, and then integrates text based on the framework. This allows the integrated cross-fragment semantic text to highly reproduce the user's complex consultation intent, enabling a quick and accurate understanding of the full picture of the user's consultation. It avoids misunderstandings caused by fragmented segments, thus enabling comprehensive and accurate handling of customer consultations. This significantly improves the efficiency and accuracy of power consultation services, thereby enhancing the user service experience.

[0134] Furthermore, the semantic analysis device for power customer service based on AI large model provided by the present invention will be described below. The semantic analysis device for power customer service based on AI large model described below can be referred to in correspondence with the semantic analysis method for power customer service based on AI large model described above.

[0135] Optional, refer to Figure 2 , Figure 2 This is a schematic diagram of the structure of the AI-based large-scale model-based power customer service semantic analysis device provided by the present invention. The AI-based large-scale model-based power customer service semantic analysis device includes...

[0136] The text slicing module 210 is used to slice the received power consultation text based on natural language to obtain multiple consultation text fragments;

[0137] The semantic relationship parsing module 220 is used to perform semantic relationship parsing on each consultation text fragment based on the semantic understanding capabilities of the AI ​​large model, and to construct a semantic relationship graph based on the semantic relationships between words in each consultation text fragment;

[0138] The business scenario association module 230 is used to determine the target power business scenario associated with each consultation text fragment based on the semantic relationship graph and the power business knowledge graph.

[0139] The text integration module 240 is used to perform cross-fragment semantic integration on each consultation text fragment based on the logical connection between each consultation text fragment, so as to obtain cross-fragment semantically integrated text.

[0140] The intent recognition output module 250 is used to output response text based on cross-fragment semantic integration text and target power business scenario.

[0141] This invention, through the powerful semantic understanding capabilities of a large AI model, can adapt to various expression methods. As long as the semantics are the same or similar, it can accurately understand the intent. Furthermore, through text segmentation, semantic relationship parsing, and cross-segment semantic integration, it can deeply understand the complex semantics in power consultation texts. At the same time, through the identification of related business scenarios and cross-segment semantic integration, it can clearly sort out the relationships between different business scenarios, thereby accurately identifying multiple complex consultation intents in user consultation texts. This enables comprehensive and accurate handling of customer consultations, significantly improving the efficiency and accuracy of power consultation services, and thus enhancing the user service experience.

[0142] Please see Figure 3 , Figure 3 An embodiment diagram of an electronic device provided in accordance with the present invention. For example... Figure 3 As shown, this embodiment of the invention provides an electronic device 300, including a memory 310, a processor 320, and a computer program 311 stored in the memory 310 and executable on the processor 320. When the processor 320 executes the computer program 311, it performs the following steps:

[0143] The received electricity consultation text is sliced ​​based on natural language processing to obtain multiple consultation text fragments;

[0144] Based on the semantic understanding capabilities of the AI ​​large model, semantic relationships are analyzed for each consultation text fragment, and a semantic relationship graph is constructed based on the semantic relationships between words in each consultation text fragment;

[0145] The target power business scenario associated with each consultation text fragment is determined based on the semantic relationship graph and the power business knowledge graph.

[0146] Based on the logical connections between each consultation text fragment, cross-fragment semantic integration is performed on each consultation text fragment to obtain cross-fragment semantic integrated text;

[0147] The response text is output based on cross-fragment semantic integration of text and target power business scenario.

[0148] Please see Figure 4 , Figure 4 An embodiment diagram of a computer-readable storage medium provided in accordance with an embodiment of the present invention is shown. Figure 4 As shown, this embodiment provides a computer-readable storage medium 400 on which a computer program 311 is stored. When the computer program 311 is executed by a processor, it performs the following steps:

[0149] The received electricity consultation text is sliced ​​based on natural language processing to obtain multiple consultation text fragments;

[0150] Based on the semantic understanding capabilities of the AI ​​large model, semantic relationships are analyzed for each consultation text fragment, and a semantic relationship graph is constructed based on the semantic relationships between words in each consultation text fragment;

[0151] The target power business scenario associated with each consultation text fragment is determined based on the semantic relationship graph and the power business knowledge graph.

[0152] Based on the logical connections between each consultation text fragment, cross-fragment semantic integration is performed on each consultation text fragment to obtain cross-fragment semantic integrated text;

[0153] The response text is output based on cross-fragment semantic integration of text and target power business scenario.

[0154] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer is able to execute the AI-based large-model-based power customer service semantic analysis method provided by the above methods. This method includes:

[0155] The received electricity consultation text is sliced ​​based on natural language processing to obtain multiple consultation text fragments;

[0156] Based on the semantic understanding capabilities of the AI ​​large model, semantic relationships are analyzed for each consultation text fragment, and a semantic relationship graph is constructed based on the semantic relationships between words in each consultation text fragment;

[0157] The target power business scenario associated with each consultation text fragment is determined based on the semantic relationship graph and the power business knowledge graph.

[0158] Based on the logical connections between each consultation text fragment, cross-fragment semantic integration is performed on each consultation text fragment to obtain cross-fragment semantic integrated text;

[0159] The response text is output based on cross-fragment semantic integration of text and target power business scenario.

[0160] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0161] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the parts that contribute to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0162] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An AI large model-based power customer service semantic analysis method, characterized in that, The method comprises the following steps: slicing the received power consultation text based on natural language to obtain multiple consultation text segments; performing semantic relationship analysis on each consultation text segment based on the semantic understanding capability of an AI large model, and constructing a semantic relationship graph based on the semantic relationship between words in each consultation text segment; determining the target power business scenario associated with each consultation text segment based on the semantic relationship graph and a power business knowledge graph; performing cross-segment semantic integration on each consultation text segment based on the logical connection between each consultation text segment to obtain cross-segment semantic integration text; outputting a reply text based on the cross-segment semantic integration text and the target power business scenario; wherein the determination of the target power business scenario associated with each consultation text segment based on the semantic relationship graph and the power business knowledge graph comprises: combining a first graph node in the semantic relationship graph and a second graph node in the power business knowledge graph to obtain a target graph node pair; the first graph node and the second graph node satisfy a preset condition; the preset condition is that the semantic similarity between the nodes is greater than a preset similarity threshold; for each target graph node pair, performing path traversal based on the first adjacent node corresponding to the first graph node and the second adjacent node corresponding to the second graph node to obtain an initial associated path; the initial associated path is composed of the first adjacent node and the second adjacent node that satisfy the preset condition; determining a target associated path based on the semantic consistency between the first semantic relationship of the initial associated path in the semantic relationship graph and the second semantic relationship of the initial associated path in the power business knowledge graph; determining the target power business scenario of each consultation text segment based on the position of each node in the target associated path in the power business knowledge graph; the cross-segment semantic integration of each consultation text segment based on the logical connection between each consultation text segment to obtain cross-segment semantic integration text comprises: classifying the logical connection between each consultation text segment to obtain the logical connection category between each consultation text segment, and constructing a logical connection graph with each consultation text segment as a node and the logical connection category between each consultation text segment as a node edge; for each second target node in the logical connection graph, performing path traversal based on the node edges in the logical connection graph to determine the path quantity of the path starting from or / and ending at the second target node; determining the consultation text segment with a path quantity greater than a preset quantity as a key text segment; performing cross-segment semantic integration on each consultation text segment based on the key text segment and the text semantic theme of each consultation text segment to obtain the cross-segment semantic integration text. 2.The AI large model-based power customer service semantic analysis method of claim 1, wherein, the determination of the target power business scenario of each consultation text segment based on the position of each node in the target associated path in the power business knowledge graph comprises: acquire a first target node in the target association path; the first target node is within a preset range of a graph center node of the power business knowledge graph; determine a business scenario in the power business knowledge graph that has a relationship link with the first target node as an initial power business scenario; sort the initial power business scenarios based on a semantic fit degree between a first semantic feature vector of the consultation text segment and a second semantic feature vector of each initial power business scenario, to obtain sorted power business scenarios; determine a power business scenario with the highest priority in the sorted power business scenarios as a target power business scenario of each consultation text segment. 3.The AI large model-based power customer service semantic analysis method of claim 1, wherein, the cross-segment semantic integration text is obtained by performing cross-segment semantic integration on each consultation text segment based on the key text segment and the text semantic topic of each consultation text segment, including: for any two target consultation text segments, determine a semantic topic association degree of the text semantic topic of the target consultation text segments under the logical connection between the target consultation text segments; based on the position of the key text segment in the logical connection graph and the semantic topic association degree between the key text segment and other consultation text segments, construct a cross-segment semantic integration framework; the other consultation text segments are consultation text segments other than the key text segment; based on the cross-segment semantic integration framework, perform cross-segment semantic integration on each consultation text segment to obtain the cross-segment semantic integration text. 4.The AI large model-based power customer service semantic analysis method of claim 1, wherein, the semantic relationship graph is constructed according to the semantic relationship between the words in each consultation text segment based on the semantic understanding ability of the AI large model, including: for each consultation text segment, based on the AI large model, perform syntax analysis on the consultation text segment to identify words of various parts of speech in the consultation text segment; based on the third semantic feature vector of each word, perform semantic relationship analysis on the consultation text segment to identify the preliminary semantic relationship between each word in the consultation text segment; the third semantic feature vector of each word represents the position of each word in the semantic space corresponding to the consultation text segment, which is determined based on the contribution degree and semantic feature value of each semantic feature in the word; based on the context information of each word, adjust the preliminary semantic relationship to obtain the strengthened semantic relationship between each word in the consultation text segment; based on the AI large model, determine the semantic relationship direction between each word according to the vector direction of the third semantic feature vector between each word, and determine the semantic relationship level of the strengthened semantic relationship between each word according to the semantic relationship direction and the strengthened semantic relationship between each word; the semantic relationship graph is obtained by taking the words in each consultation text segment as graph nodes, and constructing edges between nodes based on the strengthened semantic relationship, semantic relationship direction and semantic relationship level between each word in each consultation text segment. 5.The AI large model-based power customer service semantic analysis method according to any one of claims 1 to 4, characterized in that, the received power consultation text is sliced based on natural language to obtain a plurality of consultation text segments, including: The power consultation text is segmented according to punctuation to obtain initial text segments; Adjacent text segments in the initial text segments with a semantic coherence degree greater than or equal to a preset coherence degree threshold are merged to obtain merged text segments; the semantic coherence degree is determined according to the text theme and text context between adjacent text segments; Power keywords of each merged text segment are identified based on a preset power keyword dictionary, and theme clustering of each merged text segment is performed based on the power keywords and semantic information of each merged text segment to obtain a theme clustering result of each merged text segment; Based on a preset association word dictionary, the association words associated with each merged text segment in the power consultation text are identified, and the power consultation text is divided into a text structure level based on the association words and the theme clustering result of each merged text segment to obtain a text structure level of each merged text segment; Merged text segments with the same theme clustering result and text structure level are divided into the same segment to obtain a plurality of consultation text segments. 6.A device for AI large model-based power customer service semantic analysis, characterized in that, The application is applied to the AI large model-based power customer service semantic analysis method in any one of claims 1 to 5; the AI large model-based power customer service semantic analysis device comprises: A text slicing module for slicing the received power consultation text based on natural language to obtain a plurality of consultation text segments; A semantic relationship analysis module for analyzing the semantic relationship of each consultation text segment based on the semantic understanding ability of the AI large model, and constructing a semantic relationship graph based on the semantic relationship between words in each consultation text segment; A business scenario association module for determining the target power business scenario associated with each consultation text segment based on the semantic relationship graph and the power business knowledge graph; A text integration module for cross-segment semantic integration of each consultation text segment based on the logical connection between each consultation text segment to obtain cross-segment semantic integration text; An intent recognition output module for outputting a reply text based on the cross-segment semantic integration text and the target power business scenario; The determination of the target power business scenario associated with each consultation text segment based on the semantic relationship graph and the power business knowledge graph comprises: Combining a first graph node in the semantic relationship graph and a second graph node in the power business knowledge graph to obtain a target graph node pair; the first graph node and the second graph node satisfy a preset condition; the preset condition is that the semantic similarity between the nodes is greater than a preset similarity threshold; For each target graph node pair, a path traversal is performed based on a first adjacent node corresponding to the first graph node and a second adjacent node corresponding to the second graph node to obtain an initial association path; the initial association path is composed of the first adjacent node and the second adjacent node that satisfy the preset condition; A target association path is determined based on the semantic consistency between the first semantic relationship of the initial association path in the semantic relationship graph and the second semantic relationship of the initial association path in the power business knowledge graph. determine a target power service scenario of each consultation text segment based on a position of each node in the target association path in the power service knowledge graph; the cross-segment semantic integration text is obtained by performing cross-segment semantic integration on each consultation text segment based on logical connections between the consultation text segments, including: classifying the logical connections between each consultation text segment to obtain a logical connection category between each consultation text segment, and constructing a logical connection graph by taking each consultation text segment as a node and taking the logical connection category between each consultation text segment as a node edge; for each second target node in the logical connection graph, performing path traversal on each node edge in the logical connection graph to determine a path quantity of a path starting from or / and ending at the second target node; determining a key text segment from the consultation text segments with a path quantity greater than a preset quantity; performing cross-segment semantic integration on each consultation text segment based on the key text segment and a text semantic topic of each consultation text segment to obtain the cross-segment semantic integration text.

7. An electronic device, comprising: a memory for storing a computer software program; a processor for reading and executing the computer software program, wherein the processor, when executing the computer software program, implements the power customer service semantic analysis method based on an AI large model according to any one of claims 1 to 5.

8. A non-transitory computer readable storage medium having stored therein a computer software program, characterized in that, the computer software program, when executed by the processor, implements the power customer service semantic analysis method based on an AI large model according to any one of claims 1 to 5.

Images