Method and terminal for embedding exclusive AI interpretation for shared documents

Abstract

The present invention provides a method and terminal for implanting exclusive AI interpretation for shared documents, the method for implanting exclusive AI interpretation for shared documents comprising: receiving a document forwarded by a forwarder on a chat tool, extracting the text content of the document, calling an AI knowledge base model to analyze the text content of the document, and obtaining key information; generating a summary of the text content according to a generative summary algorithm, identifying the theme of the text content using an LDA algorithm, and performing dependency syntax analysis on the text content to extract the main information of the sentences in the text content; integrating the key information, summary, theme, and main information of the sentences, generating an interpretation result of the document, and sending the interpretation result to the receiver. The present invention enables the receiver to quickly obtain the core content and structure of the document without reading the full text, thereby improving the efficiency of document processing and information transmission.

Description

Technical Field

[0001] The present invention relates to the field of document processing technology, and in particular to a method and a terminal for implanting exclusive AI interpretation into shared documents. Background Art

[0002] In the current information age, the creation and sharing of documents have become very frequent, especially through email, instant messaging, social media and other digital platforms. However, as the number and length of documents increase, when the forwarder forwards the document to the recipient, the recipient may encounter information overload when faced with a large amount of text, and it is difficult to quickly identify and extract key information from a large amount of text, affecting the efficiency of interaction with the forwarder.

[0003] In the technical solution with application number CN202311218741.2, although it can automatically classify and store texts according to the trained general language model, thereby improving document management efficiency, it is unable to quickly identify documents and extract key information.

[0004] In the technical solution with application number CN202211085411.6, although it realizes the function of setting document forwarding tasks in online documents, allowing users to conduct multi-person circulation of online documents without relying on third-party management tools, thereby improving the convenience of online document sending, it also fails to realize automatic interpretation of forwarded documents, affecting the interaction efficiency between recipients and forwarders. Summary of the invention

[0005] The present invention provides a method and a terminal for embedding exclusive AI interpretation into shared documents, so as to quickly identify and interpret key information of forwarded documents, thereby improving the interaction efficiency between receivers and forwarders.

[0006] To solve the above problems, the present invention adopts the following technical solutions:

[0007] The present invention provides a method for embedding exclusive AI interpretation into a shared document, comprising:

[0008] Receive documents forwarded by the forwarder on the chat tool;

[0009] Extract the text content of the document, call the AI ​​knowledge base model to analyze the text content of the document, and obtain key information;

[0010] Generate a summary of the text content according to a generative summary algorithm;

[0011] Using the LDA algorithm to identify the subject of the text content, and performing dependency syntactic analysis on the text content to extract the main information of the sentences in the text content;

[0012] After integrating the key information, abstract, theme and main information of the sentence, an interpretation result of the document is generated and sent to the receiver.

[0013] Furthermore, after sending the interpretation result to the receiver, the method further includes:

[0014] Obtaining the recipient's question information about the document;

[0015] Performing vectorization processing on the question information to obtain a vector corresponding to the question information;

[0016] Calling the AI ​​knowledge base model to analyze and process the vector corresponding to the question information, and screening out the answer information whose vector corresponding to the question information has a similarity greater than a preset degree;

[0017] The reply information is sent to the recipient.

[0018] Furthermore, after sending the reply information to the recipient, the method further includes:

[0019] When it is determined that the receiver is a target customer in the customer list, after the receiver finishes asking all questions, all interaction contents with the receiver are obtained;

[0020] Analyze all interactive contents to determine the interest preferences of the recipient;

[0021] Query product information that matches the interest preference, and send the product information and all interactive content that matches the interest preference to the sender.

[0022] Preferably, querying product information matching the interest preference includes:

[0023] determining a user tag describing the interest preferences of the recipient;

[0024] Vectorize the user label to obtain a user label vector;

[0025] Extract product tags describing product information from the knowledge base;

[0026] Vectorize the product label to obtain a product label vector;

[0027] Calculate the cosine distance between the user label vector and the product label vector, and use the product label vector whose cosine distance with the user label vector is greater than a preset cosine distance as the target product label vector;

[0028] Determine the product information corresponding to the target product label vector, and obtain product information matching the interest preference.

[0029] Preferably, the AI ​​knowledge base model is called to analyze the text content of the document to obtain key information, including:

[0030] Calling the AI ​​knowledge base model to perform semantic analysis on the text content of the document to obtain semantic information of the text content;

[0031] Inputting the semantic information and the text content into the Bayesian network layer of the AI ​​knowledge base model for probability calculation to obtain the threshold probability of each node in the Bayesian network layer;

[0032] Based on the threshold probability of each node, a target node whose threshold probability is greater than a preset threshold probability is selected from multiple nodes;

[0033] Determine the information corresponding to the target node and obtain key information.

[0034] Preferably, extracting the text content of the document includes:

[0035] identifying structural features of the document;

[0036] Determine a segmentation mode according to the structural characteristics of the document;

[0037] Constructing a corresponding regular expression according to the segmentation pattern, using the regular expression to match segmentation points of the document, and dividing the document into multiple document segments according to the matched segmentation points;

[0038] After screening the multiple document fragments, a target document fragment is obtained;

[0039] The text content of the target document fragment is extracted.

[0040] Preferably, extracting the text content of the document includes:

[0041] When it is determined that the document contains a target text in a picture or PDF format, the text content of the target text is identified and extracted using OCR technology.

[0042] Preferably, performing dependency syntactic analysis on the text content to extract the main information of the sentences in the text content includes:

[0043] Calling a dependency syntactic analysis tool to perform dependency syntactic analysis on the text content to obtain a dependency tree, wherein the dependency tree is a graphical structure for representing dependency relationships between words in a sentence, each node in the dependency tree represents a word, and directed edges represent dependency relationships between words;

[0044] identifying a plurality of functional words in the dependency tree, the functional words comprising conjunctions, prepositions, and articles;

[0045] Calculate the contribution of the directed edge corresponding to each of the functional words to the trunk structure of the dependency tree, remove the functional words and the corresponding directed edges whose contribution is less than a preset threshold, and generate a target dependency tree;

[0046] Determining the dependency relationship of each word in the text content according to the target dependency tree, wherein the dependency relationship includes a subordinate relationship between words;

[0047] According to the dependency relationship of each word in the text content, the main information of the sentence in the text content is extracted.

[0048] Preferably, the AI ​​knowledge base model is called to analyze the text content of the document to obtain key information, including:

[0049] After word segmentation, stop word removal and grammatical annotation of the text content of the document, the target text content is obtained;

[0050] The target text content is input into the AI ​​knowledge base model for analysis to obtain key information.

[0051] The present invention provides a terminal, including a memory and a processor, wherein the memory stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, the processor executes the steps of the method for implanting exclusive AI interpretation into shared documents as described in any of the above items.

[0052] Compared with the prior art, the technical solution of the present invention has at least the following advantages:

[0053] The method and terminal for implanting exclusive AI interpretation for shared documents provided by the present invention utilize an AI knowledge base model to analyze documents, quickly identify key information and key points in documents, and apply a generative summary algorithm to create a concise summary for documents, so that recipients can quickly grasp the main points of the documents; by using the LDA algorithm to identify the dominant theme of the document, the understanding of the document content is enhanced, and at the same time, the document is subjected to dependency syntactic analysis to extract the main information of the sentences, further clarify the sentence structure and core meaning, and finally, by integrating information of different dimensions such as key information, summaries, themes and sentence trunks, a comprehensive document interpretation result is formed, and the integrated information provides a multi-dimensional perspective of the document, which helps recipients to have a deeper understanding of the document content, and also helps recipients to quickly obtain the core content and structure of the document without reading the full text, thereby improving the efficiency of document processing and information transmission. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 This is a flowchart of an embodiment of a method for embedding exclusive AI interpretation into shared documents according to the present invention;

[0055] Figure 2 This is a module block diagram of an embodiment of a device for embedding exclusive AI interpretation into shared documents according to the present invention;

[0056] Figure 3 FIG. 4 is a block diagram of the internal structure of a terminal in one embodiment of the present invention. DETAILED DESCRIPTION

[0057] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0058] In some of the processes described in the specification and claims of the present invention and the above-mentioned figures, multiple operations that appear in a specific order are included, but it should be clearly understood that these operations may not be executed in the order in which they appear in this article or executed in parallel. The sequence numbers of the operations, such as S11, S12, etc., are only used to distinguish between different operations, and the sequence numbers themselves do not represent any execution order. In addition, these processes may include more or fewer operations, and these operations may be executed in sequence or in parallel. It should be noted that the descriptions of "first", "second", etc. in this article are used to distinguish different messages, devices, modules, etc., do not represent the order of precedence, and do not limit the "first" and "second" to be different types.

[0059] It will be understood by those of ordinary skill in the art that, unless expressly stated, the singular forms "a", "an", "said" and "the" used herein may also include plural forms. It should be further understood that the term "comprising" used in the specification of the present invention refers to the presence of the features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof. It should be understood that when we refer to an element as being "connected" or "coupled" to another element, it may be directly connected or coupled to the other element, or there may be intermediate elements. In addition, the "connection" or "coupling" used herein may include wireless connection or wireless coupling. The term "and / or" used herein includes all or any unit and all combinations of one or more associated listed items.

[0060] It will be understood by those skilled in the art that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as generally understood by those skilled in the art in the art to which the present invention belongs. It should also be understood that terms such as those defined in common dictionaries should be understood to have meanings consistent with the meanings in the context of the prior art, and will not be interpreted with idealized or overly formal meanings unless specifically defined as herein.

[0061] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention, wherein the same or similar reference numerals throughout represent the same or similar elements or elements with the same or similar functions. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0062] See also Figure 1 The present invention provides a method for embedding exclusive AI interpretation into a shared document, comprising:

[0063] S11, receiving the document forwarded by the forwarder on the chat tool;

[0064] S12, extracting the text content of the document, calling the AI ​​knowledge base model to analyze the text content of the document, and obtaining key information;

[0065] S13, generating a summary of the text content according to a generative summary algorithm;

[0066] S14, using the LDA algorithm to identify the subject of the text content, and performing dependency syntax analysis on the text content to extract the main information of the sentences in the text content;

[0067] S15. After integrating the key information, abstract, theme and main information of the sentence, generate an interpretation result of the document, and send the interpretation result to a receiver.

[0068] As described in the above step S11, the recipient receives the document forwarded by the forwarder through a chat tool, which includes WeChat, WhatsApp or QQ, and the format of the document may include PDF, Word document and other formats.

[0069] As described in the above step S12, a document parsing tool may be used to extract plain text content from documents of various formats, and the document parsing tool includes a TexSmart or DocAI document intelligent analysis engine.

[0070] Calling an AI knowledge base model, such as a classifier or keyword extraction algorithm based on machine learning, identifies key information and main concepts from the text. The key information may include words or phrases that reflect the core theme of the document, the main viewpoints or arguments supporting the theme of the document, numerical data or charts used to support the arguments, action guidelines or suggestions based on the analysis and conclusions, professional terms and definitions for specific fields or topics, etc. The AI ​​knowledge base model is a pre-trained neural network model that is used to analyze text content and extract key information. It should be noted that the present invention does not specifically limit the training process of the AI ​​knowledge base model.

[0071] As described in step S13 above, the generative summary algorithm is an automatic text summarization technology that can not only extract key sentences from the text content, but also use the language model to generate new sentences to form a summary, thereby creating a fluent, concise and easy-to-understand text summary while retaining the core meaning of the original text.

[0072] Specifically, features such as keywords, phrases, sentence structures, etc. are extracted from the text content, and the features are processed using a sequence-to-sequence (Seq2Seq) model, which consists of two parts: an encoder and a decoder. The encoder is responsible for reading and understanding the input text, and the decoder is responsible for generating a summary, which is usually a sequence of words or phrases. At the same time, an attention mechanism is introduced to help the model better focus on the part of the input features that is most relevant to generating the summary.

[0073] In addition, the generated summary can be optimized, including grammatical correction, removal of duplicate content, and ensuring the coherence and fluency of the summary.

[0074] As described in step S14 above, this step uses the LDA algorithm to perform topic modeling on the text content, identify the main themes or topics in the text content, and perform dependency syntactic analysis on the text to identify the dependency relationship between words in the sentence, thereby extracting the main information of the sentence, namely the subject, predicate and object, etc.

[0075] Among them, dependency syntactic analysis is used to determine the dependency relationship between words in a sentence. This analysis method can reveal the grammatical structure of a sentence, including the subordinate relationship between words, such as subject, predicate, object, etc.

[0076] As described in step S15 above, this embodiment integrates information of different dimensions such as key information, abstract, theme and sentence trunk to form a coherent and comprehensive document interpretation result, and sends the interpretation result to the recipient through the chat tool, thereby providing a fast and convenient way of information transmission. Among them, the generated interpretation result is a high-level summary and explanation of the document content, so that the recipient can quickly grasp the main idea and structure of the document.

[0077] The method provided by the present invention for implanting exclusive AI interpretation for sharing documents utilizes the AI ​​knowledge base model to analyze the document, quickly identifies the key information and main points in the document, and applies the generative summary algorithm to create a concise summary for the document, so that the recipient can quickly grasp the main idea of ​​the document; by using the LDA algorithm to identify the dominant theme of the document, the understanding of the document content is enhanced, and at the same time, the document is subjected to dependency syntactic analysis to extract the main information of the sentence, further clarify the sentence structure and core meaning, and finally, by integrating information of different dimensions such as key information, summary, theme and sentence trunk, a comprehensive document interpretation result is formed, and the integrated information provides a multi-dimensional perspective of the document, which helps the recipient to understand the document content more deeply, and also helps the recipient to quickly obtain the core content and structure of the document without reading the full text, thereby improving the efficiency of document processing and information transmission.

[0078] In one embodiment, after sending the interpretation result to the recipient, the method further includes:

[0079] Obtaining the recipient's question information about the document;

[0080] Performing vectorization processing on the question information to obtain a vector corresponding to the question information;

[0081] Calling the AI ​​knowledge base model to analyze and process the vector corresponding to the question information, and screening out the answer information whose vector corresponding to the question information has a similarity greater than a preset degree;

[0082] The reply information is sent to the recipient.

[0083] In this embodiment, the recipient has specific questions about the document or needs more information. They can ask questions in the chat tool dialog box by text or voice. The system obtains the recipient's question information about the document and converts the recipient's question information into a numerical vector using natural language processing technology. For example, Word2Vec, GloVe or BERT is used to convert the question information into a format that the model can understand.

[0084] The AI ​​knowledge base model receives the vector representation of the question information and searches the knowledge base for information related to the vector corresponding to the question information. By comparing the vector corresponding to the question information with the existing answer vectors in the knowledge base, it filters out the answer information that exceeds the preset similarity threshold, thereby accurately matching the answer information and sending the filtered answer information back to the recipient to answer the recipient's question.

[0085] For example, suppose the recipient has a question about a scientific paper on climate change and asks, "What are the main findings of this paper?" The system first obtains the question information, then uses NLP technology to convert the question into a vector form. The AI ​​knowledge base model receives this vector and searches in a vector database containing the paper's abstract, key points, and conclusions. The model calculates the similarity between the question vector and each answer vector in the database. The system selects answers with a similarity higher than a preset threshold (for example, 0.8), including a summary of the main findings of the paper. The system sends these selected answers to the recipient.

[0086] This embodiment uses advanced NLP technology to ensure that questions are accurately understood and answered, and by providing timely and accurate responses, the efficiency and effectiveness of the interaction between the recipient and the document content are improved. At the same time, the system can also provide personalized responses based on the recipient's specific questions.

[0087] In one embodiment, after sending the reply information to the recipient, the method further includes:

[0088] When it is determined that the receiver is a target customer in the customer list, after the receiver finishes asking all questions, all interaction contents with the receiver are obtained;

[0089] Analyze all interactive contents to determine the interest preferences of the recipient;

[0090] Query product information that matches the interest preference, and send the product information and all interactive content that matches the interest preference to the sender.

[0091] When the recipient is confirmed to belong to the target customer group in the pre-built customer list, after the recipient finishes all questions, the system obtains all interaction records with the recipient, including questions, feedback, purchase history, etc. This embodiment can use text analysis and data mining technology to analyze the collected interaction content to determine the recipient's interest preferences, such as attention to a certain type of product or specific needs.

[0092] Based on the recipient's interest preferences, the system searches the product database for matching product information and sends the matching product information and all interactive content to the sender (which can be a marketing team or customer service representative) to understand the customer's interests or demand intentions, so that the customer service representative can follow up and gain insight into business opportunities.

[0093] For example, suppose a customer expresses interest in a certain brand of smartwatch and asks about the product features and price several times in a customer service chat. The customer service representative confirms the customer's interest in the smartwatch and collects all interactions after the customer's questions are completed. Using NLP technology, sentiment analysis and topic modeling are performed on the customer's questions and feedback to determine that the customer is particularly interested in the health monitoring function of the smartwatch. Based on the analysis results, the system searches the product database for smartwatch products with health monitoring functions. The system finds several smartwatches that meet the customer's preferences and extracts detailed information about these products, such as price, feature introduction, user reviews, etc. The system sends this product information and all the customer's interactions to the customer service representative so that they can provide more personalized services and recommendations, and improve the success rate of commodity transactions.

[0094] This embodiment can increase sales opportunities and conversion rates by providing product information that meets customer needs, and the enterprise can also have a deeper understanding of customer interests and preferences, providing guidance for product development and marketing strategies. At the same time, the enterprise can focus marketing resources on the customer groups that are most likely to generate returns, thereby improving resource utilization efficiency.

[0095] In one embodiment, querying product information matching the interest preference includes:

[0096] determining a user tag describing the interest preferences of the recipient;

[0097] Vectorize the user label to obtain a user label vector;

[0098] Extract product tags describing product information from the knowledge base;

[0099] Vectorize the product label to obtain a product label vector;

[0100] Calculate the cosine distance between the user label vector and the product label vector, and use the product label vector whose cosine distance with the user label vector is greater than a preset cosine distance as the target product label vector;

[0101] Determine the product information corresponding to the target product label vector, and obtain product information matching the interest preference.

[0102] This embodiment can determine user tags that describe the recipient's interest preferences, such as "technology enthusiasts", "healthy life", etc. by analyzing the recipient's interactive content, purchase history, browsing behavior, etc. The user tag is converted into a vector form through Word2Vec to obtain a user tag vector so that the computer can process and analyze it. Product tags that describe product information, such as "smart wearables", "low-sugar foods", etc., are extracted from a pre-built knowledge base. Similarly, the product tag is converted into a vector form through Word2Vec to obtain a product tag vector, and the cosine distance between the user tag vector and the product tag vector is calculated to evaluate the similarity between them. According to the preset cosine distance threshold, the product tag vector that is closer to the user tag vector is screened out, that is, the target product tag vector; finally, according to the screened target product tag vector, the corresponding product information is determined, and these products are most likely to meet the recipient's interest preferences.

[0103] For example, suppose a user often mentions content related to healthy eating during interaction. By analyzing user behavior, the system determines that the user tags are "healthy eating", "nutrition", and "organic food". These user tags are vectorized and converted into user tag vectors. The system extracts product tags from the knowledge base, such as "low calorie", "whole wheat", "organic certification", etc., and the product tags are also vectorized to form product tag vectors. Calculate the cosine distance between the user tag vector and all product tag vectors. If the cosine distance threshold is set to 0.7, the product tag vectors with a cosine distance greater than the threshold are filtered out. Finally, the system determines the products that match the "healthy eating" user tag, such as "organic whole wheat bread".

[0104] This embodiment can accurately match user needs and product features in the form of vectors, increase the possibility of purchase, provide users with product recommendations that are highly relevant to their interests and preferences, and improve user satisfaction.

[0105] In one embodiment, the AI ​​knowledge base model is called to analyze the text content of the document to obtain key information, including:

[0106] Calling the AI ​​knowledge base model to perform semantic analysis on the text content of the document to obtain semantic information of the text content;

[0107] Inputting the semantic information and the text content into the Bayesian network layer of the AI ​​knowledge base model for probability calculation to obtain the threshold probability of each node in the Bayesian network layer;

[0108] Based on the threshold probability of each node, a target node whose threshold probability is greater than a preset threshold probability is selected from multiple nodes;

[0109] Determine the information corresponding to the target node and obtain key information.

[0110] In this embodiment, the AI ​​knowledge base model is used to conduct an in-depth analysis of the text content of the document to understand the deep meaning and context of the text. After analysis, the AI ​​knowledge base model extracts the semantic information of the text content, including keywords, concepts, entity relationships, etc.

[0111] At the same time, the AI ​​knowledge base model is pre-configured with a Bayesian network layer. Then, the extracted semantic information and text content are input into the Bayesian network layer of the AI ​​knowledge base model. On the Bayesian network layer, the probability of each node is calculated to obtain the threshold probability of each node in the Bayesian network layer. These nodes represent different concepts or attributes in the text, and the threshold probability represents the possibility that the information corresponding to the node belongs to key information. The Bayesian network is a graphical model used to represent the dependency relationship between variables.

[0112] Based on the calculated threshold probability, target nodes whose threshold probability is greater than the preset threshold probability are screened out. These target nodes are considered to be key information in the text, and the specific information corresponding to the target nodes is determined to obtain the key information of the document.

[0113] For example, suppose there is a research report on emerging technology trends. First, use the AI ​​knowledge base model to perform semantic analysis on the report to identify the technology trends and related concepts discussed in the report. The semantic information extracted by the model includes keywords such as "artificial intelligence", "machine learning", and "deep learning". This semantic information and the report content are input into the Bayesian network layer of the model for probability calculation. Assuming that there are multiple nodes in the Bayesian network layer, such as "technology maturity", "market potential", "application field", etc., the model calculates the threshold probability of each node. Set the threshold probability to 0.8 and filter out nodes with a threshold probability greater than 0.8, such as the "market potential" node. Determine the information corresponding to the "market potential" node, such as the "artificial intelligence application in the medical field has huge market potential" mentioned in the report.

[0114] This embodiment can more deeply understand the document content and context through semantic analysis; through probability calculation and target node screening through the Bayesian network layer, key information can be accurately and effectively extracted from the document. At the same time, the Bayesian network can also continuously update the probability according to new information to improve the adaptability and accuracy of the model.

[0115] In one embodiment, extracting the text content of the document includes:

[0116] identifying structural features of the document;

[0117] Determine a segmentation mode according to the structural characteristics of the document;

[0118] Constructing a corresponding regular expression according to the segmentation pattern, using the regular expression to match segmentation points of the document, and dividing the document into multiple document segments according to the matched segmentation points;

[0119] After screening the multiple document fragments, a target document fragment is obtained;

[0120] The text content of the target document fragment is extracted.

[0121] This embodiment implements an optimized processing flow of a document, aiming to extract the text content of a specific document segment by identifying the structural features of the document and segmenting it using regular expressions.

[0122] Specifically, first identify the structural characteristics of the document, that is, analyze the layout and format of the document, and identify structural elements such as titles, paragraphs, lists, and tables. According to the structural characteristics of the document, determine the appropriate segmentation criteria, such as blank lines, specific separators, heading levels, etc. According to the determined segmentation pattern, construct the corresponding regular expression to identify the segmentation points in the document.

[0123] Among them, the regular expression is a text pattern matching method used for string search, replacement, text verification and other operations. It consists of a series of characters that define a search pattern, which can be used to check whether a string conforms to a certain format, or to extract strings that conform to a specific pattern from a text. In the document processing process, regular expressions can be used to identify the segmentation points in the document so that the document can be divided into multiple fragments. The segmentation points can be specific text patterns in the document, such as titles, specific delimiters, blank lines, etc.

[0124] Use regular expressions to scan the document and find all matching segmentation points. According to the identified segmentation points, split the document into multiple independent document fragments, filter the segmented document fragments, retain the target document fragments according to specific conditions (such as fragments containing specific keywords or topics), and extract text content from the filtered target document fragments for further processing or analysis, so as to accurately filter out text fragments with substantive meaning without analyzing and processing all text content, thereby improving the system's analysis efficiency of text content.

[0125] For example, suppose there is a Word document with multiple chapters, and each chapter starts with a title, such as "Introduction to Chapter 1". First, identify the structural features of the document, noting that the title of each chapter is in the form of "Chapter X" and is followed by a blank line. Then determine the segmentation pattern and decide to use the title of each chapter and the subsequent blank line as the segmentation point. Construct a regular expression, such as Chapter*s*$, to match lines that start with "Chapter" and end with "Chapter", including the blank line behind. Use the regular expression to scan the entire document and identify the titles and blank lines of all chapters. Based on the identified segmentation points, split the document into independent chapters, filter out chapters containing specific keywords or information, such as chapters containing "Market Analysis", and extract the text content of the filtered chapters for further analysis or summary generation.

[0126] This embodiment can make document processing more structured and systematic by identifying the structural characteristics of the document. The segmentation mode based on the structural characteristics also improves the accuracy of document segmentation. At the same time, the automated regular expression matching and document segmentation reduce the need for manual operation. In addition, by screening specific document fragments, information extraction is more focused on actual needs, which improves the efficiency of document processing and information extraction and saves time.

[0127] In one embodiment, extracting the text content of the document includes:

[0128] When it is determined that the document contains a target text in a picture or PDF format, the text content of the target text is identified and extracted using OCR technology.

[0129] When a document contains text in the form of images, or when the text embedded in a PDF file cannot be copied directly, optical character recognition (OCR) technology can be used to identify and extract the text content. Specifically, first determine whether the image or PDF in the document contains the target text to be extracted, use OCR technology to process these images or PDF files, identify the text content, and correct and format the text recognized by OCR technology to improve accuracy and readability. Among them, the OCR technology converts the text in the image into an editable and searchable text format.

[0130] This embodiment converts the text in the image or PDF into an editable format, thereby improving the accessibility of the document. The automated OCR process also reduces the time and labor cost of manually inputting text, as well as possible errors.

[0131] In one embodiment, performing dependency syntactic analysis on the text content to extract the main information of the sentences in the text content includes:

[0132] Calling a dependency syntactic analysis tool to perform dependency syntactic analysis on the text content to obtain a dependency tree, wherein the dependency tree is a graphical structure for representing dependency relationships between words in a sentence, each node in the dependency tree represents a word, and directed edges represent dependency relationships between words;

[0133] identifying a plurality of functional words in the dependency tree, the functional words comprising conjunctions, prepositions, and articles;

[0134] Calculate the contribution of the directed edge corresponding to each of the functional words to the trunk structure of the dependency tree, remove the functional words and the corresponding directed edges whose contribution is less than a preset threshold, and generate a target dependency tree;

[0135] Determining the dependency relationship of each word in the text content according to the target dependency tree, wherein the dependency relationship includes a subordinate relationship between words;

[0136] According to the dependency relationship of each word in the text content, the main information of the sentence in the text content is extracted.

[0137] In this embodiment, the text content can be cleaned to remove irrelevant characters and punctuation marks, and then a dependency parsing tool such as Stanford NLP, SpaCy or NLTK is used to process the cleaned text to generate a dependency tree. The dependency tree is a graphical structure. In dependency parsing, the dependency tree decomposes a sentence into a directed graph, in which each node represents a word and the directed edges represent the dependency relationship between words.

[0138] Specifically, the dependency tree has the following characteristics: The dependency tree reflects the hierarchical structure of the sentence, and each word is directly dependent on its function or grammatical role in the sentence. At the same time, the edges in the dependency tree are directional, indicating the direction of dependency between words. For example, a verb may depend on its subject and object. The dependency tree has a root node, which is usually the predicate verb of the sentence, and the dependency structure of the entire sentence is built around this root node. Each node (term) in the dependency tree is usually accompanied by a part-of-speech tag, such as noun, verb, adjective, etc. In addition, the edges in the dependency tree are usually accompanied by dependency labels, which describe the specific types of dependency relationships, such as "nsubj" (nominal subject), "dobj" (direct object), etc. In the dependency tree, a word can also have multiple dependencies, which allows the model to capture more complex syntactic structures.

[0139] Identify functional words in the dependency tree, such as conjunctions ("and", "but"), prepositions ("in", "by"), and articles ("a", "the"). Evaluate the contribution of the directed edges corresponding to each functional word to the backbone structure of the dependency tree. For example, calculate the contribution by analyzing the type of the edge, the frequency of the word, or its role in the sentence. According to a preset contribution threshold, remove those functional words with low contributions and their corresponding directed edges, thereby generating a more concise target dependency tree.

[0140] In one embodiment, when calculating the contribution of the directed edges corresponding to each functional word to the backbone structure of the dependency tree, weights can be assigned to different types of dependency relationships. For example, the subject-predicate relationship may contribute more to the backbone structure than the adverbial relationship. Or, different importance levels can be given according to the part of speech. Generally, nouns, verbs, and adjectives contribute more to the meaning of the sentence than adverbs and prepositions. Or, analyze the sentence components to determine the role of the functional word in the sentence. If the functional word is the core part of the sentence component, its contribution is higher. Or, count the frequency of the functional word appearing in the document. Words with high frequencies may contribute more to the overall structure.

[0141] Specifically, the system can calculate the dependency path length from the sentence subject to other words for each of the functional words, and determine the contribution of the directed edges corresponding to each of the functional words to the backbone structure of the dependency tree according to the dependency path length. This contribution is inversely correlated with the dependency path length, that is, functional words with shorter paths may contribute more to the backbone structure.

[0142] For example, assume there is a dependency tree for the following sentence: "He walks in the park."

[0143] Identify the functional words "in" (preposition) and "里" (preposition).

[0144] Analyze the type of dependency relationship of "in" and "里", which may be "prep" (prepositional phrase).

[0145] Assign a weight to the "prep" dependency relationship type, such as 0.5.

[0146] Calculate the path length and syntactic depth of "in" and "里", and find that "in" is closer to the subject "he", and "里" is farther from the subject "he".

[0147] Based on the above factors, it is evaluated that the contribution of "in" to the backbone structure is higher, and the contribution of "里" to the backbone structure is lower.

[0148] Through the above method, it is possible to accurately determine which functional words contribute less to the backbone structure of the dependency tree, and then consider removing them when generating the target dependency tree, which helps to simplify the sentence structure and highlight the core information.

[0149] Next, analyze the target dependency tree to determine the dependency relationships of each word in the text, including subordination relationships. Based on the dependency relationships, extract the main information of the sentence, which usually includes the subject, predicate, and object.

[0150] For example, assume there is the following sentence: "He passed the exam through hard study." Then use a dependency parsing tool to analyze the sentence and obtain the dependency tree. Identify functional words such as "through" (preposition) and "了" (auxiliary word) in the dependency tree. Calculate the contribution degree of the directed edge corresponding to "through" to the main structure and find that its impact on the main structure "He - studied hard - passed the exam" is small. Remove "through" and its corresponding directed edge to generate the target dependency tree. Analyze the target dependency tree to determine that "He" is the subject, "studied hard" is the predicate, and "passed the exam" is the object. Extract the main sentence: "He studied hard passed the exam."

[0151] This embodiment can help extract key information more accurately through dependency parsing, and through contribution degree calculation and comparison with a preset threshold, precisely screen and remove functional words with less impact, simplify the sentence structure, make the main structure clearer, and at the same time help improve the understanding efficiency of the text content by quickly identifying the main sentence of the text.

[0152] In one embodiment, call the AI knowledge base model to analyze the text content of the document to obtain key information, including:

[0153] After performing word segmentation, removing stop words, and grammar annotation on the text content of the document, obtain the target text content;

[0154] Input the target text content into the AI knowledge base model for analysis to obtain key information.

[0155] In this embodiment, the system performs word segmentation on the text content of the document, that is, divides a continuous text string into meaningful words or phrases. Removing stop words is to remove common small words in the language (such as "的", "和", "是", etc.), which usually do not carry important information in text analysis. Perform grammar annotation on the text content to identify the part-of-speech (noun, verb, adjective, etc.) and sentence structure in the text.

[0156] Finally, input the preprocessed target text content into the AI knowledge base model to obtain key information, such as "market demand growth", "changes in consumer behavior", etc. in a Word document about market trend analysis.

[0157] Please refer to Figure 2 , an embodiment of the present invention also provides a device for implanting an exclusive AI interpretation into a shared document, including:

[0158] The receiving module 11 is used to receive the document forwarded by the forwarder on the chat tool;

[0159] An extraction module 12 is used to extract the text content of the document, call the AI ​​knowledge base model to analyze the text content of the document, and obtain key information;

[0160] A generating module 13, used for generating a summary of the text content according to a generative summary algorithm;

[0161] The recognition module 14 is used to identify the subject of the text content by using the LDA algorithm, and perform dependency syntax analysis on the text content to extract the main information of the sentences in the text content;

[0162] The integration module 15 is used to integrate the key information, abstract, theme and main information of the sentence, generate an interpretation result of the document, and send the interpretation result to the receiver.

[0163] Regarding the device in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated here.

[0164] A terminal provided by the present invention includes a memory and a processor, wherein the memory stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, the processor executes the steps of the method for implanting exclusive AI interpretation into shared documents as described in any of the above items.

[0165] In one embodiment, the terminal provided in one embodiment of the present application refers to Figure 3 The terminal may be a computer device, and its internal structure may be as follows Figure 3 As shown. The computer device includes a processor, a memory, a network interface, a display screen and an input device connected via a system bus. Among them, the processor designed by the computer is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store relevant data of the method for implanting exclusive AI interpretation into shared documents. The network interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, the method for implanting exclusive AI interpretation into shared documents described in the above embodiment is implemented.

[0166] In one embodiment, the present invention further proposes a storage medium storing computer-readable instructions, which, when executed by one or more processors, enables one or more processors to execute the above-mentioned method of embedding exclusive AI interpretation into shared documents. The storage medium may be ROM, random access memory (RAM), CD-ROM, magnetic tape, floppy disk, optical data storage device, etc.

[0167] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program, and the computer program can be stored in a storage medium. When the program is executed, it can include the processes of the embodiments of the above-mentioned methods. The aforementioned storage medium can be a non-volatile storage medium such as a disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

[0168] Based on the above embodiments, it can be seen that the greatest beneficial effect of the present invention is:

[0169] The method and terminal for implanting exclusive AI interpretation for shared documents provided by the present invention utilize an AI knowledge base model to analyze documents, quickly identify key information and key points in documents, and apply a generative summary algorithm to create a concise summary for documents, so that recipients can quickly grasp the main points of the documents; by using the LDA algorithm to identify the dominant theme of the document, the understanding of the document content is enhanced, and at the same time, the document is subjected to dependency syntactic analysis to extract the main information of the sentences, further clarify the sentence structure and core meaning, and finally, by integrating information of different dimensions such as key information, summaries, themes and sentence trunks, a comprehensive document interpretation result is formed, and the integrated information provides a multi-dimensional perspective of the document, which helps recipients to have a deeper understanding of the document content, and also helps recipients to quickly obtain the core content and structure of the document without reading the full text, thereby improving the efficiency of document processing and information transmission.

[0170] The technical features of the above-described embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above-described embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0171] The above-mentioned embodiments only express several implementation methods of the present invention, and the description thereof is relatively specific and detailed, but it cannot be understood as limiting the scope of the patent of the present invention. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.

Claims

1. A method for embedding exclusive AI interpretation into shared documents, characterized in that: include: Receive documents forwarded by the forwarder on the chat tool; Extract the text content of the document, call the AI ​​knowledge base model to analyze the text content of the document, and obtain key information; Generate a summary of the text content according to a generative summary algorithm; Using the LDA algorithm to identify the subject of the text content, and performing dependency syntactic analysis on the text content to extract the main information of the sentences in the text content; After integrating the key information, abstract, theme and main information of the sentence, generating an interpretation result of the document, and sending the interpretation result to a receiver; The performing dependency syntactic analysis on the text content to extract the main information of the sentences in the text content includes: Calling a dependency syntactic analysis tool to perform dependency syntactic analysis on the text content to obtain a dependency tree, wherein the dependency tree is a graphical structure for representing dependency relationships between words in a sentence, wherein each node in the dependency tree represents a word, and directed edges represent dependency relationships between words, identifying multiple functional words in the dependency tree, wherein the functional words include conjunctions, prepositions, and articles, calculating the contribution of directed edges corresponding to each functional word to the trunk structure of the dependency tree, removing functional words and corresponding directed edges whose contribution is less than a preset threshold, generating a target dependency tree, determining the dependency relationship of each word in the text content according to the target dependency tree, wherein the dependency relationship includes a subordinate relationship between words, and extracting trunk information of sentences in the text content according to the dependency relationship of each word in the text content; The calculating the contribution of the directed edge corresponding to each functional word to the trunk structure of the dependency tree includes: The length of the dependency path from the subject of the sentence to other words of each functional word is calculated, and the contribution of the directed edge corresponding to each functional word to the trunk structure of the dependency tree is determined according to the dependency path length, and the contribution is inversely correlated with the dependency path length.

2. The method for embedding exclusive AI interpretation into shared documents according to claim 1, characterized in that: After the interpretation result is sent to the recipient, the method further includes: Obtaining the recipient's question information about the document; Performing vectorization processing on the question information to obtain a vector corresponding to the question information; Calling the AI ​​knowledge base model to analyze and process the vector corresponding to the question information, and screening out the answer information whose vector corresponding to the question information has a similarity greater than a preset degree; The reply information is sent to the recipient.

3. The method for embedding exclusive AI interpretation into shared documents according to claim 2, characterized in that: After sending the reply information to the recipient, the method further includes: When it is determined that the receiver is a target customer in the customer list, after the receiver finishes asking all questions, all interaction contents with the receiver are obtained; Analyze all interactive contents to determine the interest preferences of the recipient; Query product information that matches the interest preference, and send the product information and all interactive content that matches the interest preference to the sender.

4. The method for embedding exclusive AI interpretation into shared documents according to claim 3, characterized in that: Query product information that matches the interest preferences, including: determining a user tag describing the interest preferences of the recipient; Vectorize the user label to obtain a user label vector; Extract product tags describing product information from the knowledge base; Vectorize the product label to obtain a product label vector; Calculate the cosine distance between the user label vector and the product label vector, and use the product label vector whose cosine distance with the user label vector is greater than a preset cosine distance as the target product label vector; Determine the product information corresponding to the target product label vector, and obtain product information matching the interest preference.

5. The method for embedding exclusive AI interpretation into shared documents according to claim 1, characterized in that: The AI ​​knowledge base model is called to analyze the text content of the document to obtain key information, including: Calling the AI ​​knowledge base model to perform semantic analysis on the text content of the document to obtain semantic information of the text content; Inputting the semantic information and the text content into the Bayesian network layer of the AI ​​knowledge base model for probability calculation to obtain the threshold probability of each node in the Bayesian network layer; Based on the threshold probability of each node, a target node whose threshold probability is greater than a preset threshold probability is selected from multiple nodes; Determine the information corresponding to the target node and obtain key information.

6. The method for embedding exclusive AI interpretation into shared documents according to claim 1, characterized in that: Extracting the text content of the document, including: identifying structural features of the document; Determine a segmentation mode according to the structural characteristics of the document; Constructing a corresponding regular expression according to the segmentation pattern, using the regular expression to match segmentation points of the document, and dividing the document into multiple document segments according to the matched segmentation points; After screening the multiple document fragments, a target document fragment is obtained; The text content of the target document fragment is extracted.

7. The method for embedding exclusive AI interpretation into shared documents according to claim 1, characterized in that: Extracting the text content of the document, including: When it is determined that the document contains a target text in a picture or PDF format, the text content of the target text is identified and extracted using OCR technology.

8. The method for embedding exclusive AI interpretation into shared documents according to claim 1, characterized in that: The AI ​​knowledge base model is called to analyze the text content of the document to obtain key information, including: After word segmentation, stop word removal and grammatical annotation of the text content of the document, the target text content is obtained; The target text content is input into the AI ​​knowledge base model for analysis to obtain key information.

9. A terminal, characterized in that: It includes a memory and a processor, wherein the memory stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, the processor executes the steps of the method for implanting exclusive AI interpretation for shared documents as described in any one of claims 1 to 8.

Images