A text understanding method and system based on parallel multi-dimensional semantic signal output

Abstract

The application discloses a text understanding method and system based on parallel multi-dimensional semantic signal output, and belongs to the technical field of natural language processing. The method simultaneously performs parallel analysis on literal dimension, intention dimension and emotion dimension of input text, and fuses signals of the three dimensions into comprehensive cognitive signal output. The intention dimension adopts sentence structure recognition and does not need to depend on a large-scale language model. The method simulates the instantaneous understanding process of the human brain through a parallel multi-dimensional architecture, and outputs cognitive signals that can be directly used by the brain, rather than fragmented labels.

Description

Technical Field

[0001] This invention belongs to the field of artificial intelligence and natural language processing technology, specifically relating to a text understanding method and system that is fully localized, does not rely on external large models, and is based on parallel multi-dimensional semantic signal output. Background Technology

[0002] Existing natural language understanding technologies primarily rely on large language models, achieving semantic understanding by calling cloud APIs or deploying large-parameter models locally. These solutions suffer from the following drawbacks: reliance on external models leads to data security risks; enormous computational power consumption makes them difficult to deploy on personal devices; the semantic understanding process is a black box operation, lacking interpretability; and the output is a single label or vector, failing to fully represent multi-dimensional semantic information. Therefore, there is an urgent need for a text understanding method that is fully localized, interpretable, and capable of parallel outputting multi-dimensional semantic signals. Summary of the Invention

[0003] The core innovation of this invention lies in abandoning the traditional serial processing flow and adopting a parallel architecture to simultaneously output semantic signals in three dimensions: literal, intentional, and emotional, thereby simulating the instantaneous multidimensional understanding process of the human brain when receiving language. Specifically, this method performs three parsing operations on the input text simultaneously: 1. Literal dimension analysis: Identify the calling object, extract core concepts and label semantic categories (time, space, action, reference, emotion, etc.). 2. Intent Dimension Analysis: Sentence structure is used to identify and determine the overall intent. For example, ending with "hello" indicates a greeting, beginning with "I am" indicates a statement, and containing an interrogative word and ending with a question mark indicates a question. Intent types are categorized into only five types: greeting, question, statement, farewell, and instruction. 3. Emotional Dimension Analysis: Based on a preset set of emotional words, detect emotional tendencies and output positive, negative, or neutral emotional signals. The signals from the three dimensions are output in parallel within the same processing cycle, fused into a comprehensive cognitive signal, and then transmitted to the decision-making module. The decision-making module receives not fragmented labels, but a complete, multi-dimensional cognitive picture. Attached Figure Description

[0004] Figure 1 System architecture diagram (three parsing modules process in parallel and converge to the signal fusion module) (The three modules are arranged in parallel: user input simultaneously enters the literal parsing module, the intent parsing module, and the sentiment parsing module. The outputs of the three modules converge to the signal fusion module, which outputs a comprehensive cognitive signal.) Figure 2Semantic understanding workflow diagram (input → parallel three-dimensional parsing → signal fusion → comprehensive cognitive output) (flow: user input → parallel three-dimensional parsing (literal dimension, intention dimension, and emotional dimension are performed simultaneously) → signal fusion → comprehensive cognitive signal output → transmitted to the brain's decision-making layer.) Detailed Implementation Example 1: Greeting Scenario User input: "Good morning, Xinyao" Literal Analysis: The target of the call is "Xinyao," and the core concepts are "morning" (time concept) and "good" (evaluation concept). Output literal signal: {Target: Xinyao, Concepts: [morning (time), good (evaluation)]}. Intent dimension analysis: The sentence ends with "okay," matching the greeting pattern rule. Output intent signal: {intent: greeting}. Sentiment dimension analysis: No negative sentiment words were hit, output sentiment signal: {sentiment: positive}. The three signals are merged into a comprehensive cognitive signal: {Target: Xinyao, Intent: Greeting, Emotion: Positive, Concept: [Morning, Good]}. Upon receiving this signal, the decision-making module directly triggers the greeting response process, generating a greeting message. Technical effect Compared with the prior art, the present invention has the following beneficial effects: 1. Fully localized operation: No need for any external large models or network connections, and data is autonomously controllable. 2. Parallel and efficient: Simultaneous analysis across three dimensions results in a fast response time, simulating the instantaneous understanding process of the human brain. 3. Highly interpretable: The source of the signal in each dimension is clearly traceable, and there is no black box. 4. Decision-making friendly: The comprehensive cognitive signals output can be directly used for decision-making without secondary processing.

Claims

1. A text understanding method based on parallel multi-dimensional semantic signal output, characterized in that, Includes the following steps: (1) Receive text information input by the user; (2) Perform literal dimension parsing on the text information, extract the calling object, core concepts and their semantic category labels, and output the literal dimension signal; (3) Perform intent dimension analysis on the text information, identify and judge the overall intent type based on sentence structure, and output the intent dimension signal; (4) Perform sentiment dimension analysis on the text information, detect the sentiment tendency contained therein, and output the sentiment dimension signal; (5) The literal dimension signal, intention dimension signal and emotional dimension signal are fused in parallel into a comprehensive cognitive signal and output to the decision module.

2. The method according to claim 1, characterized in that, The literal dimension parsing in step (2) includes: (2a) Identify the calling object in the text information, wherein the calling object is a word or phrase in the text used to address the other party in the dialogue; (2b) Extract the core concepts from the text information, wherein the core concepts include at least one of time concept, space concept, action concept, referential concept, and emotion concept; (2c) Label each core concept with a semantic category label.

3. The method according to claim 1, characterized in that, The intent dimension parsing in step (3) uses sentence structure recognition, including: If the sentence ends with a greeting marker (such as "hello"), then the intention is determined to be a greeting; If the sentence begins with a self-introduction marker (such as "I am"), then the intention is determined to be a statement; If a sentence contains interrogative markers (such as "what", "how", "who") and ends with an interrogative symbol or interrogative particle, then the intention is to ask a question. If a sentence ends with a farewell marker (such as "la", "ba", "oh") and contains words of farewell, then the intention is determined to be a farewell. If the sentence is in the imperative form, then the intention is determined to be an instruction.

4. The method according to claim 3, characterized in that, The output of the intent dimension parsing is one of five types of intent: greeting, question, statement, farewell, and instruction.

5. The method according to claim 1, characterized in that, The emotional dimension analysis in step (4) includes: The text information is compared with a preset set of positive sentiment words and a preset set of negative sentiment words; If a positive sentiment word is matched, a positive sentiment signal is output. If a negative sentiment word is matched, a negative sentiment signal will be output. If none of the above conditions are met, a neutral sentiment signal will be output.

6. The method according to claim 1, characterized in that, Parallel fusion in step (5) means that the literal dimension signal, intention dimension signal and emotion dimension signal are output simultaneously in the same processing cycle, rather than being output step by step in sequence.

7. The method according to claim 1, characterized in that, The method further includes: The comprehensive cognitive signal is matched with a preset knowledge base to obtain matching knowledge entries, and these matching knowledge entries are added to the comprehensive cognitive signal as supplementary information.

8. A text understanding system based on parallel multi-dimensional semantic signal output, characterized in that, include: The literal parsing module is used to parse the literal information of the input text and output the literal dimension signal; The intent parsing module is used to identify and determine the overall intent of the input text based on sentence structure and output intent dimension signals; The sentiment analysis module is used to detect the sentiment tendency of the input text and output the sentiment dimension signal; The signal fusion module is used to fuse the signals from the three dimensions in parallel into a comprehensive cognitive signal and output it.

9. A computer-readable storage medium having a computer program stored thereon that, when executed by a processor, implements the method of any one of claims 1 to 7.

Images