Interaction text analysis method and system based on large model agent

Abstract

The application relates to the technical field of natural language processing, in particular to an interactive text analysis method and system based on a large model agent, which comprises the following steps: obtaining each word group in the interactive text of the large model agent, and extracting the parts of speech of each word group; obtaining the approximate word groups of each word group in the interactive text through the word groups with the same part of speech as each word group in the interactive text; obtaining the content importance coefficient of each word group in the interactive text; obtaining the word group combination appearance probability and the part of speech combination appearance probability of each word group in the interactive text by obtaining the appearance probability of each part of speech in the sentence where each word group is located; obtaining the negligible coefficient of each word group in the interactive text by combining the dispersion degree of the part of speech word group in the sentence where each word group is located; and obtaining the information key degree of each word group in the interactive text, which is used for extracting the keywords in the interactive text, and then analyzing the interactive text of the large model agent. The application can improve the analysis accuracy of the interactive text of the large model agent.

Description

Technical Field

[0001] This application relates to the field of natural language processing technology, specifically to an interactive text parsing method and system based on large model intelligent agents. Background Technology

[0002] With the development of generative artificial intelligence, the application scenarios of human-computer interaction are gradually increasing. Through question-and-answer interactions with large generative AI models, decision-making in various fields can be assisted. Currently, users mainly input information in various ways. The large model agent transforms and parses the input to generate output text, thus completing the entire interaction process. For example, input speech data can be converted into text data, natural language processing algorithms can be used to parse the semantics of the text to understand the input content, and then text can be generated for output.

[0003] In current text parsing, to achieve rapid response in human-computer interaction, the input content is typically preprocessed to extract keywords, quickly identifying key terms. Semantic dependency analysis and contextual understanding algorithms are then used to rapidly understand the key points of the input content, leading to a quick response and improved interaction efficiency. Existing technologies primarily use the TF-IDF (Term Frequency-Inverse Document Frequency) algorithm for keyword extraction. This algorithm analyzes the term frequency and inverse document frequency of words in the interactive text to determine the contribution and criticality of each word, ultimately selecting the keywords.

[0004] However, in practical applications, due to the wide range of applications and complex scenarios of large model agents, the input process may be subject to various needs and purposes, resulting in a lot of uncertainty in the converted interactive text. Consequently, the words in the text are not judged based on the magnitude of word frequency or inverse document frequency, which will cause important information to be missed after text parsing and ultimately affect the effectiveness of agent interaction. Summary of the Invention

[0005] To address the aforementioned technical problems, the purpose of this application is to provide an interactive text parsing method and system based on large-scale intelligent agents. The specific technical solution adopted is as follows:

[0006] This application provides an interactive text parsing method based on a large model intelligent agent, including the following steps:

[0007] Obtain each word group from the interactive text of a large model agent and extract the part-of-speech tag for each word group;

[0008] By analyzing the differences between each phrase in the interactive text and all other phrases of the same part of speech, we obtain the approximate phrases of each phrase in the interactive text. We then count the number of all approximate phrases of each phrase in the interactive text, as well as the number of phrases in the sentences containing each phrase. Finally, we combine the differences between each phrase in the interactive text and the first and last phrases in the sentences containing each phrase to obtain the content importance coefficient of each phrase in the interactive text.

[0009] The number of phrases in the sentences containing each phrase in the interactive text and the number of phrases of each part of speech in the sentences are counted. The probability of each phrase of each part of speech appearing in the sentences containing each phrase in the interactive text is obtained. Combined with the phrases and phrases of each part of speech in the interactive text with the phrases before and after them, and the probability of phrase combination and phrase combination appearing in the interactive text are obtained based on the number of phrase combinations and the number of phrase combinations. Then, the negligible coefficient of each phrase in the interactive text is obtained.

[0010] Based on the ignorability coefficient of each phrase in the interactive text, and combined with the content importance coefficient of each phrase in the interactive text, the information criticality of each phrase in the interactive text is obtained. This information is then used to extract all keywords from the interactive text, so as to use keywords to parse the interactive text of the large model agent.

[0011] Preferably, the approximate word groups of each phrase in the interactive text further include:

[0012] The edit distance between each word group in the interactive text and all other word groups with the same part of speech is calculated. All the edit distances are then thresholded to obtain the first segmentation threshold. Word groups whose edit distances to each word group are less than or equal to the first segmentation threshold are considered as approximate word groups of each word group.

[0013] Preferably, the method for calculating the content importance coefficient of each phrase in the interactive text is as follows:

[0014]

[0015] In the formula, A i C is the content importance coefficient of the i-th phrase in the interactive text; i B is the total number of phrases in the sentence containing the i-th phrase in the interactive text; i b is the distance between the i-th phrase in the interactive text and the first phrase in the sentence containing it; i D is the distance between the i-th phrase in the interactive text and the last phrase in the sentence containing it; i Z represents the total number of approximate word groups existing for the i-th word group in the interactive text; min(·) is the function that selects the minimum value; |·| is the absolute value operation; iLet be the part of speech of the i-th phrase in the interactive text; n. be the part of speech of a noun; v. be the part of speech of a verb; α be a constant to avoid a denominator of 0.

[0016] Preferably, the method for obtaining the probability of occurrence of each part of speech in the sentence containing each phrase in the interactive text is as follows:

[0017] The ratio of the total number of phrases of each part of speech in the sentence containing each phrase in the interactive text to the total number of phrases in the sentence containing each phrase is taken as the probability of each phrase of each part of speech appearing in the sentence containing each phrase in the interactive text.

[0018] Preferably, the probability of phrase combinations and the probability of part-of-speech combinations appearing in the interactive text further include:

[0019] Each pair of adjacent word groups is combined into a word group combination. The preceding word group combination and the following word group combination of each word group are extracted. The ratio of the total number of word group combinations in the interactive text that are the same as the preceding word group combination and the following word group combination of each word group to the total number of word groups in the interactive text is used as the probability of the preceding word group combination and the following word group combination of each word group appearing in the interactive text. The sum of the two is used as the probability of the word group combination of each word group appearing.

[0020] Accordingly, the part-of-speech of each phrase is obtained using the same process as the probability of phrase combination occurrence, thus obtaining the probability of part-of-speech combination occurrence of each phrase in the interactive text.

[0021] Preferably, the method for extracting the preceding and following word combinations of each phrase is as follows:

[0022] Each phrase in the interactive text is combined with its preceding and following phrases to obtain the preceding and following phrase combinations for each phrase.

[0023] Preferably, the method for calculating the ignorability coefficient of each phrase in the interactive text is as follows:

[0024]

[0025] In the formula, E i F is the ignorability coefficient of the i-th phrase in the interactive text; i G represents the variance of the probability of each part of speech appearing in the sentence containing the i-th phrase in the interactive text; i g represents the probability of part-of-speech combinations of the i-th phrase in the interactive text. i Let be the probability of the i-th word combination appearing in the interactive text.

[0026] Preferably, the method for calculating the information criticality of each phrase in the interactive text is as follows:

[0027]

[0028] In the formula, H i A represents the informational importance of the i-th phrase in the interactive text. i E represents the content importance coefficient of the i-th phrase in the interactive text. i is the negligible coefficient of the i-th word group in the interactive text; norm(·) is the normalization function; α is a constant to avoid a denominator of 0.

[0029] Preferably, all keywords in the interactive text further include:

[0030] A second segmentation threshold is obtained by thresholding the information criticality of each phrase in the interactive text. Phrases whose information criticality is greater than or equal to the second segmentation threshold are regarded as keywords in the interactive text.

[0031] This application also provides an interactive text parsing system based on a large model intelligent agent, including a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the steps of any of the above-described interactive text parsing methods based on a large model intelligent agent.

[0032] As can be seen from the above, the interactive text parsing method and system based on large-scale intelligent agents provided in this application have at least the following beneficial effects:

[0033] This application addresses the issue of diverse text types input into large-scale intelligent agents. By analyzing the positional and approximate features of phrases within sentences in the converted text received by the large-scale intelligent agent, the importance of individual phrases is determined. Further analysis of part-of-speech confusion and the probability of phrase and part-of-speech combinations is conducted to assess the ignorability of individual phrases. Combined with the traditional TF-IDF algorithm, all keywords within a single interactive text are filtered out, thereby improving the accuracy of large-scale intelligent agents in parsing interactive text. This solves the problem that the traditional TF-IDF algorithm, which only considers word frequency and inverse document frequency as key criteria for keyword filtering, cannot handle the diversity of text types in large-scale intelligent agents. Attached Figure Description

[0034] To more clearly illustrate the technical solutions and advantages in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0035] Figure 1A flowchart illustrating the steps of the interactive text parsing method based on a large model intelligent agent provided in this application;

[0036] Figure 2 A flowchart illustrating the steps of the method for obtaining the negligible coefficient provided in this application;

[0037] Figure 3 A block diagram of the interactive text parsing system based on a large model intelligent agent provided in this application. Detailed Implementation

[0038] To further illustrate the technical means and effects adopted by this application to achieve the intended inventive purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of the interactive text parsing method and system based on large-model intelligent agents proposed in this application. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0039] Unless otherwise specified and limited, terms such as “comprising,” “including,” or any other variations thereof are intended to cover a non-exclusive inclusion, such that a circuit structure, article, or device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such an article or device. Without further limitation, an element defined by the phrase “comprising one…” does not exclude the presence of other identical elements in the article or device that includes said element. Furthermore, the term “and / or” as used herein includes any and all combinations of one or more of the associated listed items. All technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.

[0040] The following section, in conjunction with the accompanying drawings, details the specific scheme of the interactive text parsing method and system based on large-scale intelligent agents provided in this application.

[0041] Please see Figure 1 The diagram illustrates a flowchart of an interactive text parsing method based on a large model intelligent agent according to an embodiment of this application, including the following steps:

[0042] Step 1: Obtain each phrase in the interactive text of the large model agent and extract the part-of-speech tag for each phrase.

[0043] This embodiment takes Chinese input by a user as an example. After the large model agent's data receiving system acquires the user's input, it converts the interactive text according to the type of input data. Text data cleaning removes irrelevant characters such as special symbols and punctuation from the interactive text. Furthermore, a word segmentation tool is used to segment the interactive text into phrases, tagging the parts of speech of each phrase, and completing sentence segmentation. Preferably, the text data cleaning method is not limited to regular expressions and predefined stop word lists. In this embodiment, regular expressions are used for text data cleaning, and the Jieba word segmentation algorithm is used. The specific process will not be elaborated further.

[0044] Furthermore, the part-of-speech tagging of phrases uses n. for nouns, v. for verbs, a. for adjectives, and z. for other parts of speech. For the key content of interactive text, nouns and verbs are far more critical than other parts of speech phrases. Therefore, this embodiment mainly calculates the tags for nouns and verbs, while other phrases are processed using the traditional TF-IDF algorithm. Preferably, in this embodiment, the TF-IDF algorithm is a well-known technique, and its specific process will not be elaborated further.

[0045] Step 2: By analyzing the differences between each phrase in the interactive text and all other phrases of the same part of speech, we obtain the approximate phrases for each phrase in the interactive text. We then count the number of all approximate phrases for each phrase in the interactive text, as well as the number of phrases in the sentences containing each phrase. Finally, we combine the differences between each phrase in the interactive text and the first and last phrases in the sentences containing each phrase to obtain the content importance coefficient of each phrase in the interactive text.

[0046] During user interaction with the large-scale intelligent model, the types of input text vary. These may include professional documents and books, or non-written text. Different types of text may describe the same thing in significantly different ways. For example, if the input is a manual for a liquid chromatograph, then "liquid chromatograph" will appear frequently as the main content. However, for non-written text, the level of rigor in the description may be relatively lower. The same content may be described using different phrases; for instance, "liquid chromatograph" might be described as "chromatograph," "instrument," or "measuring instrument." This means that while some phrases may describe the same content, the overall frequency of these synonymous phrases is low. If these phrases are judged as unimportant, it may affect the accuracy of the text parsing.

[0047] Typically, a sentence contains three main parts: subject, verb, and object, along with some modifiers. The subject and object are usually located close to the beginning and end of the sentence, while the verb is located farther away. Specifically, if a phrase is a noun, the closer it is to the first and last phrases in the sentence, the more likely it is to be the subject or object, and the higher its importance. Similarly, the closer a phrase is to the middle of the sentence, the more likely it is to be the verb, and the higher its importance. Furthermore, the longer the sentence, the more modifiers it contains, the less varied the overall subject-verb-object distribution, and the more prominent the phrase becomes. Additionally, in interactive text, synonymous phrases with different descriptions often share a certain similarity and are used to express the same meaning. The more similar phrases exist within an interactive text, the more important that phrase is.

[0048] To characterize the above features, for each statement in the interactive text, the total number of phrases contained in each statement is counted, and the total number of phrases contained between any two phrases in each statement is measured to obtain the distance between phrases in each statement. For example, if a statement is "liquid chromatograph / should / carefully / perform / check / chromatographic column", then the distance between the phrases "liquid chromatograph" and "check" is 4, meaning it contains "should", "carefully", "perform", and "check". Further, for each phrase in the interactive text, the edit distance of each phrase in the interactive text is calculated with all other phrases of the same part of speech, and all edit distances are subjected to cross-validation for threshold segmentation to obtain a first segmentation threshold. When the edit distance of each phrase with other phrases is less than or equal to the first segmentation threshold, then the two phrases are considered approximate phrases, thus obtaining approximate phrases for each phrase in the interactive text. Preferably, in this embodiment, the edit distance uses an edit distance algorithm, and the cross-validation uses K-fold cross-validation.

[0049] Based on the above analysis, the number of all similar phrases existing in each phrase in the interactive text, as well as the number of phrases in the sentences containing each phrase in the interactive text, are counted. Combined with the differences between each phrase in the interactive text and the first and last phrases of the sentences containing them, the content importance coefficient of the i-th phrase in the interactive text is calculated. In this embodiment, the specific calculation formula is as follows:

[0050]

[0051] In the formula, A i C is the content importance coefficient of the i-th phrase in the interactive text; i B is the total number of phrases in the sentence containing the i-th phrase in the interactive text; ib is the distance between the i-th phrase in the interactive text and the first phrase in the sentence containing it; i D is the distance between the i-th phrase in the interactive text and the last phrase in the sentence containing it; i Z represents the total number of approximate word groups existing for the i-th word group in the interactive text; min(·) is the function that selects the minimum value; |·| is the absolute value operation; i denoted as the part of speech of the i-th phrase in the interactive text; n. is the part of speech of a noun; v. is the part of speech of a verb; α is a constant to avoid a denominator of 0, where the value of α ranges from (0, 0.1), and in this embodiment, α is 0.01.

[0052] Specifically, when the i-th phrase in the interactive text is a noun, the closer it is to the first or last phrase in the sentence it contains, the longer the sentence it contains, and the more similar phrases it contains in the entire text, the more synonymous but different phrases it contains in the interactive text, and the higher its importance in the interactive text. When the i-th phrase in the interactive text is a verb, the closer it is to the first and last phrase in the sentence it contains, the longer the sentence it contains, and the more similar phrases it contains in the interactive text, the more likely it is to be a synonymous but different phrase in the interactive text, and the higher its importance in the interactive text.

[0053] Step 3: Count the number of phrases in the sentences containing each phrase in the interactive text, as well as the number of phrases of each part of speech in the sentences containing each phrase. Obtain the probability of each part of speech appearing in the sentences containing each phrase in the interactive text. Combine each phrase and each part of speech in the interactive text with the phrases before and after it, and with the number of phrase combinations and the number of part of speech combinations, obtain the probability of phrase combinations and the probability of part of speech combinations appearing in the interactive text. Then, obtain the negligible coefficient of each phrase in the interactive text.

[0054] Furthermore, due to the diverse data types of user input, if the original input is an audio file, converting it into interactive text using an audio-to-text tool may result in incomplete interactive text due to various noise interferences in the original audio. These interactive texts may contain missing words, grammatical errors, or recognition mistakes. Consequently, during subsequent word segmentation and sentence recognition, some sentences may appear disorganized or incomplete, resulting in missing phrases. However, some phrases may have high frequency because they are at the beginning or end of a sentence or are part of the content described in the interactive text. In such cases, relying solely on the traditional TF-IDF algorithm, or failing to consider the defects in the interactive text and judging it in a conventional way, may misidentify these phrases as keywords, leading to an inaccurate transmission of the original meaning of the sentences during interactive text parsing and affecting the accuracy of the parsing. The audio-to-text tool is not limited to TurboScribe, MemoAI, or RealtimeSTT; this embodiment uses MemoAI.

[0055] Specifically, when the above situations occur, some phrases in the interactive text will have a low degree of relevance to the surrounding phrases. After sentence segmentation of the interactive text, these phrases may be assigned to sentences with low content relevance, causing changes in the original part-of-speech combinations of the phrases in the entire sentence. In addition, due to omissions in the interactive text, the collocation of individual phrases with other phrases may also be abnormal. For example, the original phrase "observe the chromatogram column" may become "change the chromatogram column," etc. Besides changes in phrase collocation, deviations in part-of-speech collocation may also occur. For example, the original adjective and noun collocation may become a noun-noun collocation, etc. Specifically, the part-of-speech distribution of phrases in a single sentence will show significant differences. For example, a sentence may contain 10 phrases, of which 9 are noun phrases, and the probability of individual phrase combinations and part-of-speech combinations appearing in the entire text is low.

[0056] Therefore, taking a single sentence in the interactive text as an example, we obtain the part of speech of all phrases in each sentence in the interactive text, count the total number of phrases of each part of speech in each sentence, and use the ratio between the total number of phrases of each part of speech and the total number of phrases in each sentence as the probability of the occurrence of phrases of each part of speech in the sentence containing each phrase in the interactive text.

[0057] Furthermore, taking a single phrase as an example, we combine adjacent phrases into a single phrase combination. The combination of each phrase in the interactive text with its preceding phrase is considered the preceding phrase combination; and the combination of each phrase in the interactive text with its following phrase is considered the following phrase combination. We then count the total number of phrase combinations identical to the preceding phrase combination within the interactive text. The ratio of this total to the total number of phrases in the interactive text is taken as the probability of the preceding phrase combination appearing in the interactive text. Similarly, the probability of the following phrase combination appearing in the interactive text can be obtained. Finally, we sum the probabilities of the preceding and following phrase combinations in the interactive text to obtain the probability of each phrase combination appearing in the interactive text.

[0058] Accordingly, the method described above in this embodiment is repeated, and the part-of-speech of a single phrase is analyzed to obtain the probability of the occurrence of part-of-speech combinations of each phrase in the interactive text.

[0059] Based on the above analysis, and considering the dispersion of part-of-speech phrases within the sentences containing each phrase in the interactive text, as well as the fluctuations in part-of-speech combinations and phrase combinations, the negligible coefficient of each phrase in the interactive text is calculated. In this embodiment, the specific calculation formula is as follows:

[0060]

[0061] In the formula, E i F is the ignorability coefficient of the i-th phrase in the interactive text; i G represents the variance of the probability of each part of speech appearing in the sentence containing the i-th phrase in the interactive text; i g represents the probability of part-of-speech combinations of the i-th phrase in the interactive text. i Let be the probability of the i-th word combination appearing in the interactive text.

[0062] Specifically, the larger the variance of the probability of the i-th phrase appearing in the sentence containing the phrase of a particular part of speech, and the lower the probability of the phrase's part-of-speech combination and its combination in the interactive text, the higher the degree of confusion in the sentence containing the phrase. This indicates that the position of the phrase is more likely to have sentence content confusion and low sentence completion, and therefore the impact of the phrase on the article content should be ignored. Furthermore, the flowchart of the method for obtaining the negligible coefficient provided in this embodiment is as follows: Figure 2 As shown.

[0063] Step 4: Based on the ignorability coefficient of each phrase in the interactive text, and combined with the content importance coefficient of each phrase in the interactive text, obtain the information keyness of each phrase in the interactive text. This information is then used to extract all keywords from the interactive text, and to use these keywords to parse the interactive text of the large model agent.

[0064] Furthermore, based on the content importance coefficient of each phrase in the interactive text, and combined with the ignorability coefficient of each phrase in the interactive text, the information criticality of each phrase in the interactive text is calculated. In this embodiment, the specific calculation formula is as follows:

[0065]

[0066] In the formula, H i A represents the informational importance of the i-th phrase in the interactive text; i E represents the content importance coefficient of the i-th phrase in the interactive text. i is the negligible coefficient of the i-th word group in the interactive text; norm(·) is the normalization function; α is a constant to avoid a denominator of 0.

[0067] Specifically, the higher the importance of the content contained in the i-th phrase in the interactive text and the lower the degree of ignorability, the more important the content represented by the phrase in the interactive text is, and the more it should be listed as a keyword during the interactive text parsing process.

[0068] The information criticality of each phrase in the interactive text can be calculated using the above method. Using all information criticality values ​​as input, cross-validation is employed to threshold the information criticality, resulting in a second segmentation threshold. When the information criticality is greater than or equal to the second segmentation threshold, the phrase is considered a keyword and marked; otherwise, it is considered negligible, thus filtering out all keywords in the interactive text. Preferably, in this embodiment, K-fold cross-validation is used; the specific process will not be elaborated further.

[0069] Furthermore, for the remaining word groups in the interactive text, excluding verb phrases and noun phrases, the TF-IDF algorithm is used to calculate and further determine the keyword phrases in the remaining word groups, thereby obtaining all keywords in the interactive text. Preferably, in this embodiment, the TF-IDF algorithm is a well-known technology, and the specific process will not be described in detail.

[0070] Preferably, in this embodiment, the interactive text parsing system for the large model agent includes a text parsing module, an interaction module for the large model agent, and a deep learning-based intent recognition module. Specifically, the system first identifies all keywords in each sentence of the interactive text and inputs the keyword-annotated interactive text into the text parsing module. The text parsing module uses dependency parsing, short syntax parsing, and contextual understanding models to parse the interactive text based on the keywords annotated in the interactive text. After error correction and optimization, it obtains the parsing result of the interactive text and inputs the parsing result into the interaction module of the large model agent. Further, the interaction module, based on the parsing result and in conjunction with the deep learning-based intent recognition module, identifies the user's input intent and then uses it with a generative artificial intelligence model and a predefined interactive text database to generate output interactive text. The algorithm models described above are all well-known technologies. Implementers can choose the appropriate model based on actual accuracy requirements, response speed requirements, hardware performance, etc. Dependency parsing is not limited to corpus-based, statistical model-based, or neural network-based dependency parsing methods. Context understanding models are not limited to BERT or RoBERTa. In this embodiment, corpus-based dependency parsing is used for dependency parsing, PCFG model is used for phrase parsing, and BERT model is used for context understanding. The specific process will not be described in detail.

[0071] Based on the same inventive concept as the above method, this application also provides an interactive text parsing system based on a large model intelligent agent, including a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the steps of any of the above-described interactive text parsing methods based on a large model intelligent agent.

[0072] Specifically, the interactive text parsing system for large model agents in this embodiment includes a text parsing module, an interaction module for large model agents, and an intent recognition module based on deep learning. The block diagram of the interactive text parsing system based on large model agents is shown below. Figure 3 As shown.

[0073] It is understood that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.

[0074] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0075] The above description is merely an embodiment of this application and is not intended to limit the scope of this application. Any equivalent structural or procedural transformations made based on the description and drawings of this application, or direct or indirect applications in other related technical fields, are similarly included within the protection scope of this application.

Claims

1. An interactive text parsing method based on large-scale intelligent agents, characterized in that, Includes the following steps: Obtain each word group from the interactive text of a large model agent and extract the part-of-speech tag for each word group; By analyzing the differences between each phrase in the interactive text and all other phrases of the same part of speech, we obtain the approximate phrases of each phrase in the interactive text. We then count the number of all approximate phrases of each phrase in the interactive text, as well as the number of phrases in the sentences containing each phrase. Finally, we combine the differences between each phrase in the interactive text and the first and last phrases in the sentences containing each phrase to obtain the content importance coefficient of each phrase in the interactive text. The number of phrases in the sentences containing each phrase in the interactive text and the number of phrases of each part of speech in the sentences are counted. The probability of each phrase of each part of speech appearing in the sentences containing each phrase in the interactive text is obtained. Combined with the phrases and phrases of each part of speech in the interactive text with the phrases before and after them, and the probability of phrase combination and phrase combination appearing in the interactive text are obtained based on the number of phrase combinations and the number of phrase combinations. Then, the negligible coefficient of each phrase in the interactive text is obtained. Based on the ignorability coefficient of each phrase in the interactive text, and combined with the content importance coefficient of each phrase in the interactive text, the information keyness of each phrase in the interactive text is obtained, which is used to extract all keywords in the interactive text, so as to use keywords to parse the interactive text of the large model agent. The method for calculating the content importance coefficient of each phrase in the interactive text is as follows: ； In the formula, is the content importance coefficient of the i-th phrase in the interactive text; This represents the total number of phrases in the sentence containing the i-th phrase in the interactive text. The distance between the i-th phrase in the interactive text and the first phrase in the sentence containing it; The distance between the i-th phrase in the interactive text and the last phrase in the sentence containing it; This represents the total number of all similar word groups that exist for the i-th word group in the interactive text. A function that selects the minimum value; This is for absolute value operations; Let i be the part of speech of the i-th phrase in the interactive text; It is a noun. It is a verb. To avoid constants with a denominator of 0; The method for calculating the ignorability coefficient of each phrase in the interactive text is as follows: ； In the formula, is the negligible coefficient of the i-th word group in the interactive text; Let $\mathbf{i}$ be the variance of the probability of each part of speech appearing in the sentence containing the $i$-th phrase in the interactive text. Let be the probability of the part-of-speech combination of the i-th word group in the interactive text; Let be the probability of the i-th word combination appearing in the interactive text.

2. The interactive text parsing method based on a large model intelligent agent as described in claim 1, characterized in that, The similar word groups in the interactive text further include: The edit distance between each word group in the interactive text and all other word groups with the same part of speech is calculated. All the edit distances are then thresholded to obtain the first segmentation threshold. Word groups whose edit distances to each word group are less than or equal to the first segmentation threshold are considered as approximate word groups of each word group.

3. The interactive text parsing method based on a large model intelligent agent as described in claim 1, characterized in that, The method for obtaining the probability of each part of speech appearing in the sentence containing each phrase in the interactive text is as follows: The ratio of the total number of phrases of each part of speech in the sentence containing each phrase in the interactive text to the total number of phrases in the sentence containing each phrase is taken as the probability of each phrase of each part of speech appearing in the sentence containing each phrase in the interactive text.

4. The interactive text parsing method based on a large model intelligent agent as described in claim 1, characterized in that, The probability of phrase combinations and the probability of part-of-speech combinations in the interactive text further include: Each pair of adjacent word groups is combined into a word group combination. The preceding word group combination and the following word group combination of each word group are extracted. The ratio of the total number of word group combinations in the interactive text that are the same as the preceding word group combination and the following word group combination of each word group to the total number of word groups in the interactive text is used as the probability of the preceding word group combination and the following word group combination of each word group appearing in the interactive text. The sum of the two is used as the probability of the word group combination of each word group appearing. Accordingly, the part-of-speech of each phrase is obtained using the same process as the probability of phrase combination occurrence, thus obtaining the probability of part-of-speech combination occurrence of each phrase in the interactive text.

5. The interactive text parsing method based on a large model intelligent agent as described in claim 4, characterized in that, The method for extracting the preceding and following word combinations of each phrase is as follows: Each phrase in the interactive text is combined with its preceding and following phrases to obtain the preceding and following phrase combinations for each phrase.

6. The interactive text parsing method based on a large model intelligent agent as described in claim 1, characterized in that, The method for calculating the information criticality of each phrase in the interactive text is as follows: ； In the formula, The importance of the i-th phrase in the interactive text; is the content importance coefficient of the i-th phrase in the interactive text; is the negligible coefficient of the i-th word group in the interactive text; This is the normalization function; To avoid constants with a denominator of 0.

7. The interactive text parsing method based on a large model intelligent agent as described in claim 1, characterized in that, All keywords in the interactive text further include: A second segmentation threshold is obtained by thresholding the information criticality of each phrase in the interactive text. Phrases whose information criticality is greater than or equal to the second segmentation threshold are regarded as keywords in the interactive text.

8. An interactive text parsing system based on a large model intelligent agent, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the interactive text parsing method based on a large model agent as described in any one of claims 1-7.

Images