Chinese language and literature text theme analysis system based on big data

Abstract

The invention relates to the field of natural language processing, and discloses a Chinese language and literature text topic analysis system based on big data, and the system comprises a core semantic primitive library construction module which is used for extracting and constructing a knowledge base containing semantics, culture and timestamps from a meta-poeia corpus offline; the text multi-dimensional analysis module is used for extracting four-dimensional features of grammar, core semantics, intertext and emotion time sequence of the to-be-analyzed text in parallel; the feature collaborative analysis module is used for fusing multi-dimensional features, performing context dynamic activation on primitives, and innovatively quantifying and detecting deep semantic conflicts by calculating the dispersion of feature vectors in a semantic space; and the topic report generation module is used for synthesizing a macroscopic topic, integrating all analysis results and generating a structured multi-level report. The system can objectively and deeply reveal complex and even contradictory theme connotation in literature works through modular cooperation, and the accuracy and depth of text analysis are remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of natural language processing, and particularly to a Chinese literature text theme analysis system based on big data. BACKGROUND

[0002] Theme analysis of Chinese literature works is a core task in the field of humanities research, aiming to deeply mine the author's intention, thoughts, emotions and cultural values hidden behind the text. For a long time, the analysis work in this field mainly relies on the deep learning and literature accomplishment of researchers, and is carried out through artificial close reading and interpretation. However, the limitations of this traditional method are increasingly prominent when facing the growing digital literature resources. The analysis conclusion often has strong subjective color, and different researchers may have different interpretations of the same text, lacking reproducible objective standards. At the same time, the artificial analysis is low in efficiency, and it is difficult to systematically and mode-ly study a large-scale text set, and it is easy to overlook the macro rules hidden in the data across texts.

[0003] With the development of computer technology, natural language processing (NLP) technology is introduced into literary text analysis, such as using keyword frequency statistics, theme model (such as LDA) and other methods to assist theme discovery. Although these technologies have improved the objectivity and efficiency of analysis to some extent, there are still significant deficiencies in the depth and accuracy of analysis. Most of the existing computing methods are limited to shallow statistics of words, treating text as a bag-of-words, ignoring syntactic structure and contextual information that determine the meaning of text, and thus it is difficult to accurately grasp the dynamic semantics of words in specific context. More importantly, they generally lack the ability to handle the complexity of literary works, and cannot effectively identify and interpret the semantic tension brought by the artistic techniques such as deep contradictions, irony or symbolism within the text. The cultural connotation of a word, the dynamic change of its emotional color, and the intertextual relationship between the text and other classic works, etc. These multi-dimensional information are crucial for understanding the theme of literature, and are often fragmented or directly ignored in existing technologies.

[0004] Therefore, there is an urgent need in the art for a new technical solution that can efficiently and objectively process large-scale text data, and deeply integrate multi-dimensional text features to reveal the complex and even contradictory deep theme connotation hidden in literary works. SUMMARY

[0005] The technical problem to be solved by the present application is that the existing technology is difficult to effectively integrate multi-dimensional features such as semantics, emotions and intertextuality of text when performing text theme analysis, so as to reveal the deep semantic conflicts and complex theme connotation caused by the discordance of these features within the text.

[0006] To address the aforementioned technical problems, this invention provides a Chinese language and literature text theme analysis system based on big data.

[0007] The first aspect of this invention provides a Chinese language and literature text theme analysis system based on big data, comprising:

[0008] The core semantic primitive library construction module is used to extract core semantic primitives from the meta-canonical corpus and construct primitive identity vectors containing semantic vectors, cultural attribute vectors and timestamps for the core semantic primitives to form the core semantic primitive library;

[0009] The text multidimensional analysis module is used to perform parallel syntactic structure analysis, core semantic primitive recognition, intertextuality analysis, and sentiment temporal analysis on the text to be analyzed in order to obtain four corresponding feature information.

[0010] The feature collaborative analysis module, connected to the text multidimensional analysis module, is used to dynamically activate the primitive identity vectors in the core semantic primitive library based on the context of the text to be analyzed, and to compare the feature information from the four analyses in parallel to detect semantic conflicts between them.

[0011] The topic report generation module is used to locate the core topic and generate a structured analysis report based on the semantic consistency or semantic conflict detected by the feature collaborative analysis module.

[0012] Preferably, the core semantic primitive library construction module is trained on the Yuan Dian corpus using a word embedding model to obtain the semantic vector; and the cultural attribute vector is formed by encoding predefined cultural attributes into numerical values.

[0013] Preferably, the intertextuality analysis performed by the text multidimensional analysis module is achieved by running an intertextual gravity field model. The model uses the source text S from the Yuandian corpus... j Represented as the gravitational center vector And calculate the text segment vectors in the text to be analyzed. The intertextual gravitational value F between the gravitational center vector and the said gravitational force. grav The calculation formula is:

[0014]

[0015] Preferably, when the feature collaborative analysis module dynamically activates the primitive identity vector, it encodes the local context of the core semantic primitives in the text to be analyzed into a context adjustment vector. And use this vector to analyze the cultural attribute vector in the primitive identity vector. Adjustments are made to generate activation vectors. The calculation formula is as follows:

[0016]

[0017] wherein, ⊙ denotes Hadamard product, M w is a preset weight matrix.

[0018] Preferably, when the feature coordination analysis module is dynamically activated, it further adjusts the semantic vector according to the timestamp in the base element identity vector and the creation year of the text to be analyzed by a preset adjustment function, so as to realize semantic drift correction of the core semantic base element.

[0019] Preferably, when the feature coordination analysis module is performing semantic drift correction, it further adjusts the correction degree of the adjustment function according to the intertextual attraction value obtained by the intertextuality analysis performed by the text multi-dimensional analysis module.

[0020] Preferably, when the feature coordination analysis module detects semantic conflicts, the feature information obtained by the four kinds of analysis is mapped into a signal vector respectively, and the variance of the group of signal vectors is calculated; when the variance is greater than a preset conflict threshold, the corresponding text position is marked as a semantic conflict point. The signal vector used for calculation, corresponding to the core semantic base element recognition, is the signal corresponding to the activated vector generated after the context dynamic activation of the base element identity vector.

[0021] Preferably, when receiving the marking of the semantic conflict point, the theme report generation module generates a hypothetical explanation about the specific artistic method of expressing semantic inconsistency used by the text to be analyzed at this position.

[0022] Preferably, the structured analysis report generated by the theme report generation module adopts the form of argument-persuasion, takes the analyzed theme as the argument, and presents the feature information obtained by the four kinds of analysis as the persuasion.

[0023] The second aspect of the present application provides a Chinese literature text theme analysis method based on big data, comprising the following steps:

[0024] Step one: constructing a core semantic base element library, which includes extracting core semantic base elements from a meta-class corpus and constructing a base element identity vector containing semantic vectors, cultural attribute vectors and timestamps for the core semantic base elements;

[0025] Step two: multi-dimensional analysis of the input text to be analyzed, which includes parallelly performing syntax structure analysis, core semantic base element recognition, intertextuality analysis and emotion time sequence analysis to obtain four kinds of corresponding feature information;

[0026] Step 3: Perform feature co-analysis. This step includes dynamically activating the primitive identity vectors in the core semantic primitive library based on the context of the text to be analyzed, and comparing feature information from the four analyses in parallel to detect semantic conflicts between them; and

[0027] Step 4: Based on the detected semantic consistency or semantic conflict, locate the core topic and generate a structured analysis report.

[0028] This invention provides a Chinese language and literature text topic analysis system based on big data. It has the following beneficial effects:

[0029] 1. This invention improves the accuracy of analyzing symbolic and dynamically evolving semantics in literary texts. By constructing a core semantic primitive library containing cultural attribute vectors and timestamps, and utilizing a feature-based collaborative analysis module to dynamically activate primitives based on textual context and the year of creation, this system can handle the symbolic meaning changes of the same core semantic primitive in different contexts and semantic drift across different historical periods. This allows the analysis results to transcend literal meaning and more closely approximate the deeper implications of the text.

[0030] 2. This invention enhances the ability to identify and quantify potential intertextual relationships between texts. By employing an intertextual gravity field model, the source text is represented as a gravity center vector, and the similarity between the text segment to be analyzed and this vector is calculated to obtain the intertextual gravity value. This system can identify conceptual or thematic references that do not rely on direct lexical overlap. This provides a quantifiable technical basis for analyzing the inheritance and development relationships between literary works.

[0031] 3. This invention provides a technical approach to identifying and interpreting specific artistic techniques. By processing feature information across four dimensions—grammar, sentiment, core semantic primitives, and intertextuality—in parallel, and utilizing a feature co-analysis module to calculate semantic conflicts between multi-dimensional features, this system can objectively locate the structural tensions in a text caused by semantic inconsistencies. This makes it possible to technically interpret artistic techniques such as irony that rely on semantic contradictions, thus deepening the level of thematic analysis. Attached Figure Description

[0032] Figure 1 This is a schematic diagram of the functional modules of a Chinese language and literature text theme analysis system based on big data, according to an embodiment of the present invention.

[0033] Figure 2 This is a functional diagram of a text multidimensional analysis module according to an embodiment of the present invention;

[0034] Figure 3 This is a flowchart of a text topic analysis method according to an embodiment of the present invention.

[0035] The module includes: 10. Core semantic primitive library construction module; 20. Text multidimensional analysis module; 21. Syntactic structure analysis submodule; 22. Core semantic primitive recognition submodule; 23. Intertextuality analysis submodule; 24. Sentiment temporal analysis submodule; 30. Feature collaborative analysis module; and 40. Topic report generation module. Detailed Implementation

[0036] To make the objectives, technical solutions, and advantages of the present invention clearer, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be noted that, unless otherwise specified, the embodiments and features described herein can be combined with each other. The specific embodiments described herein are merely illustrative of the present invention and not intended to limit the scope of the invention.

[0037] See attached document Figure 1 , Figure 1 This is a functional module diagram of a Chinese language and literature text topic analysis system based on big data, according to an embodiment of the present invention. The present invention provides a Chinese language and literature text topic analysis system based on big data, which can be deployed on a single server, a distributed server cluster, or a cloud computing environment. The system may include: a core semantic primitive library construction module 10, a multi-dimensional text analysis module 20, a feature collaborative analysis module 30, and a topic report generation module 40.

[0038] The core semantic primitive library construction module 10 performs an offline construction process. This module receives one or more meta-classical corpora as input, extracts core semantic primitives from the corpora using a preset extraction algorithm, and generates a primitive identity vector for each extracted core semantic primitive. All core semantic primitives and their corresponding primitive identity vectors are stored to form the core semantic primitive library. This core semantic primitive library is then output to the feature collaborative analysis module 30.

[0039] The text multidimensional analysis module 20 receives an externally input Chinese literary text to be analyzed. This module integrates four parallel analysis sub-modules, which respectively perform grammatical structure analysis, core semantic primitive recognition, intertextuality analysis, and sentiment temporal analysis on the text to be analyzed. The output of this module is four corresponding feature information sets, which are then transmitted to the feature co-analysis module 30.

[0040] The feature collaborative analysis module 30 has its input terminals connected to the output terminals of the core semantic primitive library construction module 10 and the text multidimensional analysis module 20, respectively. This module receives the core semantic primitive library provided by the core semantic primitive library construction module 10, and four types of feature information provided by the text multidimensional analysis module 20.

[0041] The feature co-analysis module 30 performs two core computational tasks.

[0042] The first task is to dynamically activate the primitive identity vectors corresponding to the core semantic primitives identified in the text, based on the context of the text to be analyzed.

[0043] The second task is to compare feature information from four analyses in parallel to detect whether there are semantic conflicts between feature information from different dimensions at specific locations in the text.

[0044] The output of this module is a set of analysis results, which includes a determination of the semantic consistency of different parts of the text and the location markers of all detected semantic conflict points. This set of analysis results is transmitted to the topic report generation module 40.

[0045] The topic report generation module 40 has its input end connected to the output end of the feature co-analysis module 30. This module receives the analysis results transmitted by the feature co-analysis module 30. Its function is to locate the core topic of the text based on the semantic consistency regions and semantic conflict points marked in the analysis results, and generate a structured analysis report as the final output according to a preset report template.

[0046] In a specific workflow, the system first constructs the core semantic primitive library through the core semantic primitive library construction module 10. When text analysis is required, the text to be analyzed is input into the system and processed by the text multidimensional analysis module 20, generating four types of feature information: syntactic structure, core semantic primitives, intertextuality, and sentiment temporal sequence. Subsequently, the feature co-analysis module 30 receives these four types of feature information and the pre-constructed core semantic primitive library, performs dynamic activation of primitives and detection of semantic conflicts, and sends the judgment results to the topic report generation module 40. Finally, the topic report generation module 40 generates and outputs the report based on these judgment results.

[0047] In this embodiment, the specific functions of the core semantic primitive library construction module 10 are implemented through the following steps:

[0048] Step 1: Construction and preprocessing of the Yuandian Corpus.

[0049] This step gathers classical Chinese literature from one or more pre-defined databases. The literature includes texts from the four categories of classics, history, philosophy, and literature, as well as anthologies of poetry and prose from various dynasties and representative works of important literary schools. After acquiring the classical literature, the system performs standardized preprocessing operations. These operations include: converting all text characters to simplified Chinese; automatically punctuating the text and segmenting it according to classical Chinese grammar rules to form standardized word sequences; and removing words from a pre-defined stop word list.

[0050] Step 2: Based on the preprocessed meta-canonical corpus, extract the core semantic primitives.

[0051] The extraction process is automated through a hybrid scoring algorithm, which aims to select high-value words or phrases with high recurrence rates and stable semantic structures from the corpus.

[0052] First, the system extracts all N-grams of length 1 to N from the corpus as a candidate set, where N is a preset integer, such as 4.

[0053] For each candidate core semantic primitive p, the system calculates its comprehensive importance score, Score(p). This score is obtained by a weighted sum of three components, calculated using the following formula:

[0054] Score(p) = α·S tfidf (p)+β·S pmi (p)+γ·S tr (p);

[0055] Where p represents a candidate core semantic primitive. S tfidf (p) is the term frequency-inverse document frequency (TF-IDF) score of candidate primitive p, used to measure its importance in a single document and its prevalence in the entire corpus. pmi (p) is the mean of the internal pointwise mutual information of the candidate primitive p. If p is a word containing multiple words (w1, w2, ..., w...), then p is a pointwise mutual information of the candidate primitive p. k The component of a phrase is used to measure the degree of cohesion between its constituent words. tr (p) is the TextRank score of candidate primitive p, obtained by constructing a word graph of the text and running a ranking algorithm, used to measure its importance in the text structure. α, β, and γ are preset, non-negative weight coefficients, and their sum is 1, used to adjust the proportion of the three components in the final score.

[0056] After calculating the overall importance score of all candidate primitives, the system sets a score threshold T. score All satisfying Score(p)≥T score The candidate primitive p is officially identified as the core semantic primitive and stored in an initial list for use in subsequent primitive identity vector construction steps.

[0057] Step 3: After extracting the core semantic primitives, the core semantic primitive library construction module 10 generates a corresponding primitive identity vector V for each core semantic primitive p in the list. p .

[0058] This vector is used to provide an initial, static characterization description of the core semantic primitive from multiple dimensions.

[0059] Primitive identity vector V p It is a data structure composed of three cascaded parts, and its structure is as follows:

[0060]

[0061] Among them, V p The primitive identity vector representing the core semantic primitive p. It is a high-dimensional semantic vector. It is a cultural attribute vector. t orig It is a timestamp. || represents a cascading operation between vectors or data.

[0062] In this embodiment, semantic vectors The generation process is as follows:

[0063] The system employs a pre-defined word embedding model, such as Word2Vec or GloVe, and trains the model on a pre-processed meta-canonical corpus.

[0064] After training, for each core semantic primitive p, it is input into the trained model. The model outputs a dense numerical vector of fixed dimensions (e.g., 300 dimensions), which is the semantic vector of the core semantic primitive p.

[0065] The position and orientation of this vector represent the basic semantics of the primitive in the overall context of the meta-canonical corpus within the vector space.

[0066] In this embodiment, the cultural attribute vector The generation process is as follows:

[0067] The system predefines a list containing k cultural attributes, such as [homesickness, sorrow of parting, boldness, satire, loyalty to the emperor, historical reflection, ...].

[0068] Cultural attribute vector It is a k-dimensional vector, where each dimension corresponds one-to-one with a cultural attribute in the list.

[0069] The value w of the vector in the i-th dimension i This represents the initial association strength between the core semantic primitive and the i-th cultural attribute.

[0070] This value is determined in the following way:

[0071] First, a seed set containing several strongly related words is manually defined for each cultural attribute;

[0072] Then calculate the semantic vector of the core semantic primitive p. The cosine similarity between the word and the mean semantic vector of all words in each cultural attribute seed set is used, and this similarity value is used as the initial association strength w in the corresponding dimension. i .

[0073] In this embodiment, the timestamp t orig The acquisition process is as follows:

[0074] Each document in the Yuandian Corpus is accompanied by metadata about the date of its creation.

[0075] The system searches the entire corpus to find the earliest date in which the core semantic primitive p first appears in all literature.

[0076] The year is recorded as a numerical value, such as in the Gregorian calendar, and assigned to the timestamp t. orig .

[0077] This timestamp is used to characterize the initial historical period when the core semantic primitive entered literary use and formed a stable meaning.

[0078] Finally, the system generates semantic vectors from the above steps. Cultural attribute vector and timestamp t orig The primitive identity vectors are concatenated in a predetermined order to form a complete primitive identity vector V. p .

[0079] Each core semantic primitive p and its corresponding primitive identity vector V p Each record constitutes a single record, and all records are stored together in the database to form the final core semantic primitive library.

[0080] See attached document Figure 2 , Figure 2 This is a functional diagram of a text multidimensional analysis module according to an embodiment of the present invention. Upon receiving externally input text to be analyzed, the text multidimensional analysis module 20 activates its four internally integrated parallel analysis submodules. The text multidimensional analysis module 20 includes: a syntactic structure analysis submodule 21, a core semantic primitive recognition submodule 22, an intertextuality analysis submodule 23, and a sentiment temporal analysis submodule 24.

[0081] The syntax structure analysis submodule 21 is used to perform syntax structure analysis on the text to be analyzed. In a specific embodiment, the implementation process of syntax structure analysis is as follows:

[0082] First, the text to be analyzed is segmented into independent sentence units. For each sentence, the grammatical structure analysis submodule 21 calls a pre-trained dependency parsing model for processing. This model has been trained on a dataset labeled with the grammatical structures of Chinese literary texts to enable it to handle specific syntactic structures in classical or modern literary works.

[0083] The dependency parsing model takes a sentence as input and outputs the dependency parsing tree of that sentence.

[0084] This tree-like data structure identifies the part of speech of each word in a sentence, as well as the dependency relationships between words, such as subject-verb (SBV), verb-object (VOB), and attributive-head (ATT).

[0085] After the analysis is completed, the syntax structure analysis submodule 21 does not directly output the original syntax tree, but converts the analysis results into a structured feature information.

[0086] Specifically, for each word in the text to be analyzed, the system generates a corresponding grammatical feature tuple. This tuple contains at least the part-of-speech of the word and its dependency relation type in the sentence. This feature information is organized into a sequence of data aligned with the original word sequence and then transmitted to the feature co-analysis module 30 as one of the four input signals for subsequent semantic conflict detection.

[0087] In parallel with the grammatical structure analysis, the core semantic primitive recognition submodule 22 in the text multidimensional analysis module 20 performs a recognition operation on the same text to be analyzed. The purpose of this operation is to locate all text fragments in the text that match the entries already stored in the core semantic primitive library 10.

[0088] In a specific embodiment, in order to implement the recognition function, the recognition submodule 22 first reads all the core semantic primitives from the core semantic primitive library 10 during the initialization phase.

[0089] Subsequently, this submodule uses these core semantic primitives as a dictionary to construct an Aho-Corasick automaton. This automaton is a finite state machine that contains structural information of all core semantic primitives and pre-computes failure pointers for state transitions to support single-pass recognition of the input text.

[0090] After receiving the text to be analyzed, the recognition submodule 22 inputs the segmented text word sequence word by word into the constructed Aho-Corasick automaton. The automaton performs state transitions based on the current word and the current state.

[0091] When the automaton reaches one or more states while processing the text stream, and these states are marked as the end of one or more core semantic primitives in the dictionary, the recognition submodule 22 records a successful match. For each match, the system generates a recognition record. This record contains a unique identifier of the recognized core semantic primitive (which corresponds to an entry in the core semantic primitive library 10), as well as the start and end indexes of the primitive in the text to be analyzed.

[0092] After traversing all the text to be analyzed, the recognition submodule 22 compiles all the generated recognition records into a list. This list constitutes the feature information of the core semantic primitive dimension and is transmitted to the feature co-analysis module 30 as one of the four input signals for subsequent analysis.

[0093] In parallel with the aforementioned analysis, the intertextuality analysis submodule 23 within the text multidimensional analysis module 20 performs intertextuality analysis on the same text to be analyzed. The purpose of this analysis is to quantify the potential correlation between the text to be analyzed and the source texts in the meta-canonical corpus by running an intertextual gravity field model.

[0094] In one specific embodiment, the construction and operation of this intertextual gravitational field model includes the following steps:

[0095] First, during the offline phase, the system selects a set of pre-defined, representative source texts S from the Yuandian Corpus. j , where j is the index of the source text. For each source text S j The system segments the content into words and uses the same word embedding model as in the core semantic primitive library building module 10 to convert all words in the source text into semantic vectors. By calculating the average of these semantic vectors, a gravity center vector representing the overall semantics of the source text is generated. The gravitational center vector of all source texts It is stored.

[0096] Upon receiving the text to be analyzed, the intertextuality analysis submodule 23 first segments the text into continuous text fragments d according to preset rules. i , where i is the index of the text segment. The segmentation rule can be segmentation by paragraph, or segmentation using a sliding window with a fixed number of sentences.

[0097] For each text segment d i The system uses the same method as generating the gravitational center vector to convert it into a text fragment vector. Subsequently, the system calculates the vector of the text fragment. With each pre-stored gravitational center vector The intertextual attraction value F between them grav. This value is obtained by calculating the cosine similarity between two vectors, and the calculation formula is:

[0098]

[0099] where F grav (d i , S j ), S j ) is the intertextual gravitational value of the text segment d i relative to the source text S j . is the vector representation of the text segment d i . is the gravitational center vector of the source text S j . · represents the dot product of vectors. ||·|| represents the Euclidean norm of vectors.

[0100] For each text segment d i , the system traverses all the source texts S j and finds a source text that maximizes its intertextual gravitational value F grav . The maximum gravitational value and its corresponding source text identifier are used as the result of the intertextuality analysis of the text segment d i .

[0101] Finally, the intertextuality analysis sub-module 23 outputs a sequence of feature information aligned with the text segment sequence. Each element in this sequence contains the maximum intertextual gravitational value of the corresponding text segment and the source text identifier with the strongest association. This sequence of feature information is transmitted to the feature collaborative analysis module 30 as one of the four input signals for subsequent analysis.

[0102] In parallel with the foregoing analysis, the sentiment time series analysis sub-module 24 in the text multi-dimensional analysis module 20 performs sentiment time series analysis on the same text to be analyzed. This analysis aims to generate a sequence of changes in sentiment intensity corresponding to the text process.

[0103] In a specific embodiment, the function of the sentiment time series analysis sub-module 24 is based on a pre-constructed Chinese language and literature sentiment dictionary. This dictionary stores a series of word entries, each entry containing a sentiment word, a basic sentiment score representing the sentiment polarity and intensity (for example, a floating point number between -1.0 and +1.0, where a negative number represents negative sentiment and a positive number represents positive sentiment), and a词性标签. This dictionary also specifically marks negative words (such as "不", "未") and degree adverbs (such as "极", "略", "甚").

[0104] After receiving the text to be analyzed, the sentiment time series analysis sub-module 24 first divides the text into independent sentence units according to the same rules as the syntactic structure analysis sub-module 21. 请注意,原文中“词性标签”未给出具体英文表述,我保留了中文。你可根据实际情况替换。

[0105] For each sentence unit, submodule 24 executes a sentiment score calculation process. This process iterates through every word in the sentence. When a sentiment word from the dictionary is detected, the system extracts its base sentiment score. Simultaneously, the system checks if a negative word or degree adverb exists within a preset window (e.g., the first two words) preceding the sentiment word. If a negative word exists, the base sentiment score of that word is multiplied by -1. If a degree adverb exists, the base sentiment score is adjusted multiplicatively according to the predefined weighting coefficient of that adverb (e.g., 1.5 for "extremely" and 0.8 for "slightly").

[0106] After calculating the adjusted scores of all sentiment words in the sentence, the system calculates the overall sentiment score of the sentence unit using an aggregation function (such as averaging or weighted summation).

[0107] This process is repeated for all sentence units. Finally, the sentiment temporal analysis submodule 24 outputs a sequence of sentiment scores, where each score corresponds one-to-one with a sentence unit in the text to be analyzed. This feature information sequence is transmitted to the feature co-analysis module 30 as one of the four input signals for subsequent semantic conflict detection.

[0108] After receiving the four types of feature information generated by the text multidimensional analysis module 20 and the core semantic primitive library provided by the core semantic primitive library construction module 10, the feature collaborative analysis module 30 first performs a context dynamic activation operation on each core semantic primitive identified in the text.

[0109] In a specific embodiment, when the core semantic primitive recognition submodule 22 identifies a core semantic primitive p at a specific location in the text to be analyzed, the feature collaborative analysis module 30 retrieves the static primitive identity vector corresponding to p from the core semantic primitive library 10. Meanwhile, module 30 extracts a sequence of words within a preset window surrounding the core semantic primitive p from the text to be analyzed. This window may contain, for example, five words before and after p.

[0110] Based on this word sequence, the system calculates a context adjustment vector. This calculation is performed by obtaining the semantic vector of each word within the window (using the same word embedding model used when building the core semantic primitive library) and calculating the arithmetic mean of these semantic vectors. The resulting vectors... This is the quantitative representation of the core semantic primitive in the current specific context.

[0111] Next, the system utilizes the context adjustment vector Cultural attribute vector in primitive identity vector Dynamic adjustments are made to generate an activated cultural attribute vector. This process is calculated using the following formula:

[0112]

[0113] in, It is the original cultural attribute vector obtained from the primitive identity vector. This is the context adjustment vector obtained from the aforementioned calculation. w `Context` is a pre-trained weight matrix used to map the dimensions of the context conditioning vector to the same dimensions as the cultural attribute vector. `sigmoid` is a standard activation function with an output range of (0,1), used to generate a conditioning weight. `⊙` represents the Hadamard product, which is the element-wise multiplication of corresponding vectors. This operation uses the output of the `sigmoid` function as a gating signal to either enhance or suppress each dimension of the original cultural attribute vector.

[0114] Meanwhile, the feature collaborative analysis module 30 analyzes the semantic vectors in the primitive identity vectors. Perform semantic drift correction. This correction process first obtains the timestamp t from the primitive identity vector. orig Metadata on the creation date of the text to be analyzed text And calculate the time difference between the two Δt = t text -t orig The system uses a preset adjustment function f. drift Calculate a semantic drift vector

[0115] To further improve the accuracy of the correction, the system introduces the intertextual attraction value F calculated by the intertextuality analysis submodule 23. grav This gravity value is used to adjust the degree of semantic drift correction. A high intertextuality gravity value indicates that the current text fragment is highly correlated with a certain metatextual text, and therefore more of the original semantics of that primitive should be preserved. The final, fully corrected semantic vector... The calculation formula is:

[0116]

[0117] in, It is the original semantic vector. It is a semantic drift vector calculated based on the time difference. F grav It is the intertextuality attraction value of the text segment containing the core semantic primitive, and its value range is [0,1]. When F grav When the value approaches 1, it indicates extremely strong intertextuality and a high drift correction value. Approaching zero, thus preserving the original semantics; when F grav When the value approaches 0, a full drift correction is applied.

[0118] After completing the above calculations, the resulting activated cultural attribute vector is... and fully corrected semantic vectors Together, these constitute the dynamic feature representation of the core semantic primitive in the current specific context. This dynamic feature representation will serve as one of the inputs for subsequent semantic conflict detection steps.

[0119] In parallel with the dynamic activation of the core semantic primitive context, the feature co-analysis module 30 executes a semantic conflict detection mechanism. This mechanism is performed one by one on each text segment (e.g., a single sentence or clause) determined by grammatical structure analysis. Its purpose is to quantify the degree of semantic consistency of feature information obtained from different dimensions of analysis within the same text segment.

[0120] In one specific embodiment, for a given text segment, the module first maps its multidimensional feature information to the same high-dimensional semantic vector space. This process includes the following steps:

[0121] Step 1: The module obtains the feature representations of the core semantic primitives that have been dynamically activated within the text segment, specifically their fully corrected semantic vectors.

[0122] Step 2: The module obtains the gravitational center vector of the source text most strongly associated with the text segment, provided by the intertextuality analysis submodule 23.

[0123] Step 3: The module will calculate the scalar sentiment score E for the text segment, which belongs to the sentiment time series analysis submodule 24. score Convert into an emotion vector

[0124] This transformation is achieved by multiplying the emotional score by a preset unit vector representing the main direction of the emotion. To complete, that is This unit vector The direction in the semantic space is predefined to represent the emotional axis from the most negative to the most positive.

[0125] After completing the above mapping, the system obtains a set of feature vectors V corresponding to the current text segment. seg This set contains one or more vectors generated by the preceding steps, for example...

[0126] Subsequently, the module applies this feature vector set V segThe spatial dispersion of the vector set is quantified to obtain the semantic conflict score of the text segment. First, the centroid vector of the vector set is calculated. The calculation formula is as follows:

[0127]

[0128] in, It is set V seg The k-th eigenvector in V. m is a set. seg The total number of vectors contained therein.

[0129] Then, based on this centroid vector, the semantic conflict score C is calculated. seg This score is defined as the average cosine distance between all vectors in the set and the centroid vector, calculated using the following formula:

[0130]

[0131] Among them, C seg This is the semantic conflict score of the current text segment, with a value range of [0, 2]. A higher score indicates lower directional consistency of features across different dimensions in the semantic space. · represents the vector dot product. ||·|| represents the Euclidean norm of the vector.

[0132] Finally, the calculated semantic conflict score C seg With a preset numerical threshold T conflict Compare. If C seg ≥T conflict If a semantic conflict is detected, the feature co-analysis module 30 marks the location of the conflict in that text segment. All marked semantic conflict points and their location information will be transmitted to the topic report generation module 40 as part of the analysis results.

[0133] After receiving the analysis results output by the feature collaborative analysis module 30, the topic report generation module 40 executes the final report generation process.

[0134] This module first aggregates all data from the preceding module.

[0135] Specifically, it receives a list of semantic conflict markers, where each marker contains its position index in the text to be analyzed and its corresponding semantic conflict score C. seg Simultaneously, it receives dynamic feature representations associated with each identified core semantic primitive in the text, including fully corrected semantic vectors. and activated cultural attribute vector In addition, the module also receives raw four-dimensional feature information generated by the text multidimensional analysis module 20 for cross-referencing and corroboration.

[0136] For each text segment marked as semantically conflicting, the topic report generation module 40 performs a conflict source deconstruction analysis.

[0137] This analysis aims to determine the reasons for the high conflict score C. seg The main feature dimensions. For a conflict point, the module re-examines its corresponding feature vector set V. seg and its centroid vector By calculating each feature vector in the set one by one With the centroid vector By analyzing the cosine distance between the two points, the system can identify the feature vector furthest from the centroid. The analytical dimension represented by this vector (e.g., sentiment timeline, intertextuality, or core semantics) is identified as the primary source of the conflict. The system then attaches this source information, such as a label identifying it as "sentiment-semantic conflict," to the record of that conflict point.

[0138] In parallel, this module performs macro-thematic component synthesis of the text.

[0139] This process is based on the activation cultural attribute vectors of all identified core semantic primitives in the text. The system employs an unsupervised clustering algorithm, such as K-Means clustering, to cluster this set of vectors. The number of clusters K can be preset or dynamically determined using metrics such as silhouette coefficients. After clustering, the centroid vector of each cluster is defined as a topic component vector. The system further associates one or more of the most salient cultural attribute labels with each topic component vector by examining the average weights of each dimension of the cultural attribute vectors in each cluster.

[0140] Finally, the thematic report generation module 40 integrates and structures all the above analysis results to generate a multi-level analysis report.

[0141] The report's data structure includes the following parts:

[0142] The first part is the metadata summary, which contains overall statistical information about the text to be analyzed, such as the total number of core semantic primitives identified, the density of semantic conflict points (i.e., the number of conflict points divided by the total number of sentences), and a list of main topic components obtained through topic component synthesis.

[0143] The second part is a list of conflict details, listing all semantic conflict points one by one in textual order. Each entry includes the conflict location and conflict score C. seg The conflict type labels derived from the conflict source deconstruction analysis, and the references to the feature vectors that lead to the conflict.

[0144] The third part details the topic analysis, listing each synthesized topic component and showing the core semantic primitives that contribute the most to the topic and their positions in the text.

[0145] The generated reports are output in a standardized data format (such as JSON or XML) to facilitate subsequent programmatic processing or visualization in a graphical user interface. For example, the results of sentiment time series analysis can be formatted as time series data for plotting sentiment curves; the relationships between topic components and their associated core semantic primitives can be formatted as network graph data.

[0146] See attached document Figure 3 , Figure 3 This is a flowchart of a text topic analysis method according to an embodiment of the present invention. The text topic analysis method of the present invention mainly includes the following steps in its execution flow:

[0147] Step S100, Preparation stage: Building the core knowledge base.

[0148] Before the online analysis process begins, the system executes an offline knowledge base construction process. This process involves in-depth mining and statistical analysis of a large-scale, high-quality corpus of classical texts to extract core semantic primitives with stable cultural connotations and high reusability. For each extracted core semantic primitive, the system generates a primitive identity vector containing a high-dimensional semantic vector, a multi-dimensional cultural attribute vector, and an initial formation timestamp. All core semantic primitives and their corresponding identity vectors are stored uniformly, forming a core semantic primitive library, providing prior knowledge support for subsequent online analysis.

[0149] Step S200: Receive the text to be analyzed and start parallel multidimensional feature extraction.

[0150] Once an external user or system inputs a text to be analyzed, this method initiates its online analysis process. The text is simultaneously fed into four parallel analysis engines to deconstruct it from different dimensions and generate mutually aligned feature information streams:

[0151] 1. Grammatical Structure Analysis: The dependency parsing model is invoked to parse each sentence in the text, obtaining the part of speech of each word and the syntactic components it undertakes (such as subject, object, etc.).

[0152] 2. Core semantic primitive identification: Using an Aho-Corasick automaton built on a core semantic primitive library, a single scan of the text is performed to quickly and accurately identify all core semantic primitives referenced in the text and their specific locations within the text.

[0153] 3. Intertextuality analysis: The text to be analyzed is segmented into fragments, and the semantic similarity between each fragment and each source text in the Yuandian Corpus is calculated using the intertextuality gravity field model. This determines the source text with the strongest association and the corresponding intertextuality gravity value for each fragment.

[0154] 4. Sentiment Time Series Analysis: Based on a sentiment dictionary that includes sentiment polarity and intensity, and considering the influence of negative words and degree adverbs, the sentiment score of the text is calculated sentence by sentence, generating a sentiment curve data that reflects the emotional fluctuations of the entire text.

[0155] Step S300: Perform feature co-analysis and semantic conflict detection.

[0156] This method aggregates the four heterogeneous feature information streams generated in the previous step and initiates feature collaborative analysis. First, for each core semantic primitive identified in the text, the method performs a context-dynamic activation operation. That is, it uses the local contextual word information of its occurrence position, the global intertextuality value, and the time difference with the text creation date to dynamically adjust and correct its static primitive identity vector, generating a dynamic feature representation that better reflects its specific meaning in the current context.

[0157] Subsequently, this method performs semantic conflict detection on each paragraph or sentence of the text. This process maps different dimensional features within the text segment (such as the dynamic semantics of primitives, the semantics of intertextual source texts, sentiment vectors, etc.) into a high-dimensional semantic space. A semantic conflict score is quantified by calculating the dispersion of these feature vectors in this space (i.e., the mean cosine distance to the centroid of the vector set). If this score exceeds a preset threshold, the text segment is considered to have a potential deep semantic conflict caused by inconsistencies in multi-dimensional information, and is marked accordingly.

[0158] Step S400: Perform the synthesis of thematic components and generate a structured analysis report.

[0159] In the final stage of the method, the system first synthesizes macro-topics based on the context-activated cultural attribute vectors of all core semantic primitives in the text, using unsupervised learning algorithms such as K-means clustering to identify one or more potential topic components in the text. Then, the system performs attribution analysis on each labeled semantic conflict point to determine the main feature dimensions that triggered the conflict.

[0160] Finally, this method integrates all analysis results, including macro-thematic components, a detailed list of semantic conflict points (including location, score, and conflict source), and raw feature data of each dimension, to generate a structured, multi-level comprehensive analysis report, which is output in a standard data format (such as JSON).

[0161] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A Chinese language and literature text theme analysis system based on big data, characterized in that: include: The core semantic primitive library construction module is used to extract core semantic primitives from the meta-canonical corpus and construct primitive identity vectors containing semantic vectors, cultural attribute vectors and timestamps for the core semantic primitives to form the core semantic primitive library; The text multidimensional analysis module is used to perform parallel syntactic structure analysis, core semantic primitive recognition, intertextuality analysis, and sentiment temporal analysis on the text to be analyzed, in order to obtain four corresponding feature information. The feature collaborative analysis module, connected to the text multidimensional analysis module, is used to dynamically activate the primitive identity vectors in the core semantic primitive library based on the context of the text to be analyzed, and to compare the feature information from the four analyses in parallel to detect semantic conflicts between them. The topic report generation module is used to locate the core topic and generate a structured analysis report based on the semantic consistency or semantic conflict detected by the feature collaborative analysis module.

2. The Chinese language and literature text theme analysis system based on big data according to claim 1, characterized in that, The core semantic primitive library construction module is specifically used for: The semantic vector is obtained by training a word embedding model to capture the basic word meaning of the core semantic primitives; The predefined cultural attributes are encoded into numerical values ​​to form the cultural attribute vector, which represents the initial symbolic meaning of the core semantic primitive.

3. The Chinese language and literature text theme analysis system based on big data according to claim 1, characterized in that, The intertextuality analysis performed by the text multidimensional analysis module is achieved by running an intertextuality gravity field model, which is specifically used for: Represent the source text in the Yuandian Corpus as a gravity center vector; Calculate the vector similarity between the text segment vector in the text to be analyzed and the gravity center vector, and use the similarity as the intertextual gravity value to measure the intertextual association between the text to be analyzed and the source text.

4. The Chinese language and literature text theme analysis system based on big data according to claim 1, characterized in that, The process by which the feature collaborative analysis module dynamically activates the primitive identity vector includes: The local context of the core semantic primitives in the text to be analyzed is encoded into a context adjustment vector; The cultural attribute vector in the primitive identity vector is adjusted using the context adjustment vector to generate an activation vector that reflects the cultural attributes in the current context.

5. The Chinese language and literature text theme analysis system based on big data according to claim 4, characterized in that, The process of dynamically activating the primitive identity vector by the feature collaborative analysis module also includes: Based on the timestamp in the primitive identity vector and the creation year of the text to be analyzed, the semantic vector in the primitive identity vector is adjusted to achieve semantic drift correction of the core semantic primitive.

6. The Chinese language and literature text theme analysis system based on big data according to claim 1, characterized in that, The feature co-analysis module, when performing the semantic drift correction, is also used for: The degree of semantic drift correction is adjusted by weighting the intertextuality attraction value obtained from the intertextuality analysis performed by the text multidimensional analysis module.

7. The Chinese language and literature text theme analysis system based on big data according to claim 1, characterized in that, The process of detecting semantic conflicts by the feature co-analysis module includes: The feature information obtained through the four analyses is mapped to a unified semantic expression space to form a set of signal vectors; Calculate the variance of the signal vector set. When the variance is greater than a preset conflict threshold, mark the corresponding text position as a semantic conflict point.

8. The Chinese language and literature text theme analysis system based on big data according to claim 7, characterized in that, The topic report generation module is also used for: When the marker of the semantic conflict point is received, a hypothetical explanation is generated regarding whether the text to be analyzed uses a specific artistic technique of expressing semantic inconsistency at that location.

9. The Chinese language and literature text theme analysis system based on big data according to claim 1, characterized in that, The structured analysis report generated by the topic report generation module adopts an argument-evidence format, presenting the analyzed topic as the argument and the feature information obtained through the four types of analysis as evidence.

10. A method for thematic analysis of Chinese language and literature texts based on big data, implemented by the system described in any one of claims 1-9, characterized in that, Includes the following steps: Constructing a core semantic primitive library, the construction process includes extracting core semantic primitives from the meta-canonical corpus, and constructing primitive identity vectors for the core semantic primitives, which include semantic vectors, cultural attribute vectors, and timestamps; The input text to be analyzed is subjected to multidimensional analysis, which includes parallel syntactic structure analysis, core semantic primitive identification, intertextuality analysis, and sentiment temporal analysis to obtain four corresponding feature information. Feature collaborative analysis is performed, which includes dynamically activating the primitive identity vectors in the core semantic primitive library based on the context of the text to be analyzed, and comparing feature information from the four analyses in parallel to detect semantic conflicts between them. Based on the detected semantic consistency or semantic conflict, the core topic is located and a structured analysis report is generated.

Images