Scientific and technical literature viewpoint sentence recognition method and system based on emotion enhancement

Abstract

The invention provides a scientific and technical literature viewpoint sentence recognition method and system based on emotion enhancement, and relates to the technical field of natural language processing, and the method comprises the steps: obtaining a target literature to be subjected to viewpoint sentence recognition; a viewpoint sentence recognition model is constructed, and the viewpoint sentence recognition model comprises a context feature extraction component, a local text feature extraction component and a weighted emotion feature fusion enhancement component; and inputting the target literature into the viewpoint sentence recognition model, and outputting and obtaining a viewpoint sentence recognition result. The technical problems that in the prior art, a viewpoint sentence recognition model mainly focuses on viewpoint sentence mining in the fields of social media, product comments and the like, depends on dominant sentiment words for analysis, viewpoint sentences which are generally lack of dominant sentiment words in the scientific and technical literatures are generally hard to migrate directly, and adaptability to the scientific and technical literatures is insufficient are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of natural language processing, and particularly to a method and system for identifying opinion sentences in scientific and technological literature based on emotion enhancement. Background Art

[0002] Identifying opinion sentences in scientific and technological literature is a natural language processing (NLP) task aimed at automatically identifying sentences in a text that contain the author's subjective evaluation, reasoning, or judgment. These sentences often reflect the author's subjective views, speculations, or evaluations, rather than simply stating objective facts. Opinion sentences are usually based on the author's personal understanding, research findings, or theoretical assumptions, and express the author's attitude, opinion, or conclusion towards a specific research question.

[0003] In scientific and technological literature, opinion sentences not only have subjectivity but may also contain complex reasoning, academic reviews, or theoretical discussions. They are usually distributed in core parts such as abstracts, introductions, backgrounds, discussions, and conclusions, and involve the evaluation of the advantages and disadvantages of research methods, the interpretation of the significance of experimental results, the comparative analysis with existing research, and the prediction of future research directions. The accurate identification of opinion sentences is of great value to academic literature, as it can not only reveal the author's academic stance and research attitude but also provide crucial support for information extraction, literature summarization, and knowledge discovery.

[0004] The existing opinion sentence identification has the following limitations:

[0005] 1. Limitations in research fields: Most of the current opinion sentence mining techniques mainly analyze subjective information in fields such as product reviews and social media. These methods usually focus on sentences with explicit emotional expressions, such as positive or negative evaluations. However, the opinions expressed in scientific literature are often more complex, involving implicit reasoning, professional terms, and academic arguments, which are significantly different from the opinion expressions in popular texts. Existing technologies are difficult to directly apply to the task of identifying opinion sentences in scientific literature.

[0006] 2. Lack of specialized research on opinion sentences in scientific literature: Opinion sentences in scientific literature usually contain complex reasoning logics and technical languages, and present the author's subjective opinions implicitly. This complexity makes it difficult for existing opinion sentence identification technologies to accurately capture the author's academic attitude and implicit judgments when dealing with scientific literature. In addition, there is limited research on opinion mining methods closely related to the research content and academic background in scientific texts. Summary of the Invention

[0007] This application provides a method and system for identifying opinion sentences in scientific and technological literature based on emotion enhancement, aiming to make up for the deficiencies in the prior art in this field. The current opinion sentence recognition models mainly focus on fields such as social media and product reviews, relying on explicit emotion words for analysis, and it is difficult to effectively transfer to scientific and technological literature involving implicit reasoning, technical terms, and academic arguments, thus having limitations in adaptability. To address this issue, this application provides a more adaptable recognition solution to improve the recognition performance in scientific and technological literature.

[0008] In the first aspect disclosed in this application, a method for identifying opinion sentences in scientific and technological literature based on emotion enhancement is provided. The method includes: obtaining a target document to be identified for opinion sentences; constructing an opinion sentence recognition model, where the opinion sentence recognition model includes a context feature extraction component, a local text feature extraction component, and a weighted emotion feature fusion and enhancement component; inputting the target document into the opinion sentence recognition model and outputting the opinion sentence recognition result.

[0009] In the second aspect disclosed in this application, a system for identifying opinion sentences in scientific and technological literature based on emotion enhancement is provided. The system is used for the above-mentioned method for identifying opinion sentences in scientific and technological literature based on emotion enhancement. The system includes: a target document acquisition module for obtaining a target document to be identified for opinion sentences; an identification model construction module for constructing an opinion sentence recognition model, where the opinion sentence recognition model includes a context feature extraction component, a local text feature extraction component, and a weighted emotion feature fusion and enhancement component; an identification result acquisition module for inputting the target document into the opinion sentence recognition model and outputting the opinion sentence recognition result.

[0010] One or more technical solutions provided in this application have at least the following beneficial effects:

[0011] By extracting the target documents to be analyzed from scientific and technological literature, the efficient management of input data is achieved, ensuring the pertinence of the data input into the model; a viewpoint sentence recognition model is constructed. Among them, the viewpoint sentence recognition model includes a context feature extraction component, a local text feature extraction component, and a weighted sentiment feature fusion and enhancement component, which can achieve multi-level text feature extraction and processing. This design helps to comprehensively capture the semantic information, context relevance, and sentiment features in the target documents, providing multi-dimensional feature support for viewpoint sentence recognition; the context feature extraction component can accurately capture the implicit information in the sentence, improving the model's ability to capture domain terms and context, and enhancing the accuracy of viewpoint sentence recognition; the local text feature extraction component can accurately capture the local semantic features at different scales in the sentence, especially suitable for complex language expressions in viewpoint sentences, and can effectively extract the deep semantic information spanning multiple words in the viewpoint sentence, improving the understanding ability of complex viewpoint sentences; the weighted sentiment feature fusion and enhancement component can accurately identify the implicit sentiment features in the viewpoint sentence, improving the model's ability to recognize viewpoint sentences; after the target documents are input into the model, the context feature extraction, local text feature extraction, and weighted sentiment feature fusion and enhancement components work together to improve the recognition effect of viewpoint sentences.

[0012] The above description is only an overview of the technical solution of this application. In order to be able to understand the technical means of this application more clearly, it can be implemented according to the content of the specification. And in order to make the above and other purposes, features, and advantages of this application more obvious and understandable, the following specific embodiments of this application are specifically given. Brief Description of the Drawings

[0013] Figure 1 It is a schematic flow chart of a method for recognizing viewpoint sentences in scientific and technological literature based on sentiment enhancement provided by an embodiment of this application.

[0014] Figure 2 It is an exemplary structural schematic diagram of a viewpoint sentence recognition model in a method for recognizing viewpoint sentences in scientific and technological literature based on sentiment enhancement provided by an embodiment of this application.

[0015] Figure 3 It is a structural schematic diagram of a system for recognizing viewpoint sentences in scientific and technological literature based on sentiment enhancement provided by an embodiment of this application.

[0016] Description of the reference numerals: Target document acquisition module 10, recognition model construction module 20, recognition result acquisition module 30. Detailed Description of the Embodiments

[0017] The embodiments of the present application provide a method and system for identifying opinion sentences in scientific and technological literature based on emotion enhancement. Aiming at the limitations of the prior art, which is mainly applied to the fields of social media and product reviews and relies on explicit emotion words for analysis, it solves the problem of insufficient adaptability that it is difficult to directly migrate in scientific and technological literature involving implicit reasoning, technical terms, and academic arguments because opinion sentences usually lack explicit emotion words.

[0018] After introducing the basic principle of the present application, various non-limiting implementation manners of the present application will be specifically introduced below in conjunction with the accompanying drawings of the specification. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0019] Embodiment 1, as Figure 1 shown, the embodiments of the present application provide a method for identifying opinion sentences in scientific and technological literature based on emotion enhancement, and the method includes:

[0020] Step S100: Obtain a target document to be identified for opinion sentences;

[0021] Obtain the target document. The target document refers to a scientific and technological document containing opinion sentences and factual sentences to be identified, and may include academic papers, research reports, technical patents, etc. Preferably, the scientific and technological documents in the agricultural field are used as the main application scenario. Further, perform preprocessing operations on the target document, including segmenting, clause splitting, and removing irrelevant information, such as headers, footers, chart descriptions, etc., to improve the accuracy of subsequent identification.

[0022] Step S200: Construct an opinion sentence recognition model, wherein the opinion sentence recognition model includes a context feature extraction component, a local text feature extraction component, and a weighted emotion feature fusion and enhancement component;

[0023] Construct an opinion sentence recognition model for identifying opinion sentences in scientific and technological literature, including a context feature extraction component, a local text feature extraction component, and a weighted emotion feature fusion and enhancement component, as Figure 2As shown, it is an exemplary structural diagram of the opinion sentence recognition model. Among them, the context feature extraction component uses AgriBERT (Agricultural Bidirectional Encoder Representations from Transformers) as the encoder to extract the context features of the input sentence. AgriBERT can effectively capture the domain-specific language features and expression patterns in agricultural science and technology literature, providing high-quality semantic representations for opinion sentence recognition; the local text feature extraction component uses the Deep Pyramid Convolutional Neural Network (DPCNN) to be responsible for hierarchical feature extraction, and uses its hierarchical convolutional structure to capture the multi-scale opinion expression features in the sentence; the weighted sentiment feature fusion and enhancement component uses the weighted sentiment feature fusion and enhancement mechanism to organically combine the sentiment features and text features to enhance the model's ability to capture implicit sentiment and subjective expressions in opinion sentences.

[0024] Step S300: Input the target document into the opinion sentence recognition model and output the opinion sentence recognition result.

[0025] After splitting the text of the preprocessed target document into sentences, input each sentence into the opinion sentence recognition model one by one. Among them, the context feature extraction component generates the context representation vector of the sentence, the local text feature extraction component extracts the hierarchical multi-scale sentence features, the weighted sentiment feature fusion and enhancement component combines the sentiment features and text features, and the fully connected layer generates the classification probability of each sentence through the Softmax activation function. According to the classification probability, determine the category of each sentence. Finally, the model outputs the classification label of each sentence and the corresponding probability distribution to obtain the opinion sentence recognition result.

[0026] Furthermore, building the opinion sentence recognition model includes:

[0027] Step S210: Build an opinion sentence recognition corpus;

[0028] Step S220: Build the context feature extraction component, local text feature extraction component, and weighted sentiment feature fusion and enhancement component in the opinion sentence recognition model;

[0029] Step S230: Based on the opinion sentence recognition corpus, train the opinion sentence recognition model on the training set, screen the optimal model parameters on the validation set, and evaluate the performance of the final opinion sentence recognition model on the test set.

[0030] Obtain a sample document dataset from scientific and technological literature, ensure that the sample documents contain opinion sentences with a variety of writing styles and different expressions, such as directly expressed opinions and implicitly inferred opinions, annotate the opinion sentences and factual sentences in each sample document data, obtain a sample opinion sentence recognition result set, divide the sample document dataset and the sample opinion sentence recognition result set to obtain a training set, a validation set, and a test set, thereby constructing an opinion sentence recognition corpus.

[0031] AgriBERT is used as an encoder to build a context feature extraction component to extract context features of the input sentence, generate domain-specific representation vectors, and enhance the model's ability to understand agricultural terms and context. With its pre-trained context perception capabilities, AgriBERT can accurately capture implicit information in sentences, thereby improving the accuracy of opinion sentence recognition.

[0032] The deep pyramid convolutional neural network (DPCNN) is used to construct a local text feature extraction component. The hierarchical convolutional structure captures semantic features of different scales in sentences, which is particularly suitable for complex language expressions in opinion sentences. DPCNN can effectively extract deep semantic information across multiple words in opinion sentences, enhancing the model's ability to understand complex opinion sentences.

[0033] The weighted sentiment feature fusion enhancement fusion mechanism is adopted to construct a weighted sentiment feature fusion enhancement component, which organically combines sentiment features with text features to enhance the model's ability to capture implicit sentiment and subjective expressions in opinion sentences. This component can capture the implicit sentiment features in opinion sentences through the weighted sentiment feature fusion mechanism and improve the model's ability to distinguish opinion sentences from factual sentences.

[0034] Use the training set in the opinion sentence recognition corpus to train the model, calculate the loss value and update the model parameters based on the prediction results and the true label; use the validation set for mid-term evaluation and adjust the model's hyperparameters, such as the learning rate and hidden layer dimension; evaluate the model performance on the test set and calculate the final indicators, such as the F1 value. When the F1 value reaches the preset requirements, the opinion sentence recognition model is completed.

[0035] Furthermore, we construct a corpus for identifying opinion sentences, including:

[0036] Step S211: Obtain a sample document data set;

[0037] Step S212: manually annotating the sentences in each sample document data set to distinguish opinion sentences from fact sentences, so as to construct an opinion sentence recognition corpus;

[0038] Step S213: Divide the constructed opinion sentence recognition corpus according to experimental requirements to form a training set, a validation set, and a test set to support the training and evaluation of the opinion sentence recognition model.

[0039] Obtain a sample document dataset from scientific and technological literature, including academic papers, research reports, technical patent documents, etc. The dataset should have the distribution characteristics of opinion sentences and factual sentences, that is, it contains opinion sentences with various writing styles and different expressions, such as directly expressed opinions and implicitly inferred opinions. Preferably, select scientific and technological literature in the agricultural field, covering different directions, such as plant protection, soil improvement, crop yield, etc.

[0040] Manually annotate the sentences in each sample document dataset to distinguish between opinion sentences and factual sentences. Among them, an opinion sentence is a sentence that expresses the author's subjective opinion, reasoning, hypothesis, evaluation, or conclusion, such as "The research shows that this method significantly improves the yield."; a factual sentence is a sentence that states an objective fact or describes an experimental method, such as "There are three variable groups and one control group in the experiment."

[0041] Use a dedicated annotation tool for sentence-level manual annotation. Exemplarily, invite doctoral students and researchers with relevant disciplinary backgrounds in the agricultural field to participate in the annotation to ensure professionalism. Assign a "opinion sentence" (such as annotated as 1) or "factual sentence" (such as annotated as 0) label to each sentence to ensure the consistency and reliability of the annotation results. After the annotation is completed, an opinion sentence recognition corpus is obtained.

[0042] According to experimental requirements, divide the opinion sentence recognition corpus. For example, divide it into a training set, a validation set, and a test set according to a ratio of 6:2:2. Among them, the training set is used to train the model and has the largest data volume; the validation set is used for model parameter tuning; the test set is used for the final model performance evaluation. Adopt a random sampling method to ensure that opinion sentences and factual sentences are evenly distributed in each dataset, and use the division result as the opinion sentence recognition corpus.

[0043] Furthermore, construct a context context feature extraction component, a local text feature extraction component, and a weighted sentiment feature fusion and enhancement component in the opinion sentence recognition model, including:

[0044] Step S221: Construct a context context feature extraction component in the opinion sentence recognition model. Use AgriBERT as the Embedding layer to perform vector representation on the input text. For the input sequence S = {s 1 ,..., s i ,..., s L}, its output sequence representation is X = {x 1 ,..., x i ,..., x L}, where L is the length of the input sequence, s i is the i-th token in the input sequence, is the corresponding representation vector, and d is the hidden layer dimension;

[0045] Step S222: Construct a local text feature extraction component within the opinion sentence recognition model;

[0046] Step S223: Construct a weighted sentiment feature fusion and enhancement component within the opinion sentence recognition model;

[0047] Step S224: Connect the context feature extraction component, the local text feature extraction component, and the weighted sentiment feature fusion and enhancement component to obtain an opinion sentence recognition model.

[0048] The Embedding layer is used to encode the input text sequence into a representation vector with context information. In specific domain tasks, domain-specific pre-trained language models can significantly improve the performance of tasks such as text classification and information extraction by pre-training on large-scale domain corpora; AgriBERT is a pre-trained language model fine-tuned on a diverse agricultural corpus, which can capture scientific knowledge and practical application information in the agricultural domain and generate exclusive and domain context-related embedding representations. The input sequence of the constructed context feature extraction component is S = {s 1 ,..., s i ,..., s L}, and under the encoding of AgriBERT, the generated output sequence representation is X = {x 1 ,..., x i ,..., x L}, where L is the length of the input sequence, s i is the i-th token in the input sequence, x i ∈R d is the corresponding representation vector, and d is the hidden layer dimension.

[0049] The local text feature extraction component is constructed based on a deep pyramid convolutional neural network, which is an efficient hierarchical architecture designed for modeling long-range dependencies in text classification tasks. Its core structure is composed of multiple convolutional blocks with shortcut connections stacked to form a pyramid-shaped feature extraction structure, combining performance and computational efficiency.

[0050] The weighted sentiment feature fusion and enhancement component adopts a weighted sentiment feature fusion mechanism, aiming to enhance the ability to recognize opinion sentences by integrating sentiment features, subjectivity features, and context features in the text. As a subjective statement, an opinion sentence usually contains the author's sentiment tendency, subjective judgment, and context. Therefore, combining text features and sentiment features can capture the characteristics of opinion sentences more comprehensively.

[0051] Adopt a weighted fusion mechanism to perform weighted fusion on the outputs of the local text feature extraction component and the weighted sentiment feature fusion enhancement component, so as to combine them into a unified representation. This fusion representation is mapped to a two-dimensional space through a fully connected layer and the classification of the opinion sentence is completed through a Softmax layer. By organically combining sentiment features with text context features, this fusion mechanism enables the model to more accurately capture implicit emotions and subjective expressions in opinion sentences. The comprehensive representation of such multi-dimensional features enhances the robustness and classification ability of the model, especially significantly under the complex emotion expressions implicit in academic opinion sentences.

[0052] Connect the context context feature extraction component, the local text feature extraction component, and the weighted sentiment feature fusion enhancement component to obtain a complete opinion sentence recognition model.

[0053] Furthermore, constructing the local text feature extraction component within the opinion sentence recognition model includes:

[0054] Step S2221: Adopt a deep pyramid convolutional neural network, which is composed of a text region embedding layer and multiple convolutional blocks with shortcut connections stacked, presenting a pyramid-shaped feature extraction structure;

[0055] Step S2222: Construct a text region embedding layer for extracting local context features spanning multiple words from the hidden layer output of AgriBERT. The text region embedding layer uses F num filters and a convolutional kernel of size H×d, where H is the height of the convolutional kernel and d represents the embedding dimension of the hidden layer;

[0056] Step S2223: Construct a convolutional block with a shortcut connection for extracting deep text features layer by layer. Each convolutional block consists of two convolutional layers, and each layer uses F num filters and a convolutional kernel of size H×H;

[0057] Step S2224: Adopt a downsampling mechanism with a fixed feature map to control the computational complexity of the model and retain key features. After each convolutional block, the model performs downsampling through a pooling operation with a stride of 2, halving the size of the internal representation while maintaining the same number of feature maps. Among them, the repetition times T of convolution and downsampling can be calculated by the following formula:

[0058]

[0059] where L is the sequence length.

[0060] In the original deep pyramid convolutional neural network, this layer captures information of larger text chunks by converting the input text into regional embedding representations, rather than just embedding individual words. In this study, this layer is modified to extract local context features from the output of the hidden layer of AgriBERT. Specifically, F num filters and convolutional kernels of size H×d are used, where H is the size of the convolutional kernel and d is the embedding dimension of the hidden layer. This design can effectively capture local features spanning multiple words and provide high-quality inputs for subsequent convolutional operations.

[0061] Each convolutional block consists of two convolutional layers. Each layer uses F num filters and convolutional kernels of size H×d, and stable feature transfer is achieved through shortcut connections. This kind of shortcut connection adopts a pre-activation mechanism, that is, an activation function is applied before the convolutional operation, which helps to accelerate convergence and improve the stability of the training process.

[0062] After each convolutional block, a pooling layer with a stride of 2 is configured. Downsampling is performed through max-pooling operations, halving the size of the internal representation while maintaining the same number of feature maps. This pyramid-shaped downsampling not only reduces the computational cost but also retains important context information.

[0063] The operations of downsampling and convolutional blocks are repeated until the size of the internal representation reaches a feasible minimum. The number of repetitions T can be calculated by the following formula:

[0064]

[0065] where L is the sequence length.

[0066] Denote the final output of the deep pyramid convolutional neural network as where represents the feature corresponding to the i-th filter, and F num represents the total number of filters. Through this structure, the deep pyramid convolutional neural network can effectively capture hierarchical features of text at multiple scales and provide strong feature support for the deep semantic modeling of opinion sentences.

[0067] Furthermore, a weighted sentiment feature fusion enhancement component is constructed within the opinion sentence recognition model, including:

[0068] Step S2231: Extract multi-dimensional sentiment features: including positive sentiment score S pos , negative sentiment score S neg , sentiment polarity score S pol and subjectivity score S sub, to characterize the emotional tendency and subjectivity degree of the text. Among them, the positive emotion score and the negative emotion score are obtained by calculating through the VADER Sentiment tool, and the emotion polarity score and the subjectivity score are obtained by calculating through the TextBlob tool;

[0069] Step S2232: Weighted fusion of text features and emotion features: Enhance the fusion mechanism through weighted emotion feature fusion, and fuse the text feature V extracted by the local text feature extraction component text with the above-mentioned emotion feature S pos , S neg , S pol and S sub for joint representation to generate a unified feature vector as follows:

[0070] V fusion = Concat(λ text V text , λ senti ⊙ S senti );

[0071] Among them, is the learnable weight of the text feature; S senti = (S pos , S neg , S pol , S sub ) is the emotion feature vector, λ senti = (λ pos , λ neg , λ pol , λ sub ) is the learnable weight of the emotion feature vector; ⊙ is the element-wise multiplication operation, and Concat(·) is the vector concatenation operation;

[0072] Step S2233: Opinion sentence recognition based on the fused representation: Map the weighted fused vector V fusion to a two-dimensional space through a fully connected layer and perform normalization processing through the Softmax function, so as to realize opinion sentence recognition.

[0073] Construct the calculation mechanisms of the positive emotion score S pos , the negative emotion score S neg , the emotion polarity score S pol and the subjectivity score S sub to capture the emotional and subjective information in different dimensions of the text. Among them, the positive emotion score S pos , with a value range of [0, 1], represents the intensity of positive emotion; the negative emotion score S neg , with a value range of [0, 1], represents the intensity of negative emotion; the emotion polarity score S pol, with a value range of [-1, 1], represents the overall polarity of sentiment, from negative to positive; the subjectivity score S sub , with a value range of [0, 1], represents the degree of text from objective to subjective. Among them, the positive and negative sentiment scores are calculated by the VADER Sentiment tool, and the sentiment polarity score and subjectivity score are calculated by the TextBlob tool. These sentiment features can characterize the sentiment and subjectivity information of sentences from multiple dimensions.

[0074] Enhance the fusion mechanism through weighted sentiment feature fusion, combine the text features extracted by the local text feature extraction component with the output of the weighted sentiment feature fusion enhancement component to generate a unified feature vector. The fusion process is represented by the following formula:

[0075] V fusion = Concat(λ text V text , λ senti ⊙S senti );

[0076] Adopt the element-wise multiplication operation ⊙ to ensure that each sentiment feature is appropriately adjusted according to its weight. Finally, combine the weighted text features and sentiment features into a unified representation V through the Concat function fusion , and this fusion representation is mapped to a two-dimensional space through a fully connected layer. This fusion mechanism enables the model to more accurately capture the implicit sentiment and subjective expressions in opinion sentences by organically combining sentiment features with text context features.

[0077] Map the weighted fusion vector V fusion to a two-dimensional space through a fully connected layer and perform normalization processing through the Softmax function to achieve opinion sentence recognition. This comprehensive representation of multi-dimensional features enhances the robustness and classification ability of the model, especially significantly under the complex sentiment expressions implicit in academic opinion sentences.

[0078] In the comparative experiment of this study, we selected the model with the highest F1 value on the validation set for test set evaluation. The experimental results show that the AgriBERT-SentiDPCNN model performs best in the opinion sentence recognition task, and its F1 value reaches 90.19%. In contrast, the F1 value of the baseline model AgriBERT-DPCNN is 89.50%, which further verifies the effectiveness of improving the model performance through the weighted sentiment feature fusion mechanism.

[0079] Compared with traditional neural network models, AgriBERT-SentiDPCNN has significant advantages. For example, the F1 values of FastText and TextCNN are 64.31% and 67.58% respectively, indicating that models relying solely on standard embeddings and lacking large-scale pre-training support are significantly limited in complex tasks. Among pre-trained language models, AgriBERT performs best with an F1 value of 88.72%, outperforming BERT (87.97%) and DeBERTa (87.90%). This validates the effectiveness of domain-specific language models (such as AgriBERT) in domain-specific tasks, which can provide feature representations more suitable for domain tasks.

[0080] The hybrid neural network model further improves performance based on traditional models. The F1 value of the baseline model AgriBERT-DPCNN is 89.50%, which performs best among the hybrid models, demonstrating the robustness of the hierarchical feature extraction ability of DPCNN for opinion recognition tasks. In addition, the F1 values of AgriBERT-CNN and AgriBERT-RCNN are 89.23% and 88.99% respectively, also showing strong competitiveness.

[0081] Compared with Large Language Models (LLMs), the AgriBERT-SentiDPCNN model also performs better. For example, the F1 values of Llama3-70b and GPT-4 are 75.21% and 75.30% respectively. Although LLMs show high generality in multi-tasks, their performance in opinion sentence recognition tasks is lower than that of domain-specific models specifically optimized because they are only based on prompts and not fine-tuned.

[0082] In summary, the experimental results show that AgriBERT-SentiDPCNN performs best in the opinion sentence recognition task, significantly outperforming not only traditional neural network models and pre-trained language models but also unfine-tuned large language models, demonstrating the strong performance advantages of domain-specific models in complex tasks.

[0083] In summary, the technical effects of a method for recognizing opinion sentences in scientific and technological literature based on emotion enhancement provided by the embodiments of the present application are as follows:

[0084] By extracting the target documents to be analyzed from scientific and technological literature, the efficient management of input data is achieved, ensuring the pertinence of the data input into the model; a viewpoint sentence recognition model is constructed. Among them, the viewpoint sentence recognition model includes a context feature extraction component, a local text feature extraction component, and a weighted sentiment feature fusion and enhancement component, which can realize multi-level text information extraction and processing. This design helps to comprehensively capture the semantic information, context relevance, and sentiment features in the target documents, providing multi-dimensional feature support for viewpoint sentence recognition; the context feature extraction component can accurately capture the implicit information in the sentence, improving the model's ability to capture domain terms and context, and enhancing the accuracy of viewpoint sentence recognition; the local text feature extraction component can accurately capture the semantic features at different scales in the sentence, especially suitable for the complex language expressions in viewpoint sentences, and can effectively extract the deep semantic information spanning multiple words in the viewpoint sentence, improving the understanding ability of complex viewpoint sentences; the weighted sentiment feature fusion and enhancement component can accurately identify the implicit sentiment features in the viewpoint sentence, improving the model's ability to recognize viewpoint sentences; after the target documents are input into the model, the context feature extraction, local text feature extraction component, and weighted sentiment feature fusion and enhancement component work together to achieve the efficient classification of each sentence, improving the recognition effect of viewpoint sentences.

[0085] Embodiment 2, based on the same inventive concept as the method for recognizing viewpoint sentences in scientific and technological literature with emotion enhancement in the foregoing embodiment, as Figure 3 shown, an embodiment of the present application provides a system for recognizing viewpoint sentences in scientific and technological literature with emotion enhancement, and the system includes:

[0086] A target document acquisition module 10, configured to acquire a target document to be recognized for viewpoint sentences;

[0087] A recognition model construction module 20, configured to construct a viewpoint sentence recognition model, where the viewpoint sentence recognition model includes a context feature extraction component, a local text feature extraction component, and a weighted sentiment feature fusion and enhancement component;

[0088] A recognition result acquisition module 30, configured to input the target document into the viewpoint sentence recognition model and output a viewpoint sentence recognition result.

[0089] Furthermore, the recognition model construction module 20 includes:

[0090] A corpus construction unit, configured to construct a corpus for recognizing viewpoint sentences;

[0091] A module construction unit, configured to construct a context feature extraction component, a local text feature extraction component, and a weighted sentiment feature fusion and enhancement component in the viewpoint sentence recognition model;

[0092] A model training unit for training the opinion sentence recognition model on a training set based on the opinion sentence recognition corpus, screening the optimal model parameters on a validation set, and evaluating the performance of the final opinion sentence recognition model on a test set.

[0093] Furthermore, the corpus construction unit includes:

[0094] A sample literature data acquisition channel for acquiring a sample literature data set;

[0095] An annotation channel for manually annotating the sentences in each sample literature data set to distinguish opinion sentences and factual sentences, so as to construct an opinion sentence recognition corpus;

[0096] A partitioning channel for partitioning the constructed opinion sentence recognition corpus according to experimental requirements to form a training set, a validation set, and a test set to support the training and evaluation of the opinion sentence recognition model.

[0097] Furthermore, the module construction unit includes:

[0098] A context context feature extraction component construction channel for constructing a context context feature extraction component in the opinion sentence recognition model, using AgriBERT as the Embedding layer for vector representation of the input text. For the input sequence S = {s 1 ,..., s i ,..., s L}, its output sequence representation is X = {x 1 ,..., x i ,..., x L}, where L is the length of the input sequence, s i is the i-th token in the input sequence, is the corresponding representation vector, and d is the hidden layer dimension;

[0099] A local text feature extraction component construction channel for constructing a local text feature extraction component in the opinion sentence recognition model;

[0100] A weighted sentiment feature fusion enhancement component construction channel for constructing a weighted sentiment feature fusion enhancement component in the opinion sentence recognition model;

[0101] An opinion sentence recognition model acquisition channel for connecting the context context feature extraction component, the local text feature extraction component, and the weighted sentiment feature fusion enhancement component to obtain an opinion sentence recognition model.

[0102] Furthermore, the local text feature extraction component construction channel includes:

[0103] A deep convolutional neural network construction node, which is used to adopt a deep pyramid convolutional neural network. The deep pyramid convolutional neural network is stacked by a text region embedding layer and multiple convolutional blocks with shortcut connections, presenting a pyramid-shaped feature extraction structure;

[0104] A text region embedding layer construction node, which is used to construct a text region embedding layer and extract local context features spanning multiple words from the output of the hidden layer of AgriBERT. The text region embedding layer uses F num filters and a convolutional kernel of size H×d, where H is the height of the convolutional kernel and d represents the embedding dimension of the hidden layer;

[0105] A convolutional block construction node, which is used to construct a convolutional block with a shortcut connection and extract deep text features layer by layer. Each convolutional block consists of two convolutional layers, and each layer uses F num filters and a convolutional kernel of size H×H, and feature transfer is achieved through a shortcut connection;

[0106] A node for adopting a downsampling mechanism, which is used to adopt a downsampling mechanism for fixed feature maps to control the computational complexity of the model and retain key features. After each convolutional block, the model performs downsampling through a pooling operation with a stride of 2, halving the size of the internal representation while maintaining the same number of feature maps. Among them, the repetition times T of convolution and downsampling can be calculated by the following formula:

[0107]

[0108] where L is the sequence length.

[0109] Furthermore, the channel for constructing the weighted sentiment feature fusion enhancement component includes:

[0110] A multi-dimensional sentiment feature extraction node, which is used to extract multi-dimensional sentiment features: including a positive sentiment score S pos , a negative sentiment score S neg , a sentiment polarity score S pol and a subjectivity score S sub to characterize the sentiment tendency and subjectivity degree of the text. Among them, the positive sentiment score and the negative sentiment score are calculated through the VADER Sentiment tool, and the sentiment polarity score and the subjectivity score are calculated through the TextBlob tool;

[0111] A weighted fusion node, which is used to weight and fuse text features and sentiment features: through a weighted sentiment feature fusion enhancement mechanism, the text features V text extracted by the local text feature extraction component are fused with the above sentiment features S pos, S neg , S pol and S sub are jointly represented to generate a unified feature vector as follows:

[0112] V fusion = Concat(λ text V text , λ senti ⊙ S senti );

[0113] where is the learnable weight of the text feature; S senti = (S pos , S neg , S pol , S sub ) is the sentiment feature vector, λ senti = (λ pos , λ neg , λ pol , λ sub ) is the learnable weight of the sentiment feature vector; ⊙ is the element-wise multiplication operation, and Concat(·) is the vector concatenation operation;

[0114] The opinion sentence recognition node is used for opinion sentence recognition based on the fused representation: the weighted fused vector V fusion is mapped to a two-dimensional space through a fully connected layer and normalized by the Softmax function, so as to realize opinion sentence recognition.

[0115] Through the foregoing detailed description of a method for recognizing opinion sentences in scientific and technological literature based on sentiment enhancement in this specification, those skilled in the art can clearly know a system for recognizing opinion sentences in scientific and technological literature based on sentiment enhancement in this embodiment. Since it corresponds to the method disclosed in the embodiment, it is described relatively simply. For related parts, reference can be made to the description in the method part.

[0116] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be obvious to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to these embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for identifying opinion sentences in scientific and technological literature based on sentiment enhancement, characterized in that: The method comprises: Obtaining the target document for opinion sentence recognition; Constructing an opinion sentence recognition model, wherein the opinion sentence recognition model includes a context feature extraction component, a local text feature extraction component, and a weighted sentiment feature fusion enhancement component; The target document is input into the opinion sentence recognition model, and the opinion sentence recognition result is obtained by output.

2. The method for identifying opinion sentences in scientific and technological literature based on emotion enhancement according to claim 1 is characterized in that: Construct an opinion sentence recognition model, including: Construct a corpus for opinion sentence identification; Constructing a context feature extraction component, a local text feature extraction component, and a weighted sentiment feature fusion enhancement component in the opinion sentence recognition model; Based on the opinion sentence recognition corpus, the opinion sentence recognition model is trained on a training set, model parameters with optimal performance are screened on a validation set, and performance evaluation of the final opinion sentence recognition model is performed on a test set.

3. The method for identifying opinion sentences in scientific and technological literature based on emotion enhancement according to claim 2 is characterized in that: Construct a corpus for identifying opinion sentences, including: Get the sample document dataset; Manually annotate the sentences in each sample document dataset to distinguish opinion sentences from factual sentences, so as to build an opinion sentence recognition corpus; According to experimental requirements, the constructed opinion sentence recognition corpus is divided into training set, validation set and test set to support the training and evaluation of the opinion sentence recognition model.

4. The method for identifying opinion sentences in scientific and technological literature based on emotion enhancement according to claim 2 is characterized in that: Constructing the context feature extraction component, the local text feature extraction component and the weighted sentiment feature fusion enhancement component in the opinion sentence recognition model includes: The context feature extraction component in the opinion sentence recognition model is constructed, and AgriBERT is used as the Embedding layer to vectorize the input text. For the input sequence S = {s1,...,s i ,...,s L }, and its output sequence is represented as X = {x1,...,x i ,...,x L }, where L is the length of the input sequence, s i is the i-th token in the input sequence, x i ∈R d is the corresponding representation vector, d is the hidden layer dimension; Constructing a local text feature extraction component within the opinion sentence recognition model; Constructing a weighted sentiment feature fusion enhancement component within the opinion sentence recognition model; The context feature extraction component, the local text feature extraction component and the weighted sentiment feature fusion enhancement component are connected to obtain an opinion sentence recognition model.

5. The method for identifying opinion sentences in scientific and technological literature based on emotion enhancement according to claim 4 is characterized in that: Constructing a local text feature extraction component in the opinion sentence recognition model includes: A deep pyramid convolutional neural network is used. The deep pyramid convolutional neural network is composed of a text region embedding layer and a plurality of convolutional blocks with shortcut connections stacked together to present a pyramid-shaped feature extraction structure. Construct a text region embedding layer to extract local context features spanning multiple words from the hidden layer output of AgriBERT. The text region embedding layer adopts F num A filter and a convolution kernel of size H×d, where H is the height of the convolution kernel and d is the embedding dimension of the hidden layer; Construct a convolution block with shortcut connection to extract deep text features layer by layer. Each convolution block consists of two convolution layers, and each layer uses F num A filter and a convolution kernel of size H×H are used, and feature transfer is achieved through shortcut connection; A fixed feature map downsampling mechanism is used to control the computational complexity of the model and maintain key features. After each convolution block, the model performs downsampling through a pooling operation with a stride of 2 to halve the size of the internal representation while maintaining the same number of feature maps. The number of repetitions T of convolution and downsampling can be calculated by the following formula: Where L is the sequence length.

6. The method for identifying opinion sentences in scientific and technological literature based on emotion enhancement according to claim 5 is characterized in that: Constructing a weighted sentiment feature fusion enhancement component in the opinion sentence recognition model includes: Extract multi-dimensional sentiment features: including positive sentiment score S pos , negative sentiment score S neg , sentiment polarity score S pol and the subjective score S sub , to characterize the emotional tendency and subjectivity of the text. The positive emotional score and negative emotional score are calculated by the VADER Sentiment tool, and the emotional polarity score and subjectivity score are calculated by the TextBlob tool; Weighted fusion of text features and sentiment features: The text features V extracted by the local text feature extraction component are fused into a weighted sentiment feature fusion enhancement mechanism. text With the above emotional characteristics S pos , S neg , S pol and S sub Perform joint representation to generate a unified feature vector As follows: V fusion =Concat(λ text V text ,λ senti ⊙S senti ); Among them, λ text is the learnable weight of text features; S senti =(S pos ,S neg ,S pol ,S sub ) is the sentiment feature vector, λ senti =(λ pos ,λ neg ,λ pol ,λ sub ) is the learnable weight of the sentiment feature vector; ⊙ is the element-by-element multiplication operation, and Concat(·) is the vector concatenation operation; Opinion sentence recognition based on fusion representation: The weighted fusion vector V fusion It is mapped to a two-dimensional space through a fully connected layer and normalized through a Softmax function to achieve opinion sentence recognition.

7. A system for identifying opinion sentences in scientific and technological literature based on sentiment enhancement, characterized in that: A method for identifying opinion sentences in scientific and technological literature based on emotion enhancement according to any one of claims 1 to 6, the system comprising: A target document acquisition module is used to acquire the target document to be used for opinion sentence recognition; A recognition model building module, used to build an opinion sentence recognition model, wherein the opinion sentence recognition model includes a context feature extraction component, a local text feature extraction component and a weighted sentiment feature fusion enhancement component; The recognition result acquisition module is used to input the target document into the opinion sentence recognition model and output the opinion sentence recognition result.

8. The system for identifying opinion sentences in scientific and technological literature based on emotion enhancement according to claim 7 is characterized in that: The recognition model building module comprises: Corpus construction unit, used to construct opinion sentence identification corpus; A component construction unit, used to construct a context feature extraction component, a local text feature extraction component, and a weighted sentiment feature fusion enhancement component in the opinion sentence recognition model; A model training unit is used to train the opinion sentence recognition model on a training set based on the opinion sentence recognition corpus, select the model parameters with the best performance on a validation set, and perform performance evaluation on the final opinion sentence recognition model on a test set.

9. The system for identifying opinion sentences in scientific and technological literature based on emotion enhancement according to claim 8 is characterized in that: The corpus construction unit comprises: Sample document data acquisition channel, used to obtain sample document data set; The annotation channel is used to manually annotate the sentences in each sample document dataset, distinguish opinion sentences from factual sentences, and build an opinion sentence recognition corpus; The partitioning channel is used to partition the constructed opinion sentence recognition corpus into training sets, validation sets, and test sets according to experimental requirements to support the training and evaluation of the opinion sentence recognition model.

Images