CN108595407B - Evaluation method and device based on discourse structure of discussion treatise - Google Patents
Evaluation method and device based on discourse structure of discussion treatise Download PDFInfo
- Publication number
- CN108595407B CN108595407B CN201810182942.4A CN201810182942A CN108595407B CN 108595407 B CN108595407 B CN 108595407B CN 201810182942 A CN201810182942 A CN 201810182942A CN 108595407 B CN108595407 B CN 108595407B
- Authority
- CN
- China
- Prior art keywords
- preset
- article
- hierarchical
- discourse
- discussion
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Expired - Fee Related
Links
- 238000011156 evaluation Methods 0.000 title claims abstract description 59
- 238000000034 method Methods 0.000 claims abstract description 38
- 239000011159 matrix material Substances 0.000 claims description 29
- 239000013598 vector Substances 0.000 claims description 22
- 238000010276 construction Methods 0.000 claims description 15
- 238000013527 convolutional neural network Methods 0.000 claims description 13
- 238000013528 artificial neural network Methods 0.000 claims description 8
- 241000288105 Grus Species 0.000 claims description 3
- 238000012706 support-vector machine Methods 0.000 description 10
- 238000003062 neural network model Methods 0.000 description 8
- 238000011160 research Methods 0.000 description 5
- 238000010586 diagram Methods 0.000 description 4
- 230000001755 vocal effect Effects 0.000 description 4
- 230000000694 effects Effects 0.000 description 3
- 230000014509 gene expression Effects 0.000 description 3
- 238000010801 machine learning Methods 0.000 description 3
- 238000013145 classification model Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000013210 evaluation model Methods 0.000 description 2
- 238000007477 logistic regression Methods 0.000 description 2
- 239000000463 material Substances 0.000 description 2
- 238000012552 review Methods 0.000 description 2
- 230000004075 alteration Effects 0.000 description 1
- 238000010420 art technique Methods 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 238000012417 linear regression Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000003058 natural language processing Methods 0.000 description 1
- 230000006403 short-term memory Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000002123 temporal effect Effects 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/20—Natural language analysis
- G06F40/253—Grammatical analysis; Style critique
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Machine Translation (AREA)
Abstract
The invention discloses an evaluation method and device based on a discourse structure of a discussion paper, wherein the method comprises the following steps: acquiring a plurality of language elements of the discussion paper to be evaluated according to the paragraph type and the sentence type of the discussion paper to be evaluated; constructing sequential characteristics, planar characteristics or hierarchical characteristics of an article discourse structure through a plurality of language discourse elements; and obtaining an evaluation result of the discussion paper to be evaluated through a preset sequential feature model, a preset planar feature model or a preset hierarchical feature model according to the sequential feature, the preset planar feature or the preset hierarchical feature of the chapter structure of the article. The method can obtain the evaluation result of the discussion paper to be evaluated through a preset sequence type characteristic model, a preset plane type characteristic model or a preset hierarchy type characteristic model according to the sequence type characteristic, the preset plane type characteristic or the preset hierarchy type characteristic of the chapter structure of the article, and the evaluation accuracy is effectively improved.
Description
Technical Field
The invention relates to the technical field of natural language processing, in particular to an evaluation method and device based on a discourse structure of a discussion paper.
Background
A discussion paper is a common subject matter form, which is a discourse for analyzing things, discussing affairs, issuing opinions and proposing advices. Discussion is generally to analyze and review a problem or an event to show its own opinions, positions, attitudes, opinions and advices. The three elements of the discussion are argument, argumentation and demonstration. Generally speaking, the structure of the discussion is usually fixed, such as general fraction, contrast, progression, and parallel. Therefore, the recognition of the structure of the discussion paper is helpful to improve the understanding of the discussion paper and the evaluation of the quality of the articles, and the system for scoring the articles is promoted.
The related technology is based on the research on article coherence, grammar quality and topic relevance, and the research on the chapter structure of the article is relatively less. In the early twentieth century, there have been studies on methods for automatic evaluation. Such as IEA (Intelligent paper evaluation), e-Rater, etc. In addition, there are some non-commercial automatic rating systems, such as those based on semantic relevance, and the impact of sentence temporal relevance on article scoring.
The classifiers mostly used by the early automatic article evaluation models are logistic regression, and the simple and efficient models can conveniently map the article features to the corresponding scores. In recent years, models for automatically evaluating articles begin to use more complex model results to acquire article feature information more, and generally obtain better article scoring results. The related technology uses language, semantic and retrieval characteristics to construct a hierarchical classification model for article evaluation. Related art techniques have been to build improved logistic regression, linear regression models using features based on opinion expressions and topic elements. The related art also researches a classification model of a support vector machine based on the chapter structure of the article.
Disclosure of Invention
The present invention is directed to solving, at least to some extent, one of the technical problems in the related art.
Therefore, an object of the present invention is to provide an evaluation method based on the structure of the discourse of the discussion paper, which can effectively improve the accuracy of evaluation.
Another objective of the present invention is to provide an evaluation device based on the structure of the discourse of the discussion.
In order to achieve the above object, an embodiment of the present invention provides an evaluation method based on a discourse structure of a discussion paper, including the following steps: acquiring a plurality of sentence elements of the discussion paper to be evaluated according to the paragraph type and the sentence type of the discussion paper to be evaluated; constructing a sequential feature, a planar feature or a hierarchical feature of an article discourse structure through the plurality of language discourse elements; and obtaining the evaluation result of the discussion paper to be evaluated according to the sequential characteristic, the planar characteristic or the hierarchical characteristic of the article discourse structure through a preset sequential characteristic model, a preset planar characteristic model or a preset hierarchical characteristic model.
The evaluation method based on the discourse structure of the embodiment of the invention constructs an effective discourse structure representation method for the discourse, respectively adopts a sequence type, a plane type and a hierarchical type representation method for discourse characteristics of the discourse, and then constructs a neural network model for the discourse structure characteristic representation methods of different types to automatically evaluate the discourse structure, thereby effectively improving the evaluation accuracy.
In addition, the evaluation method based on the structure of the discourse section according to the above embodiment of the present invention may further have the following additional technical features:
further, in an embodiment of the present invention, the method further includes: filling chapter elements of the sentences into corresponding parts of the discussion papers, and filling 0 into corresponding vacant parts so as to represent each article as a vector of a preset dimension and obtain sequence type characteristics of the article; and constructing a classifier modeling according to the sequence type characteristics of the article, wherein an SVM (Support Vector Machine) is adopted for model classification so as to obtain the preset sequence type characteristic model.
Further, in an embodiment of the present invention, the method further includes: forming a vector by the sentences preset in each paragraph, forming a matrix by the preset paragraphs, and filling 0 in the corresponding vacant part to express each article as the preset matrix to obtain the planar characteristic of the article; and modeling according to the planar characteristic of the article and the article discourse structure score, wherein a convolutional neural network is adopted for modeling so as to obtain the preset planar characteristic model.
Further, in an embodiment of the present invention, the method further includes: forming a vector by the sentences preset in each paragraph, forming a matrix by the preset paragraphs, and filling 0 in the corresponding vacant part to express each article as the preset matrix to obtain the hierarchical characteristics of the article; and modeling according to the hierarchical features of the article and the chapter structure scores of the article, wherein a hierarchical neural network is adopted for modeling, paragraph-level information of the article is acquired through a GRU (Gated Current Unit) neural network, and a front-back connection relation between sentences of the article is acquired through an LSTM (Long Short-Term Memory network) network layer to obtain the preset hierarchical feature model.
Further, in one embodiment of the invention, the verbal element includes one of an introduction, a point of discourse, an example, a point of view, an attention, and a conclusion.
In order to achieve the above object, another embodiment of the present invention provides an evaluation apparatus based on a structure of a discourse, including: the first acquisition module is used for acquiring a plurality of language elements of the discussion paper to be evaluated through the paragraph type and the sentence type of the discussion paper to be evaluated; the first construction module is used for constructing sequential characteristics, planar characteristics or hierarchical characteristics of an article chapter structure through the plurality of language chapter elements; and the evaluation module is used for obtaining the evaluation result of the discussion paper to be evaluated through a preset sequential feature model, a preset planar feature model or a preset hierarchical feature model according to the sequential feature, the preset planar feature or the preset hierarchical feature of the article chapter structure.
The evaluation device based on the discourse structure of the embodiment of the invention constructs an effective discourse structure representation method for the discourse, adopts a sequential, planar and hierarchical representation method for discourse characteristics of the discourse, and then constructs a neural network model for the discourse structure characteristic representation methods of different types to automatically evaluate the discourse structure, thereby effectively improving the evaluation accuracy.
In addition, the evaluation device based on the structure of the discussion-paper chapters according to the above embodiment of the present invention may also have the following additional technical features:
further, in an embodiment of the present invention, the method further includes: the second acquisition module is used for filling chapter elements of the sentences into corresponding parts of the discussion papers, and filling 0 into corresponding vacant parts so as to represent each article as a vector with preset dimensionality and obtain sequence type features of the articles; and the second construction module is used for constructing a classifier for modeling according to the sequence type characteristics of the article, wherein an SVM is adopted for model classification so as to obtain the preset sequence type characteristic model.
Further, in an embodiment of the present invention, the method further includes: the third acquisition module is used for forming a vector by the preset sentences in each paragraph, forming a matrix by the preset paragraphs, and filling 0 in the corresponding vacant part so as to represent each article as the preset matrix and obtain the plane-type characteristics of the article; and the third construction module is used for carrying out modeling according to the planar characteristic of the article and the chapter structure score of the article, wherein a convolutional neural network is adopted for carrying out modeling so as to obtain the preset planar characteristic model.
Further, in an embodiment of the present invention, the method further includes: the fourth acquisition module is used for forming a vector by the preset sentences in each paragraph, forming a matrix by the preset paragraphs, and filling 0 in the corresponding vacant part so as to represent each article as the preset matrix and obtain the hierarchical characteristics of the article; and the fourth construction module is used for carrying out modeling according to the hierarchical characteristics of the article and the chapter structure scores of the article, wherein a hierarchical neural network is adopted for carrying out modeling, paragraph-level information of the article is obtained through a GRU (general-purpose Unit), and a front-back connection relation between sentences of the article is obtained through an LSTM (least Square TM) network layer, so that the preset hierarchical characteristic model is obtained.
Further, in one embodiment of the invention, the verbal element includes one of an introduction, a point of discourse, an example, a point of view, an attention, and a conclusion.
Additional aspects and advantages of the invention will be set forth in part in the description which follows and, in part, will be obvious from the description, or may be learned by practice of the invention.
Drawings
The foregoing and/or additional aspects and advantages of the present invention will become apparent and readily appreciated from the following description of the embodiments, taken in conjunction with the accompanying drawings of which:
FIG. 1 is a flowchart of an evaluation method based on the structure of a discussion discourse according to an embodiment of the present invention;
FIG. 2 is a flowchart of an evaluation method based on the structure of a discussion discourse according to an embodiment of the present invention;
FIG. 3 is a schematic diagram of a planar signature convolutional neural network model according to an embodiment of the present invention;
FIG. 4 is a schematic diagram of a planar signature convolutional neural network model according to an embodiment of the present invention;
fig. 5 is a schematic structural diagram of an evaluation based on the structure of the discussion-paper chapters according to an embodiment of the present invention.
Detailed Description
Reference will now be made in detail to embodiments of the present invention, examples of which are illustrated in the accompanying drawings, wherein like or similar reference numerals refer to the same or similar elements or elements having the same or similar function throughout. The embodiments described below with reference to the drawings are illustrative and intended to be illustrative of the invention and are not to be construed as limiting the invention.
The following describes an evaluation method and apparatus based on a structure of a discourse according to an embodiment of the present invention with reference to the accompanying drawings, and first, an evaluation method based on a structure of a discourse according to an embodiment of the present invention will be described with reference to the accompanying drawings.
Fig. 1 is a flowchart of an evaluation method based on a structure of a discussion paper chapter according to an embodiment of the present invention.
As shown in fig. 1, the evaluation method based on the structure of the discussion chapter includes the following steps:
in step S101, a plurality of sentence elements of the discussion to be evaluated are obtained according to the paragraph type and sentence type of the discussion to be evaluated.
In step S102, a sequential feature, a planar feature or a hierarchical feature of the article discourse structure is constructed from a plurality of language parts.
In step S103, an evaluation result of the discussion paper to be evaluated is obtained through a preset sequential feature model, a preset planar feature model, or a preset hierarchical feature model according to the sequential feature, the preset planar feature, or the preset hierarchical feature of the chapter structure of the article.
It can be understood that, because the research on the automatic evaluation of the article chapter structure in the related art is relatively less, the embodiment of the present invention can perform research and discussion on the discussion paper chapter structure, construct feature vectors for the article chapter structure, construct various machine learning models for different types of feature vectors, construct automatic evaluation models, and respectively express the discussion paper chapter structure as a sequence-type, a plane-type, and a hierarchical-type structure, thereby performing the automatic evaluation of the article using an SVM, a convolutional neural network, and a hierarchical neural network model.
That is, in the automatic evaluation task of the discourse structure in the discussion paper, the embodiment of the present invention can regard it as a classification task, and construct the discourse structure characteristics for the discussion paper. Firstly, using an existing paragraph and sentence type recognition model to obtain one of the sentence and sentence elements of the paragraph in the article, namely introduction, talking points, examples, viewpoints, themes and conclusions; secondly, constructing sequential, planar and hierarchical characteristic representation of the sentence chapter structure of the article by using chapter elements of sentences in the article; and finally, constructing a machine learning model for different feature representations to model.
Further, in an embodiment of the present invention, the method of an embodiment of the present invention further includes: filling chapter elements of the sentences into corresponding parts of the discussion papers, and filling 0 into corresponding vacant parts so as to represent each article as a vector of a preset dimension and obtain sequence type characteristics of the article; and constructing a classifier for modeling according to the sequence type characteristics of the article, wherein an SVM is adopted for model classification so as to obtain a preset sequence type characteristic model.
Specifically, after obtaining chapter elements of paragraphs and sentences of the article, the embodiment of the present invention can express the construction mode of the chapter structure of the discussion paper as the ordinal characteristic. Assuming that a discussion paper is represented by 10 paragraphs, each paragraph contains 10 sentences, since the chapter elements of each sentence in the article are obtained before, the embodiment of the present invention only needs to fill the corresponding element parts of the sentences into the corresponding parts of the discussion paper, and fill the corresponding vacant parts with 0 element, so that each article can be represented as a 100-dimensional vector, which is the sequential feature representation of the article.
After obtaining the sequence-wise feature representation of the article, embodiments of the invention can build a classifier for modeling. The article data set contains scores corresponding to each article and manually calibrated based on the quality of the chapter structure. Therefore, the chapter structure characteristics and the chapter structure scores of each article are in one-to-one correspondence, and only a classifier needs to be constructed for modeling. It should be noted that, because the SVM is more prominent in the previous machine learning tasks, the SVM is used for model classification.
Further, in an embodiment of the present invention, the method of an embodiment of the present invention further includes: forming a vector by the sentences preset in each paragraph, forming a matrix by the preset paragraphs, and filling 0 in the corresponding vacant part to express each article as the preset matrix to obtain the planar characteristic of the article; and modeling according to the planar characteristic of the article and the article chapter structure fraction, wherein a convolutional neural network is adopted for modeling to obtain a preset planar characteristic model.
It is understood that, since the discussion paper is a structure in a planar form, the embodiment of the present invention can represent it as a planar feature, and the specific representation form is as follows: assume that an article is represented by 10 paragraphs, each of which is represented by 10 sentences. Unlike the expression method of the sequence-based features, the planar expression method is mainly characterized in that the article features are expressed in a planar manner, namely, in a matrix. And forming a vector by 10 sentences in each paragraph, and forming a matrix by 10 paragraphs, namely representing the article as a 10-by-10 matrix, wherein similarly, the vacant part is filled with 0 element, so that the represented chapter structural characteristics can better reflect the relationship between the article paragraph contexts and the relationship between the paragraph contexts.
And after obtaining the planar features of the articles, modeling the planar features of the articles and the article chapter structure scores is also needed. This type of feature is suitable for modeling using convolutional neural networks, considering that the planar features of an article contain information on the upper and lower paragraphs of the article, context information in the paragraphs, and a two-dimensional matrix is used for representation. Convolution operation is carried out on the article structural features through the convolution layer, similar to an n-gram model of a statement layer, experimental results show that a representation method through plane features and selection of the model are a very suitable mode, and high scoring accuracy is achieved.
Further, in an embodiment of the present invention, the method of an embodiment of the present invention further includes: forming a vector by the sentences preset in each paragraph, forming a matrix by the preset paragraphs, and filling 0 in the corresponding vacant part to express each article as the preset matrix to obtain the hierarchical characteristics of the article; modeling is carried out according to the hierarchical characteristics of the articles and the chapter structure scores of the articles, wherein a hierarchical neural network is adopted for modeling, paragraph-level information of the articles is obtained through GRUs, and the front-back connection relation between sentences of the articles is obtained through an LSTM network layer, so that a preset hierarchical characteristic model is obtained.
It can be understood that, when people review articles, paragraphs of the articles are observed and identified first, and then sentences in the paragraphs are identified. Therefore, the embodiment of the invention can represent the article as a hierarchical representation structure. Similar to the planar representation method, the hierarchical feature representation method also adopts a 10 × 10 feature representation method, and the difference is mainly that the hierarchical feature representation method is mainly applied to the hierarchical neural network model, and the design mode of the hierarchical neural network model is matched with the hierarchical model.
In addition, after the hierarchical features of the articles are obtained, the features of the articles and the corresponding chapter structure scores of the articles also need to be modeled. Embodiments of the present invention may employ a hierarchical neural network model. The hierarchical model mainly uses a paragraph-sentence hierarchical model. At the paragraph level, because of the sequence correlation between paragraphs, it is more appropriate to use GRU to obtain paragraph level information, and at the sentence level, to use LSTM network layer to obtain the context between sentences. It should be noted that, the hierarchical model is designed for hierarchical features, so that the method has good directivity, and experimental results show that the method achieves good effect.
Further, in one embodiment of the invention, the verbal element includes one of an introduction, a point of discourse, an example, a point of view, an emphasis, and a conclusion.
In addition, in order to enhance the experimental effect of article scoring, the embodiment of the invention mainly integrates the previous models.
After the extracted features of each partial model network layer are obtained, the SVM directly uses the sequence features, the features can be spliced by the embodiment of the invention, so that more dimensional feature information can be obtained, and a better effect can be achieved.
For example, the steps of automatically evaluating the discourse structure of the discussion by the planar discourse structure representation method of the convolutional neural network according to the embodiment of the present invention are as follows:
step S1: the method comprises the steps of obtaining article data, calibrating chapter attributes of paragraphs and sentences in the articles, and identifying chapter roles of each sentence according to the definitions of the sentences in the articles, wherein the calibrated evaluation criteria are shown in table 1.
TABLE 1
| Language piece element | Definition of |
| Title | Article title |
| Guide sentence | Introducing the subject matter of an article and stating the standpoint and main point of view of the author |
| Example sentence | Providing examples to support the subject matter and main points of discussion |
| Viewpoint sentence | Providing insight into articles or paragraphs |
| Supporting sentence | Providing support or assistance to a point of view |
| Statement sentence | Stating one of the main points or arguments |
| Conclusion sentence | Summarize and summarize the sentences of the discussion |
| Others | Sentences not meeting the requirements of the above elements or contributing to discourse structure |
Step S2: constructing a plane characteristic, representing the article as a 10 x 10 matrix after acquiring chapter characters of sentences according to a sentence definition method, wherein each row represents a section, the missing part is supplemented with 0, the exceeding part is truncated, and the article is represented as a plane characteristic.
Step S3: the method comprises the following steps of modeling by using a convolutional neural network, and using a two-layer convolutional neural network structure, wherein the structure of the neural network is shown in FIG. 4, the evaluation indexes of the embodiment of the invention adopt absolute errors and relative errors, and the specific calculation formula is as follows:
step S4: for the classification result, the high-quality discussion paper has relatively complete and visual structures on chapter structures, such as the association relations of guide sentences, talking point sentences, discourse sentences, conclusion sentences and the like, while for the low-quality chapters, various chapter structure forms can occur, the more common situation is that the conclusion sentences appear too early, the talking point sentences are redundant, and the guide sentences appear at a later position, and the examples all show that the articles have relatively poor structures. In addition, the articles with good chapter structures also have higher ornamental value in content, and the readability is improved due to the perfect chapter structures.
According to the evaluation method based on the discourse structure of the discussion, an effective discourse structure representation method is constructed for the discussion, sequential, planar and hierarchical representation methods are respectively adopted for discourse characteristics of the discussion, and then a neural network model is constructed for different discourse structure characteristic representation methods to automatically evaluate the discourse structure, so that the evaluation accuracy is effectively improved.
Next, an evaluation device based on the structure of the discussion-based chapter will be described with reference to the drawings.
Fig. 5 is a schematic structural diagram of an evaluation apparatus based on the structure of the discussion discourse according to an embodiment of the present invention.
As shown in fig. 5, the apparatus 10 for evaluating based on the structure of the discussion chapter includes: a first acquisition module 100, a first construction module 200 and an evaluation module 300.
The first obtaining module 100 is configured to obtain a plurality of sentence elements of the discussion to be evaluated according to the paragraph type and the sentence type of the discussion to be evaluated. The first construction module 200 is used for constructing a sequential feature, a planar feature or a hierarchical feature of an article chapter structure from a plurality of language chapter elements. The evaluation module 300 is configured to obtain an evaluation result of the discussion paper to be evaluated through a preset sequential feature model, a preset planar feature model, or a preset hierarchical feature model according to the sequential feature, the preset planar feature, or the preset hierarchical feature of the chapter structure of the article. The device 10 of the embodiment of the invention can obtain the evaluation result of the discussion paper to be evaluated through the preset sequence type feature model, the preset plane type feature model or the preset layer type feature model according to the sequence type feature, the preset plane type feature or the preset layer type feature of the article discourse structure, thereby effectively improving the evaluation accuracy.
Further, in one embodiment of the present invention, the apparatus 10 of the embodiment of the present invention further comprises: the device comprises a second acquisition module and a second construction module. The second obtaining module is used for filling the chapter elements of the sentences into corresponding parts of the discussion papers, and filling 0 into corresponding vacant parts, so as to represent each article as a vector with preset dimensions, and obtain the sequential characteristics of the articles. And the second construction module is used for constructing a classifier for modeling according to the sequence type characteristics of the article, wherein an SVM is adopted for model classification so as to obtain a preset sequence type characteristic model.
Further, in one embodiment of the present invention, the apparatus 10 of the embodiment of the present invention further comprises: a third obtaining module and a third constructing module. The third obtaining module is used for forming a vector by the preset sentences in each paragraph, forming a matrix by the preset paragraphs, and filling 0 in the corresponding vacant part so as to represent each article as the preset matrix and obtain the plane-type characteristics of the article. And the third construction module is used for carrying out modeling according to the planar characteristic of the article and the article chapter structure fraction, wherein the convolutional neural network is adopted for carrying out modeling so as to obtain a preset planar characteristic model.
Further, in one embodiment of the present invention, the apparatus 10 of the embodiment of the present invention further comprises: a fourth obtaining module and a fourth constructing module. The fourth obtaining module is used for forming a vector by the preset sentences in each paragraph, forming a matrix by the preset paragraphs, and filling 0 in the corresponding vacant part so as to represent each article as the preset matrix and obtain the hierarchical characteristics of the article. The fourth construction module is used for modeling according to the hierarchical features of the articles and the article chapter structure scores, wherein the hierarchical neural network is used for modeling, paragraph-level information of the articles is obtained through GRUs, and the front-back connection relation between sentences of the articles is obtained through an LSTM network layer, so that a preset hierarchical feature model is obtained.
Further, in one embodiment of the invention, the verbal element includes one of an introduction, a point of discourse, an example, a point of view, an emphasis, and a conclusion.
It should be noted that the foregoing explanation of the embodiment of the evaluation method based on the discourse structure of the discussion paper is also applicable to the evaluation device based on the discourse structure of the discussion paper of this embodiment, and will not be repeated herein.
According to the evaluation device based on the discourse structure of the discussion, an effective discourse structure representation method is constructed for the discussion, sequential, planar and hierarchical representation methods are respectively adopted for discourse characteristics of the discussion, and then a neural network model is constructed for different discourse structure characteristic representation methods to automatically evaluate the discourse structure, so that the evaluation accuracy is effectively improved.
In the description of the present invention, it is to be understood that the terms "central," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," "clockwise," "counterclockwise," "axial," "radial," "circumferential," and the like are used in the orientations and positional relationships indicated in the drawings for convenience in describing the invention and to simplify the description, and are not intended to indicate or imply that the referenced devices or elements must have a particular orientation, be constructed and operated in a particular orientation, and are therefore not to be considered limiting of the invention.
Furthermore, the terms "first", "second" and "first" are used for descriptive purposes only and are not to be construed as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one such feature. In the description of the present invention, "a plurality" means at least two, e.g., two, three, etc., unless specifically limited otherwise.
In the present invention, unless otherwise expressly stated or limited, the terms "mounted," "connected," "secured," and the like are to be construed broadly and can, for example, be fixedly connected, detachably connected, or integrally formed; can be mechanically or electrically connected; they may be directly connected or indirectly connected through intervening media, or they may be connected internally or in any other suitable relationship, unless expressly stated otherwise. The specific meanings of the above terms in the present invention can be understood by those skilled in the art according to specific situations.
In the present invention, unless otherwise expressly stated or limited, the first feature "on" or "under" the second feature may be directly contacting the first and second features or indirectly contacting the first and second features through an intermediate. Also, a first feature "on," "over," and "above" a second feature may be directly or diagonally above the second feature, or may simply indicate that the first feature is at a higher level than the second feature. A first feature being "under," "below," and "beneath" a second feature may be directly under or obliquely under the first feature, or may simply mean that the first feature is at a lesser elevation than the second feature.
In the description herein, references to the description of the term "one embodiment," "some embodiments," "an example," "a specific example," or "some examples," etc., mean that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the invention. In this specification, the schematic representations of the terms used above are not necessarily intended to refer to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples. Furthermore, various embodiments or examples and features of different embodiments or examples described in this specification can be combined and combined by one skilled in the art without contradiction.
Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention, and that variations, modifications, substitutions and alterations can be made to the above embodiments by those of ordinary skill in the art within the scope of the present invention.
Claims (4)
1. An evaluation method based on a discourse structure of a discussion paper is characterized by comprising the following steps:
acquiring a plurality of sentence elements of the discussion paper to be evaluated according to the paragraph type and the sentence type of the discussion paper to be evaluated;
constructing a sequential feature, a planar feature or a hierarchical feature of an article discourse structure through the plurality of language discourse elements; and
obtaining an evaluation result of the discussion paper to be evaluated according to the sequential characteristic, the planar characteristic or the hierarchical characteristic of the article discourse structure through a preset sequential characteristic model, a preset planar characteristic model or a preset hierarchical characteristic model;
the evaluation method based on the discourse structure of the discussion paper is characterized by further comprising the following steps: forming a vector by the sentences preset in each paragraph, forming a matrix by the preset paragraphs, and filling 0 in the corresponding vacant part to express each article as a preset first-class matrix to obtain the planar characteristics of the article; modeling according to the planar characteristic of the article and the article chapter structure fraction, wherein a convolutional neural network is adopted for modeling to obtain a preset planar characteristic model;
the evaluation method based on the discourse structure of the discussion paper is characterized by further comprising the following steps:
filling chapter elements of the sentences into corresponding parts of the discussion papers, and filling 0 into corresponding vacant parts so as to represent each article as a vector of a preset dimension and obtain sequence type characteristics of the article; constructing a classifier for modeling according to the sequence type characteristics of the article, wherein an SVM is adopted for model classification so as to obtain the preset sequence type characteristic model;
the evaluation method based on the discourse structure of the discussion paper is characterized by further comprising the following steps:
forming a vector by the preset sentences in each paragraph, forming a matrix by the preset paragraphs, and filling 0 in the corresponding vacant part to express each article as a preset second-class matrix to obtain the hierarchical characteristics of the article;
and modeling according to the hierarchical features of the articles and the article chapter structure scores, wherein a hierarchical neural network is adopted for modeling, paragraph-level information of the articles is obtained through GRUs, and the front-back connection relation between sentences of the articles is obtained through an LSTM network layer, so that a preset hierarchical feature model is obtained.
2. The method of claim 1, wherein the plurality of linguistic elements includes one of an introduction, a point of discourse, an instance, a point of view, an emphasis, and a conclusion.
3. An evaluation device based on the structure of the discourse of discussion, which is characterized by comprising:
the first acquisition module is used for acquiring a plurality of language elements of the discussion paper to be evaluated through the paragraph type and the sentence type of the discussion paper to be evaluated;
the first construction module is used for constructing sequential characteristics, planar characteristics or hierarchical characteristics of an article chapter structure through the plurality of language chapter elements; and
the evaluation module is used for obtaining an evaluation result of the discussion paper to be evaluated through a preset sequential feature model, a preset planar feature model or a preset hierarchical feature model according to the sequential feature, the preset planar feature or the preset hierarchical feature of the article chapter structure;
the device for evaluating based on the structure of the discourse of the discussion paper is characterized by further comprising:
the third acquisition module is used for forming a vector by the preset sentences in each paragraph, forming a matrix by the preset paragraphs, and filling 0 in the corresponding vacant part so as to represent each article as a preset first-class matrix and obtain the planar characteristics of the article;
the third construction module is used for carrying out modeling according to the planar characteristic of the article and the article chapter structure fraction, wherein a convolutional neural network is adopted for carrying out modeling so as to obtain a preset planar characteristic model;
the device for evaluating based on the structure of the discourse of the discussion paper is characterized by further comprising:
the second acquisition module is used for filling chapter elements of the sentences into corresponding parts of the discussion papers, and filling 0 into corresponding vacant parts so as to represent each article as a vector with preset dimensionality and obtain sequence type features of the articles;
the second construction module is used for constructing a classifier for modeling according to the sequence type characteristics of the article, wherein an SVM is adopted for model classification so as to obtain the preset sequence type characteristic model;
the device for evaluating based on the structure of the discourse of the discussion paper is characterized by further comprising:
the fourth acquisition module is used for forming a vector by the preset sentences in each paragraph, forming a matrix by the preset paragraphs, and filling 0 in the corresponding vacant part so as to represent each article as a preset second-class matrix and obtain the hierarchical characteristics of the article;
and the fourth construction module is used for carrying out modeling according to the hierarchical characteristics of the article and the chapter structure scores of the article, wherein the modeling is carried out by adopting a hierarchical neural network, paragraph-level information of the article is obtained through a GRU (general-purpose Unit), and the front-back connection relation between sentences of the article is obtained through an LSTM network layer, so that a preset hierarchical characteristic model is obtained.
4. The apparatus of claim 3, wherein the plurality of linguistic elements includes one of an introduction, a point of discourse, an example, a point of view, an emphasis, and a conclusion.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810182942.4A CN108595407B (en) | 2018-03-06 | 2018-03-06 | Evaluation method and device based on discourse structure of discussion treatise |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810182942.4A CN108595407B (en) | 2018-03-06 | 2018-03-06 | Evaluation method and device based on discourse structure of discussion treatise |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108595407A CN108595407A (en) | 2018-09-28 |
| CN108595407B true CN108595407B (en) | 2022-03-18 |
Family
ID=63625800
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810182942.4A Expired - Fee Related CN108595407B (en) | 2018-03-06 | 2018-03-06 | Evaluation method and device based on discourse structure of discussion treatise |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108595407B (en) |
Families Citing this family (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110096710B (en) * | 2019-05-09 | 2022-12-30 | 董云鹏 | Article analysis and self-demonstration method |
| CN110795538B (en) * | 2019-10-30 | 2023-09-26 | 腾讯科技（深圳）有限公司 | Text scoring method and related equipment based on artificial intelligence |
| US11144719B2 (en) | 2019-11-27 | 2021-10-12 | International Business Machines Corporation | System and method for argument retrieval |
| CN112784878A (en) * | 2020-12-31 | 2021-05-11 | 北京华图宏阳网络科技有限公司 | Intelligent correction method and system for Chinese discussion papers |
| CN113743050B (en) * | 2021-09-07 | 2023-11-24 | 平安科技（深圳）有限公司 | Article layout evaluation method, apparatus, electronic device and storage medium |
| CN114880428B (en) * | 2022-03-07 | 2022-11-18 | 中国人民解放军国防科技大学 | Method for recognizing speech part components based on graph neural network |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO1999021106A1 (en) * | 1997-10-20 | 1999-04-29 | Microsoft Corporation | Automatically recognizing the discourse structure of a body of text |
| CN102779220A (en) * | 2011-05-10 | 2012-11-14 | 李德霞 | English test paper scoring system |
| CN105955956A (en) * | 2016-05-05 | 2016-09-21 | 中国科学院自动化研究所 | Chinese implicit discourse relation identification method |
| CN107168956A (en) * | 2017-05-26 | 2017-09-15 | 北京理工大学 | A kind of Chinese structure of an article analysis method and system based on pipeline |
| CN107273358A (en) * | 2017-06-18 | 2017-10-20 | 北京理工大学 | A kind of end-to-end English structure of an article automatic analysis method based on pipe modes |
| CN107506360A (en) * | 2016-06-14 | 2017-12-22 | 科大讯飞股份有限公司 | A kind of essay grade method and system |
-
2018
- 2018-03-06 CN CN201810182942.4A patent/CN108595407B/en not_active Expired - Fee Related
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO1999021106A1 (en) * | 1997-10-20 | 1999-04-29 | Microsoft Corporation | Automatically recognizing the discourse structure of a body of text |
| CN102779220A (en) * | 2011-05-10 | 2012-11-14 | 李德霞 | English test paper scoring system |
| CN105955956A (en) * | 2016-05-05 | 2016-09-21 | 中国科学院自动化研究所 | Chinese implicit discourse relation identification method |
| CN107506360A (en) * | 2016-06-14 | 2017-12-22 | 科大讯飞股份有限公司 | A kind of essay grade method and system |
| CN107168956A (en) * | 2017-05-26 | 2017-09-15 | 北京理工大学 | A kind of Chinese structure of an article analysis method and system based on pipeline |
| CN107273358A (en) * | 2017-06-18 | 2017-10-20 | 北京理工大学 | A kind of end-to-end English structure of an article automatic analysis method based on pipe modes |
Non-Patent Citations (5)
| Title |
|---|
| Attention-based Recurrent Convolutional Neural Network for Automatic Essay Scoring;Fei Dong 等;《Proceedings of the 21st Conference on Computational Natural Language Learning (CoNLL 2017)》;20171231;第153-162页 * |
| Discourse Element Identification in Student Essays based on Global and Local Cohesion;Wei Song 等;《Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing》;20151231;第2255-2261页 * |
| Discourse Mode Identification in Essays;Wei Song 等;《Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics》;20171231;第112-122页 * |
| Discourse Structure Identification for Knowledge Extraction;Jamie Guidry 等;《Proceedings of the 2013 Industrial and Systems Engineering Research Conference》;20131231;第214页 * |
| 基于树核的隐式篇章关系识别;徐凡 等;《软件学报》;20130515;第1022-1035页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108595407A (en) | 2018-09-28 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108595407B (en) | Evaluation method and device based on discourse structure of discussion treatise | |
| Lee et al. | Sentiment classification with word localization based on weakly supervised learning with a convolutional neural network | |
| Liu et al. | Assessing product competitive advantages from the perspective of customers by mining user-generated content on social media | |
| CN107330049B (en) | News popularity estimation method and system | |
| WO2019153737A1 (en) | Comment assessing method, device, equipment and storage medium | |
| CN107239444B (en) | A kind of term vector training method and system merging part of speech and location information | |
| CN106021223B (en) | Sentence similarity calculation method and system | |
| US10762992B2 (en) | Synthetic ground truth expansion | |
| US9224155B2 (en) | Systems and methods for managing publication of online advertisements | |
| CN107957993B (en) | English sentence similarity calculation method and device | |
| CN111966917A (en) | Event detection and summarization method based on pre-training language model | |
| Shi et al. | Sentiment analysis of Chinese microblogging based on sentiment ontology: a case study of ‘7.23 Wenzhou Train Collision’ | |
| US20120253792A1 (en) | Sentiment Classification Based on Supervised Latent N-Gram Analysis | |
| CN104137102A (en) | Non-factoid question answering system and computer program | |
| CN110442872B (en) | Text element integrity checking method and device | |
| JP2017138655A (en) | Bilingual corpus creation method, device and program for the same, and machine translation system | |
| CN109492105B (en) | Text emotion classification method based on multi-feature ensemble learning | |
| CN104504023A (en) | High-accuracy computer automatic marking method for subjective items based on domain ontology | |
| US20200192921A1 (en) | Suggesting text in an electronic document | |
| CN113962219A (en) | Semantic matching method and system for knowledge retrieval and question answering of power transformer | |
| CN109255012A (en) | A kind of machine reads the implementation method and device of understanding | |
| CN116541493A (en) | Interactive response method, device, equipment and storage medium based on intention recognition | |
| CN113434682A (en) | Text emotion analysis method, electronic device and storage medium | |
| JP6757840B2 (en) | Sentence extraction system, sentence extraction method, and program | |
| Lefevre | Dynamic bayesian networks and discriminative classifiers for multi-stage semantic interpretation |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20220318 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |