A network text public sentiment mining method and system based on supervised learning

Abstract

The application discloses a network text public emotion mining method and system based on supervised learning, and establishes a coding framework of network text multidimensional public emotion according to research literature, expert opinions and experience of dissemination science; coding training is conducted on each coder so that each coder reaches consistency; network text is captured, and each coder carries out scoring processing according to the given coding framework; a text feature expressing a topic and content in the network text is extracted by using a LIWC dictionary, and a text data set is constructed; a supervised machine learning method is used, network text scoring and the text data set are combined for training, a text analysis model is obtained through matrix parameter adjustment and cross-validation methods; and the evaluation subject attribute in new network text and multidimensional public emotion scores are obtained by using the text analysis model. The application can accurately and professionally evaluate the public emotion of short texts, and solves the problem of lack of Chinese annotation materials of short text public emotion in the evaluation of daily situations.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of natural language processing and data mining, in particular to a network text public sentiment mining method and system based on supervised learning. BACKGROUND

[0002] With the popularity of the Internet and online social media, people are increasingly used to expressing opinions and views on social media. Because of the popularity of users, the real-time nature of communication and the diversity of topics, the information produced by users on social media has become an important window for governments, companies and researchers to observe public opinion.

[0003] However, existing text processing systems do not clearly indicate the direction of the emotion. Some of the content in the online expression is directed at individuals, while some expresses the sentiment towards society. It is this sentiment towards society that is the basis of public opinion. Furthermore, the public sentiment evaluation of short texts produced by users on social media is particularly important because, on the one hand, social media is the main place for the expression of public opinion, and semantic analysis of colloquial short texts helps relevant parties to understand public opinion in a timely and targeted manner. On the other hand, public sentiment reflects users' views on the subject of the comment and also affects the spread of information and the attitude of the reading audience towards the subject of the comment. Quantifying the public sentiment of text information helps to monitor public opinion, feedback on goods and identify important opinion leaders.

[0004] The analysis of network text is also challenging. Compared with long texts and written information, network short texts are usually very short and contain more informal words. It is difficult to disambiguate due to the limited context available for reference. In addition, even if people want to express their emotional attitude towards society, they will hide direct expressions and use informal words, so direct application of existing and standardized dictionaries cannot identify the correct direction of public opinion. These factors make it difficult to distinguish between emotions towards society and emotions towards individuals in social media text. SUMMARY

[0005] In order to solve the above-mentioned technical problems, in order to quantitatively evaluate the public sentiment of short texts, the present application provides a network text public sentiment mining method and system based on supervised learning. After inputting the text, the system can quantitatively evaluate the scores of network text on each dimension of public sentiment.

[0006] The technical solutions adopted are as follows:

[0007] On the one hand, the present application provides a network text public sentiment mining method based on supervised learning, which comprises the following steps:

[0008] Step 1, according to the research literature of communication, expert opinions and experience, a multi-dimensional public emotion coding framework of network text is established;

[0009] Step 2, the coders are trained to achieve coding consistency;

[0010] Step 3, the network text is captured, and the coders score the network text according to the given coding framework;

[0011] Step 4, the LIWC dictionary is used to extract the text features expressing the topic and content in the network text, and a text data set is constructed;

[0012] Step 5, a supervised machine learning method is used, combined with the network text scoring in step 3 and the text data set constructed in step 4, the text analysis model is trained through matrix parameter adjustment and cross-validation method;

[0013] Step 6, the network text is input into the text analysis model, and the evaluation subject attribute and multi-dimensional public emotion score in the network text are obtained.

[0014] Further, the coding training of the coders in step 2 is as follows:

[0015] Step 2.1, train the coders to fully understand the coding framework;

[0016] Step 2.2, select network text, and let the coders score according to the coding framework;

[0017] Step 2.3, the consistency level of the coders is tested by the inter-rater reliability, when the score exceeds the set threshold, it is considered to have good consistency level, and the preferred threshold is 65%;

[0018] If the coder consistency meets the standard, the coder training is ended; if the coder consistency does not meet the standard, the coding framework in step 2.1 is revised, and the coders are retrained until the coder consistency meets the standard.

[0019] Preferably, the coding framework of multi-dimensional public emotion of network text established in step 1 includes the following emotional states: safe, happy, proud, hopeful, frustrated, angry, afraid, sad and disappointed.

[0020] Further, the random forest, convolutional neural network and support vector machine model are used to perform supervised tasks in step 5, and the best text analysis model is selected according to the accuracy evaluation index.

[0021] Further, the method further comprises step 7, manually verifying the evaluation subject attribute and public emotion score in the obtained network text.

[0022] In another aspect, the present application also provides a network text public sentiment mining system based on supervised learning, which runs in a computer, and the system comprises:

[0023] An evaluation system construction module, which establishes a coding framework with multi-dimensional public sentiment according to research literature of communication, expert opinions and experience;

[0024] An evaluation system evaluation module, which is used to verify the consistency of the network text coding of the coder, and if the consistency meets the standard, the coder training is ended, otherwise, the coding framework is revised and the coder is retrained until the consistency of the coder meets the standard;

[0025] A sample evaluation module, which is used to collect network text data, and according to the established coding framework, the coder respectively scores the collected network text;

[0026] A model construction module, which extracts text features expressing topics and contents in the network text data by using the LIWC dictionary, constructs a text data set based on the related screening features, and combines the scoring of the network text by the sample evaluation module to train the text by a supervised machine learning model; the text analysis model is trained by using the matrix parameter adjustment and cross-validation method;

[0027] The text analysis model inputs the captured network text data into the trained text analysis model to obtain the evaluation subject attribute and the multi-dimensional public sentiment score in the network text.

[0028] Further, the system further comprises a model verification module, which measures the training effect of the text analysis model by using the accuracy evaluation index, and the specific calculation formula is:

[0029]

[0030] Wherein: TP is the true positive result, TN is the true negative result;

[0031] FP is the false positive result, and FN is the false negative result.

[0032] The technical scheme of the present application has the following advantages:

[0033] A.The method and system provided by the application first constructs a coding framework with multi-dimensional public emotion according to communication research literature, expert opinions and experience, and trains the coders for network text coding, and in the coding process, supervised artificial coding is included, and a part of colloquial words are included in the liwc dictionary used, so that the key colloquial words existing in the network text can be accurately extracted, thereby avoiding the omission of key words in the text, and the public emotion of the short text is accurately and professionally evaluated, and the problem of lack of Chinese annotation materials of public emotion of short text in daily situation is solved.

[0034] B.The application uses expert experience and supervised machine learning effective technology to independently analyze the topic of network text and the attribute of the text, enhances the extraction ability of the implied informal words in the network text, significantly improves the ability of network text public opinion mining, provides an effective method for understanding the network text, trains the machine learning model to a high accuracy rate after the micro-blog text is coded, and the text analysis model constructed can score the public emotion of the text to obtain the public opinion direction expressed by the text. BRIEF DESCRIPTION OF DRAWINGS

[0035] In order to more clearly illustrate the specific embodiments of the application, the drawings needed in the specific embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the application, and other drawings can be obtained by those skilled in the art without creative labor.

[0036] Figure 1 is a method flowchart provided by the application;

[0037] Figure 2 is a system composition structure relationship diagram provided by the application;

[0038] Figure 3 is a network text public emotion mining flowchart provided by the application. DETAILED DESCRIPTION

[0039] The technical solutions of the application will be described in detail below with reference to the drawings. Obviously, the described embodiments are part of the embodiments of the application, not all embodiments. Based on the embodiments of the application, all other embodiments obtained by those skilled in the art without creative labor belong to the protection scope of the application.

[0040] As shown in Figure 1 and Figure 3 , the application provides a network text public emotion mining method based on supervised learning, including the following steps:

[0041]

S01

[0042]

[0043]

S02

[0044]

S021

[0045]

S022

[0046]

S023

[0047] If the consistency of the coders meets the standard, the coder training is ended; if the consistency of the coders does not meet the standard, the coding framework in

S021

[0048]

S03

[0049] When collecting data, all user information of the popular user list of each category of Sina Weibo is grabbed, and users on the list are randomly selected, and accounts with more than 1 year of micro-blog records and original micro-blog every month are selected. All original micro-blogs of the selected users are grabbed.

[0050] The coders encode the collected data, such as randomly selecting 1200 micro-blogs from the grabbed micro-blogs, and each coder encodes the text according to the coding framework.

[0051]

S04

[0052] When text features are preprocessed, the LIWC dictionary is used to process the topics and content expressed in the text separately, including some colloquial and informal language that directly reflects the public opinion tendency. For news text, the topic is the news title;

[0053] Feature selection: use a correlation-based feature selection method to screen features.

[0054]

S05

[0055] Model construction of supervised learning: use random forest, convolutional neural network, support vector machine and other models to perform supervised tasks, and select the best model according to the evaluation index.

[0056] Model parameter optimization using cross-validation: use ten-fold cross-validation method, divide all data into ten parts, then take each part as the validation set and the others as the training set to train and validate. Finally output a set of well-optimized hyperparameters. The optimized parameters will be used to train the final model to continue to perform downstream tasks.

[0057]

S06

[0058]

S07

[0059] As Figure 2The application also provides a network text public emotion mining system based on supervised learning, which is stored and run in a computer, and the system comprises an evaluation system construction module, an evaluation system evaluation module, a sample evaluation module, a model construction module, a model verification module and a text analysis model. The evaluation system construction module is used for establishing a coding framework with multi-dimensional public emotion according to communication research literature, expert opinions and experience; the evaluation system evaluation module is used for verifying the consistency of the coding of network text by coders, and if the consistency meets the standard, the coder training is ended, otherwise, the coding framework is revised and the coder is retrained until the consistency of the coder meets the standard; the sample evaluation module is used for collecting network text data, and coders respectively score the collected network text according to the established coding framework; the model construction module extracts text features expressing topics and contents in the network text data by using the LIWC dictionary, constructs a text data set based on the related screened features, and combines the scoring of the network text by the sample evaluation module to train the text by a supervised machine learning model; the text analysis model is trained by using the matrix parameter adjustment and cross-validation method; the text analysis model inputs the captured network text data into the trained text analysis model to obtain the evaluation subject attribute and the multi-dimensional public emotion score in the network text; in the model verification module, the accuracy evaluation index is used to measure the training effect of the text analysis model, and the specific calculation formula is:

[0060]

[0061] Wherein, TP is a true positive result, TN is a true negative result;

[0062] FP is a false positive result, and FN is a false negative result.

[0063] The application firstly constructs a public emotion evaluation framework according to expert opinions and practice, and trains coders; secondly, microblog text is collected, and after scoring processing by the coders, a machine learning model is trained to achieve a high accuracy; when processing network text, the application can score the public emotion of the text. The application uses expert experience and effective supervised machine learning technology to independently analyze topics and text attributes, improves the ability of network text public opinion mining, and provides an effective method for understanding network text.

[0064] The application does not describe the prior art.

[0065] Obviously, the above embodiments are only examples for clearly illustrating but not limiting the implementation. For those skilled in the art, other different forms of changes or variations can be made on the basis of the above description. Here, all the implementation is not enumerated, and the obvious changes or variations derived therefrom are still within the protection scope of the application.

Claims

1. A network text public sentiment mining method based on supervised learning, characterized in that, The method comprises the following steps: Step 1, according to the research literature of communication, expert opinions and experience, a coding framework of multi-dimensional public emotion of network text is established; Step 2, the coders are trained to achieve coding consistency; Step 3, network text is captured, and the coders score the network text according to the given coding framework; Step 4, the text features expressing topics and contents in the network text are extracted by using the LIWC dictionary, and a text data set is constructed; Step 5, a supervised machine learning method is used, and the network text scoring in step 3 and the text data set constructed in step 4 are combined for training, and the text analysis model is obtained through matrix parameter adjustment and cross-validation method; Step 6, the evaluation subject attribute and multi-dimensional public emotion score in the network text are obtained by inputting the network text into the text analysis model; In step 2, the coding training of the coders is as follows: Step 2.1, the coders are trained to fully understand the coding framework; Step 2.2, network text is selected, and the coders score the network text according to the coding framework; Step 2.3, the consistency level of the coders is tested by using the scorer consistency reliability, and when the score exceeds the set threshold, it is considered that the coder has good consistency level; If the coder consistency meets the standard, the coder training is ended; If the coder consistency does not meet the standard, the coding framework in step 2.1 is revised, and the coders are retrained until the coder consistency meets the standard. 2.The method of claim 1, wherein, The coding framework of multi-dimensional public emotion of network text established in step 1 comprises the following emotional states: safe, happy, proud, hopeful, frustrated, angry, afraid, sad and disappointed. 3.The method of claim 1, wherein, In step 5, random forest, convolutional neural network and support vector machine model are used to perform supervised task, and the best text analysis model is selected according to the accuracy evaluation index. 4.The method of claim 1, wherein, Step 7, the evaluation subject attribute and public emotion score in the obtained network text are manually verified.

5. A system for mining public sentiments from network texts based on supervised learning, operating in a computer, characterized in that, The system comprises: An evaluation system construction module for establishing a coding framework with multi-dimensional public emotion according to the research literature of communication, expert opinions and experience; An evaluation system evaluation module for verifying the consistency of the coders in coding network text, if the consistency meets the standard, the coder training is ended, otherwise, the coding framework is revised and the coders are retrained until the coder consistency meets the standard; A sample evaluation module for collecting network text data, and organizing the coders to score the collected network text according to the established coding framework; A model construction module for extracting text features expressing topics and contents in the network text data by using the LIWC dictionary, constructing a text data set based on related screening features, and combining the scoring of the network text by the sample evaluation module to train the text by using a supervised machine learning model; the text analysis model is trained by using matrix parameter adjustment and cross-validation method; A text analysis model for inputting the captured network text data into the trained text analysis model to obtain the evaluation subject attribute and multi-dimensional public emotion score in the network text; A text analysis model for inputting the captured network text data into the trained text analysis model to obtain the evaluation subject attribute and multi-dimensional public emotion score in the network text; The implementation method of the evaluation system evaluation module comprises the following steps: Step 2.1, training the encoder, so that each encoder fully understands the encoding framework; Step 2.2, selecting network text, so that each encoder scores according to the encoding framework; Step 2.3, using the scorer consistency reliability to test the consistency level of each encoder, and when the score exceeds the set threshold, it is considered to have a better consistency level; If the encoder consistency meets the standard, the encoder training is ended; if the encoder consistency does not meet the standard, the encoding framework in step 2.1 is revised, and the encoder is retrained until the encoder consistency meets the standard. 6.The system according to claim 5, wherein, The system also includes a model verification module that uses an accuracy evaluation index to measure the training effect of the text analysis model, and the specific calculation formula is: ； Where: TP is the true positive result, TN is the true negative result; FP is the false positive result, and FN is the false negative result.

Images