TW201513013A - Method of digging product evaluation words in electronic articles and system thereof - Google Patents

Abstract

A method of digging product evaluation words in electronic articles and system thereof are disclosed. A plurality of articles that receive public praises are selected from a plurality of universal test data and a universal prediction model is established. A prediction is conducted to a plurality of industrial test data according to the universal prediction model, and new positive and negative words of mouth are then determined and dug. The positive and negative words of mouth, either new or old, are accumulated and a plurality of new narrative sentences of words of mouth are extracted to establish a new prediction model. A prediction is conducted again to the industrial test data to repeat the processes of establishing a new prediction model, and digging new positive and negative words of mouth and new narrative sentences of words of mouth. All accumulated positive and negative words of mouth are not imported to a universal vocabulary database, which is used as common vocabularies, until no new positive and negative words of mouth are discovered.

Description

Product evaluation word excavation method and system thereof for electronic article

The invention relates to a method and system for excavating a product evaluation word for an electronic article, in particular to a word for examining product evaluation, and establishing a general vocabulary database for electronic measurement on the network. In the evaluation of articles in the article, it is possible to filter out the evaluation of specific products on the Internet by the public opinion words based on the product evaluation words faster and more accurately.

The Internet has become the main channel for the dissemination of commodity information. Consumers will query and compare prices online before the sale. In addition to the price, the larger part is to ask the public on the Internet to use the products they want to buy. Evaluation, even consumers will seek technical search or negative feedback after spending online, so if the owner can provide the necessary assistance or explanation at the above time, it can reduce the user. Dissatisfaction, it is more able to positively increase consumer confidence in the product to increase the consumer base of the product.

Therefore, some operators will screen the articles on the Internet to evaluate the specific products on the Internet. However, the definitions and usages of some evaluation terms will be different in different fields and industries, and even some industries (domains) have their own uniqueness. The evaluation of the words, and these terms require the field experts to spend a lot of manpower to obtain, and the general definition of the general terms, because the integrity is not high, it is not necessarily applicable to all industries.

Therefore, if you can explore the wording of product evaluation, you can gradually establish a sustainable The updated general vocabulary database can be used across domains, and after repeated tests in different fields, it is possible to continuously add new generic terms, which will enable continuous growth and learning so that the general vocabulary database can It is used in a variety of different fields or industries, and this should be an optimal solution.

The present invention provides a method and system for excavating a product evaluation word for an electronic article, which is capable of discovering a product evaluation term and establishing a general vocabulary database for conducting an electronic article on the network. In the evaluation of commodities, it is possible to screen out the evaluation of specific commodities on the Internet by the public opinion words based on the product evaluation terms.

The above-mentioned product evaluation word excavation method and system for electronic article can be achieved, wherein the method for excavating a product evaluation word for an electronic article is as follows: 1. First select a few articles containing word-of-mouth with several general test data, and take out a few positive and negative word-of-mouth words and several word-of-mouth narration sentences, then set and establish according to the words of positive and negative word-of-mouth and word-of-mouth narration. a general predictive model; 2. After that, based on the general forecasting model, predict the test data of several industrial fields, and judge and explore new positive and negative word-of-mouth terms used in the industrial field; 3. After accumulating the existing positive and negative word-of-mouth words with the new positive and negative word-of-mouth words, the accumulated positive and negative word-of-mouth words are taken out from several articles containing word-of-mouth to extract several new word-of-mouth narrative sentences, and then based on the cumulative The positive and negative word-of-mouth word and the new word-of-mouth narrative are used to set rules for labeling text to establish a new prediction model; 4. Finally, based on the new forecasting model, the test data of several industrial fields will be predicted again, and the new positive and negative word-of-mouth words and new word-of-mouth narratives will be judged again. After the sentence, the new prediction model is established and the new positive and negative word-of-mouth words and new word-of-mouth narratives are discovered. After the new positive and negative word-of-mouth words are not discovered, the excavation process is terminated.

More specifically, after the end of the excavation process, all accumulated positive and negative word-of-mouth words are imported into a common vocabulary database for use as a shared vocabulary.

More specifically, the article is a web article.

The product evaluation word excavation system for electronic articles of the present invention comprises a test data input module, a general predictive model building module, a proprietary vocabulary mining module, a vocabulary accumulating module, and a narrative sentence. The capture module and the predictive model training module, wherein the test data input module is configured to input test data, the test data includes at least a plurality of general test data and a plurality of industrial field test materials; and the universal predictive model The building module system comprises a word-of-mouth article selection module, which is connected with the test data input module, and the word-of-mouth article selection module is used to select a number of articles containing word-of-mouth from a plurality of general test materials; The word selection module is connected with the word-of-mouth article selection module, and is capable of sorting out a number of positive and negative word-of-mouth words in a number of articles containing word-of-mouth; a narrative sentence capture module, and the word-of-mouth article selection module And the word-of-mouth is connected by a word selection module, and uses a positive and negative word-of-mouth word to extract a plurality of word-of-mouth narration sentences in a number of articles containing word-of-mouth; a general prediction model training model And the word selection module and the narrative extraction module are connected to set a rule for labeling the text according to the positive and negative word-of-mouth word and the word-of-mouth sentence to establish a general prediction model; The proprietary vocabulary mining module is connected with the test data input module and the general predictive model training module, and is capable of predicting and testing a plurality of industrial field test data according to a general predictive model, and judging and extracting for use in the industrial field. New words for positive and negative word of mouth; and the word The cumulative module is connected with the proprietary vocabulary mining module and the word-of-mouth word selection module for accumulating the existing positive and negative word-of-mouth terms with the new positive and negative word-of-mouth terms; The group is connected with the word-of-mouth article selection module and the vocabulary accumulation module, and uses the accumulated positive and negative word-of-mouth words to extract a number of new word-of-mouth narrative sentences in the articles containing word-of-mouth; the prediction model training module, The system is connected with the vocabulary accumulation module, the narrative sentence capture module and the exclusive vocabulary search module, and is configured to set a rule for labeling the text according to the accumulated positive and negative word-of-mouth word and the new word-of-mouth narration sentence, A new predictive model is established, and the proprietary vocabulary mining module can predict a number of industrial field test data based on the new predictive model, and repeatedly judge and extract new positive and negative word-of-mouth terms for the industrial field. .

More specifically, the commodity evaluation word excavation system for an electronic article further includes a general vocabulary database connected to the vocabulary accumulation module, wherein the common vocabulary database has a shared positive and negative reputation. The word, and the proprietary vocabulary mining module judges that there is no new positive or negative word-of-mouth word, and then uses the vocabulary accumulation module to import the accumulated positive and negative word-of-mouth words into the general vocabulary database.

More specifically, the article is a web article.

〔this invention〕

11‧‧‧Test data input module

12‧‧‧General predictive model building module

121‧‧‧Words article selection module

122‧‧‧Words word selection module

123‧‧‧ Narrative Sentence Module

124‧‧‧General Predictive Model Training Module

13‧‧‧Special vocabulary mining module

14‧‧‧ vocabulary accumulation module

15‧‧‧ Narrative Sentence Module

16‧‧‧ Forecast Model Training Module

17‧‧‧Common vocabulary database

18‧‧‧ Domain vocabulary database

Fig. 1 is a flow chart showing the method for excavating a product for electronic evaluation of the article and the system thereof.

Fig. 2 is a schematic view showing the overall structure of a product evaluation word excavation method and system thereof for use in an electronic article.

The above and other technical contents, features and advantages of the present invention will be apparent from the following detailed description of the preferred embodiments.

Please refer to FIG. 1 , which is a schematic flowchart of a method for excavating a product evaluation word and a system thereof for an electronic article according to the present invention. The steps are as follows: 1. First, select a few pens from a plurality of general test materials. After the articles containing word-of-mouth, and taking out a number of positive and negative word-of-mouth words and several word-of-mouth narration sentences, then based on the positive and negative word-of-mouth words and word-of-mouth narration sentences, set up and establish a general prediction model 101; 2. After that, according to the general The forecasting model predicts several test data in the industrial field, and judges and discovers the new positive and negative word-of-mouth word 102 used in the industrial field; 3. Then accumulates the existing positive and negative word-of-mouth words with the new positive and negative word-of-mouth words. After the cumulative positive and negative word of mouth is used to extract several new word-of-mouth narrative sentences in several articles containing word-of-mouth, the rules for labeling the characters are set according to the accumulated positive and negative word-of-mouth words and the new word-of-mouth sentence. Establish a new forecasting model 103; and 4. Finally, based on the new forecasting model, re-predict several industrial field test data, and judge and discover new ones again. Negative word-of-mouth and new word-of-mouth narratives, after which repeated new prediction model establishments and new positive and negative word-of-mouth words and new word-of-mouth narratives were discovered, and after the new words of positive and negative word-of-mouth were not discovered, The discovery process 104 ends.

Please refer to FIG. 2 , the product evaluation word excavation system for electronic articles includes a test data input module 11 , a general predictive model building module 12 , a proprietary vocabulary excavation module 13 , and a a vocabulary accumulation module 14, a narration capture module 15, a predictive model training module 16, a general vocabulary database 17, and a domain vocabulary database 18, wherein the test The test data input module 11 is used for inputting test data, and the test data that can be input includes a plurality of general test data or a plurality of industrial field test materials; and the general predictive model building module 12 includes a word-of-mouth article selection. The module 121, the word-of-mouth word selection module 122, the narration sentence capturing module 123 and a general predictive model training module 124, the general predictive model building module 12 is mainly used to establish a general predictive model. When the test data input module 11 inputs a plurality of general test data into the word-of-mouth article selection module 121, the word-of-mouth article selection module 121 selects a plurality of general-purpose test materials and contains a number of words that have been included in the word-of-mouth. Articles, such as an article, have the following content: "My family started to drink Quaker's new beauty, but it will always be convenient, like diarrhea, and later changed to Nestle S26, drinking S26 will be more puffy (now look At that time, the photos all felt that he was strong.) And for a while he would overflow the milk, so he changed to the new Angel's nursing powder (the children of my classmates also drink this can), after changing the new Angel Drinking is not bad, now one year old has changed the Angel milk of Angel, and provide water and water reference!", because this article has the evaluation words such as Quaker, New Beauty, diarrhea, and good, so you can After the article is selected, the word-of-mouth word selection module 122 connected to the word-of-mouth article selection module 121 is used to sort out a number of positive and negative word-of-mouth words in the articles containing word-of-mouth, with the following contents as examples. Explain, "Our family originally intended to drink S26, I also think that it will be puffy, and now change to drink eugenics," which can be marked as negative word of mouth, and changed to drink as a positive word of mouth; and marked and sorted out After a number of positive and negative word-of-mouth words, the narration sentence extraction module 123 is connected with the word-of-mouth article selection module 121 and the word-of-mouth word selection module 122, and the used positive and negative word-of-mouth words are used in the number of pens to contain word of mouth. In the article, a number of word-of-mouth narration sentences are taken, and the narration sentence capture module 123 can obtain the location of the product and the word-of-mouth, which will be complete. Sentences, such as the following articles with word of mouth: <div class="body dfs"><p>You can go to the mom classroom</p><p>Pregnant women also need some activity~</p><p >Participate in the mother classroom, learn to be a mother</p><p>One can meet other mommy as a classmate~</p><p>Some courses can get small gifts. I think it is also a good added value~</ p><p>Mummy classroom courses are also full.</p><p>It’s like regenerative, eugenics, and mamaway. It’s not bad.</p><p>The knowledge is true. Very useful ~</p> Among them, "eugenics" and "good" are both positive and negative word-of-mouth. Therefore, the words "like eugenics, eugenics, mamaway" are selected in this article by "eugenics" and "good". It has been a good experience to participate in the experience, and the narrative sentence has a long sentence sentence rule and a short sentence sentence rule, in which the sentence length will be checked first to judge the long sentence sentence (in this embodiment, each sentence is set) If the length is greater than 25 characters, the long sentence is broken or the short sentence is broken (if each sentence is less than 25 characters in length, the short sentence is broken) Sentence punctuation rules is based on punctuation punctuation, word negative intercept terms coverage, whereas the phrase punctuation rules before and after the sentence is based on the phrase extension to intercept terms of coverage of the positive and negative reputation.

Then, the general predictive model training module 124 connected to the word-of-mouth word selection module 122 and the sentence sentence capturing module 123 sets a rule for labeling characters according to the positive and negative word-of-mouth word and the word-of-mouth sentence. The general predictive model training module 124 uses the Conditional Random Fields (CRF) to learn the system, and performs the action training of the input characters, and learns the knowledge used in the previous labeling by the algorithm. The previous text is labeled to create a general predictive model.

After the general predictive model is established, the proprietary vocabulary mining module 13 connected to the test data input module 11 and the universal predictive model training module 124 inputs the test data into the module according to the general predictive model. The input of several industrial field test data is predicted, and if it is not the previously used positive and negative word-of-mouth terms, it is taken out as a new positive and negative word-of-mouth term for the industrial field, and then, the model is developed with the exclusive vocabulary. The group 13 and the vocabulary accumulation module 14 connected by the word-of-mouth word selection module 122 accumulate the existing positive and negative word-of-mouth words and the new positive and negative word-of-mouth words, and the accumulated new positive and negative word-of-mouth words are more capable. Directly stored in a domain vocabulary database 18, so that domain vocabulary databases in different fields will be able to accumulate proprietary terms in different fields; and then connected to the word-of-mouth article selection module 121 and the vocabulary accumulation module 14 The narrative sentence capture module 15 can then use the accumulated positive and negative word-of-mouth words to extract a number of new word-of-mouth narrative sentences in a number of articles containing word-of-mouth, and then be tired by the vocabulary The module 14 , the narrative retrieval module 15 and the predictive model training module 16 connected to the exclusive vocabulary excavation module 13 are configured to mark the text by accumulating the positive and negative word-of-mouth terms and the new word-of-mouth narrative. The rules are used to establish a new forecasting model; after that, the proprietary vocabulary mining module 13 re-predicts the test data of the industrial field based on the new forecasting model, and repeatedly judges and extracts new products for the industrial field. The positive and negative word-of-mouth word, and the proprietary vocabulary mining module 13 judges that there is no new positive or negative word-of-mouth word, and then uses the vocabulary accumulation module 14 to import the accumulated positive and negative word-of-mouth words and narrative sentences into a common vocabulary database. In 17th, it is used as a general vocabulary data, and after being explored through different fields of test data, it will be able to make the general vocabulary database 17 more complete and later on the network. When measuring the product evaluation of an electronic article, the general commodity evaluation term will be used to screen the public opinion on the specific product more quickly and accurately.

Next, the present invention is actually described in the article of Meiqiangsheng, the brand and the milk powder industry, and the implementation steps are as follows:

1. First, about 9000 articles were selected by Meiqiangsheng, and about 500 articles containing word-of-mouth were selected, and about 300 positive and negative word-of-mouth words were sorted out. Then, about 200 narratives were taken out. After training, a general prediction model was obtained. .

2. A total of 30,000 articles (test data) from the milk powder industry were obtained, and the vocabulary of the milk powder industry was discovered. Therefore, the general prediction model was used for the first prediction, and about 100 new words were obtained.

3. After re-acquiring the word-of-mouth narrative, in the original 500 articles, after adding 100 new words, get 400 positive and negative word-of-mouth words (the existing 300 positive and negative word-of-mouth words plus 100 new words) ), and using the 400 positive and negative word-of-mouth terms, about 300 narratives were obtained in the original 500 articles, and after training, an updated prediction model was obtained.

4. Using the new forecasting model, a second forecast of 30,000 data in the milk powder industry will result in about 80 new words.

5. The original 500 articles, after adding 80 new words, get 480 positive and negative word-of-mouth words, using 480 positive and negative word-of-mouth words, get about 350 narrations in the original 500 articles, after training, get updated Forecast model.

6. Repeat steps 2 and 3, and after the fourth test, a total of 200 new words (100+80+20). After that, there are no new words, and the excavation process ends.

As can be seen from the above implementation steps, the present invention uses a generic model for articles in new fields. Predicting (exploration), the predicted results can be used to discover new vocabulary, but the vocabulary that is discovered cannot immediately become a new domain vocabulary, and it needs to be a candidate vocabulary after additional Rule screening, and can be added to artificial auxiliary judgment; The increase of vocabulary in the new field will increase the number of word-of-mouth narration sentences. Therefore, the technique of using word-of-mouth vocabulary will be able to obtain more narrative sentences as training materials. In addition, the positive and negative word-of-mouth words in step 3 above may be used. If there is an increase in the narrative sentence, train the new model, and repeat the procedures in steps 1 to 3. Each time you repeat, you can add the domain dictionary until there is no new one, but if there is no new narrative, the excavation program will Termination, qualified vocabulary can be added to the shared dictionary. After the implementation of the article on the brand and the milk powder industry, the initial accuracy is over 95%, which is better than the methods currently used in the market.

The advantages of the product evaluation word excavation method and system thereof for electronic articles provided by the present invention are compared with other conventional technologies, and the advantages thereof are as follows:

1. The invention can gradually establish a sustainable updated general vocabulary database through the discovery of commodity evaluation words, so as to be able to be used across fields, and after repeated tests of articles in different fields, it is possible to continuously add new universals. The terminology will enable continuous growth and learning so that the general vocabulary database can be used in a variety of different fields or industries.

2. With the general vocabulary database of the present invention, when a certain field is to be screened for the evaluation of a specific commodity on the Internet, it is not necessary to re-collect the word-of-mouth word, but can be directly used by the general vocabulary database or After first exploring the article in this industry, it can be put into use directly in this new field.

3. The invention establishes a general predictive model, and still needs human cooperation to operate the computer standard Note, but the program after the establishment of the general predictive model is simulated by the system itself, so the cost of human input can be minimized.

4. The larger the test data invested by the present invention and the larger the amount of data, the more accurate the results will be when it is finally used in the industry.

The features and spirit of the present invention will be more apparent from the detailed description of the preferred embodiments. On the contrary, the intention is to cover various modifications and equivalents within the scope of the invention as claimed.

Claims (6)

A method for excavating a product evaluation word for an electronic article, the steps of which are: first selecting a number of articles containing word of mouth with a plurality of general test data, and taking out a number of positive and negative word-of-mouth words and a plurality of word-of-mouth narration sentences, Then based on the positive and negative word-of-mouth and word-of-mouth narrative sentences, set up and establish a general forecasting model; then, based on the general forecasting model, predict the test data of several industrial fields, and judge and discover new positive and negative effects for the industrial field. Words of word of mouth; then accumulate the existing words of positive and negative word of mouth and the new words of positive and negative word of mouth, then use the accumulated positive and negative word of mouth to extract several new word-of-mouth narratives in several articles containing word-of-mouth, and then According to the accumulated positive and negative word-of-mouth words and the new word-of-mouth narrative sentence, the rules for labeling the characters are set to establish a new prediction model; and finally, the test data of several industrial fields are predicted again according to the new prediction model. And again judge and discover new positive and negative word-of-mouth words and new word-of-mouth narrative sentences, and then repeat new prediction model establishment. After the new words and new positive and negative word of mouth reputation narrative sentence explore, to discover and to no new positive and negative word of mouth words, the end of the excavation process. The scope of the patent application of paragraph 1 for the product of the evaluation terms of electronic article excavation method, wherein after the excavation process, then all of the positive and negative accumulated introducing a common word vocabulary word repository to serve as the common The vocabulary used. The method for excavating a commodity evaluation word for an electronic article as described in claim 1 of the patent application, wherein the article is a web article. A product evaluation word excavation system for an electronic article, comprising: a test data input module for inputting test data, the test data is at least including Universal test data and several industry test data; a general predictive model building module, comprising: a tablet article selection module, connected with the test data input module, the word-of-mouth article selection module is used by A number of articles containing word-of-mouth have been selected from several general test materials; a word-for-word selection module is connected to the word-of-mouth article selection module, which is capable of sorting out several positive and negative word-of-mouths in several articles containing word-of-mouth. a narration sentence extraction module, which is connected with the word-of-mouth article selection module and the word-of-mouth word selection module, and uses a positive and negative word-of-mouth word to extract a plurality of word-of-mouth narration sentences in a number of articles containing word-of-mouth; The general predictive model training module is connected with the word-of-mouth word selection module and the narrative sentence capture module, and is configured to set a rule for labeling text according to the positive and negative word-of-mouth word and the word-of-mouth sentence to establish a general predictive model; a proprietary vocabulary mining module coupled to the test data input module and the universal predictive model training module, capable of being based on a general predictive model Forecasting a number of industrial field test data, and judging and extracting new words used in the industrial field; a vocabulary accumulation module is associated with the proprietary vocabulary mining module and the word-of-mouth word selection module The connection is used to accumulate the existing positive and negative word-of-mouth words with the new positive and negative word-of-mouth words; a narrative sentence retrieval module is connected with the word-of-mouth article selection module and the vocabulary accumulation module, and the cumulative use is used. Positive and negative word-of-mouth uses a number of new word-of-mouth narratives in several articles containing word-of-mouth; a predictive model training module, and the vocabulary accumulation module, narrative sentence capture module and the proprietary The vocabulary mining module is connected to set a rule for labeling words based on the accumulated positive and negative word-of-mouth words and a new word-of-mouth narration sentence to establish a new prediction model, and the proprietary vocabulary mining module can According to the new forecasting model, several industrial field test data are predicted, and the new positive and negative word-of-mouth terms used in the industrial field are repeatedly judged and taken out. The product evaluation word excavation system for an electronic article described in claim 4 , further comprising a general vocabulary database connected to the vocabulary accumulation module, wherein the common vocabulary database has a common When the proprietary vocabulary mining module judges that there is no new positive or negative word-of-mouth word, the accumulated positive and negative word-of-mouth words are imported into the general vocabulary database through the vocabulary accumulation module. The product evaluation word excavation system for an electronic article as described in claim 4 of the patent application, wherein the article is a web article.