CN111639184A - Detection system for tendency inconsistency of scores and comment contents - Google Patents
Detection system for tendency inconsistency of scores and comment contents Download PDFInfo
- Publication number
- CN111639184A CN111639184A CN202010485096.0A CN202010485096A CN111639184A CN 111639184 A CN111639184 A CN 111639184A CN 202010485096 A CN202010485096 A CN 202010485096A CN 111639184 A CN111639184 A CN 111639184A
- Authority
- CN
- China
- Prior art keywords
- comment
- online
- user
- score
- training
- 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.)
- Pending
Links
- 238000001514 detection method Methods 0.000 title description 9
- 230000002996 emotional effect Effects 0.000 claims abstract description 88
- 230000008451 emotion Effects 0.000 claims abstract description 28
- 238000012545 processing Methods 0.000 claims abstract description 23
- 239000013598 vector Substances 0.000 claims abstract description 15
- 238000012549 training Methods 0.000 claims description 81
- 238000012552 review Methods 0.000 claims description 19
- 239000002131 composite material Substances 0.000 claims description 11
- 230000004927 fusion Effects 0.000 claims description 4
- 238000000034 method Methods 0.000 description 22
- 230000000694 effects Effects 0.000 description 4
- 238000011156 evaluation Methods 0.000 description 4
- 230000011218 segmentation Effects 0.000 description 3
- 238000004364 calculation method Methods 0.000 description 2
- 238000004590 computer program Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000007935 neutral effect Effects 0.000 description 2
- 238000012790 confirmation Methods 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 238000013441 quality evaluation Methods 0.000 description 1
- 238000013077 scoring method Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/35—Clustering; Classification
- G06F16/355—Class or cluster creation or modification
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/33—Querying
- G06F16/3331—Query processing
- G06F16/334—Query execution
- G06F16/3344—Query execution using natural language analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/20—Natural language analysis
- G06F40/279—Recognition of textual entities
- G06F40/284—Lexical analysis, e.g. tokenisation or collocates
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/20—Natural language analysis
- G06F40/279—Recognition of textual entities
- G06F40/289—Phrasal analysis, e.g. finite state techniques or chunking
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
- G06Q10/0639—Performance analysis of employees; Performance analysis of enterprise or organisation operations
- G06Q10/06393—Score-carding, benchmarking or key performance indicator [KPI] analysis
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Business, Economics & Management (AREA)
- Human Resources & Organizations (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Computational Linguistics (AREA)
- Artificial Intelligence (AREA)
- Educational Administration (AREA)
- Strategic Management (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Data Mining & Analysis (AREA)
- Development Economics (AREA)
- Databases & Information Systems (AREA)
- Economics (AREA)
- Entrepreneurship & Innovation (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Game Theory and Decision Science (AREA)
- Marketing (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention provides a system for detecting the trend inconsistency of scores and comment contents, which is used for detecting an online comment text with inconsistent scores of users from an online comment text of the comment object by the users and the scores of the users, and is characterized by comprising the following steps: an online comment acquisition unit which acquires an online comment text of a comment object and a user score as current comment information; the vectorization processing part is used for vectorizing the online comment text to form a corresponding comment text vector; an emotional tendency score predicting part which inputs the online comment text into a preset emotional tendency score predicting model so as to predict the emotional tendency score of the user emotion expressing the online comment text, wherein the value range of the emotional tendency score is consistent with the value range of the user score; and the inconsistent comment judgment and output part is used for sequentially judging the online comment texts with the difference values of the user scores and the emotion scores larger than a first threshold value as inconsistent comments and outputting all the inconsistent comments in the current comment information.
Description
Technical Field
The invention belongs to the field of text recognition, and relates to a system for detecting the trend inconsistency of scoring and commenting contents.
Background
For online transaction, taking a commodity as an example, in order to facilitate users, the commodity is commented in an online commenting manner, so that the users after the commodity is helped to know the quality of the commodity and other objects according to evaluation, and further, in order to facilitate the users to carry out scoring while publishing online comments, the scoring of all the users is combined to form comprehensive scoring of the commodity (store), and the comprehensive scoring is very important for buyers and sellers: from the perspective of users, the comprehensive scores determine a first impression of the commodities, and many users rely on the comprehensive score ranking to determine whether to browse and purchase the commodities; from the seller's perspective, the level of the composite score directly determines the sales volume of the goods.
However, users have strong randomness and subjectivity in scoring, so that the composite score finally obtained based on the scores cannot generally very accurately feed back the quality of the comment object. In a realistic review scenario, the case of "score-review" inconsistencies is very common, i.e., high score, with a negative trend for reviews (high poor score) or low score but positive trend for reviews (low good score).
In comparison, the comment content can reflect the real experience of the user more accurately and objectively. However, it takes a lot of time for the user to view the comment content, and generally, the user still depends on the composite score to make a decision on the comment object such as a store or a commodity. At this time, originally, a user can conveniently confirm the quality of the comprehensive score of the comment object simply according to the score, and the decision of a subsequent user is influenced due to inaccurate comprehensive score caused by the random score of the previous user.
Disclosure of Invention
In order to solve the problems, the invention provides a detection system which can automatically judge the consistency of online comments of users and user scores, thereby helping merchants and users to objectively confirm the inconsistency between the scores of good and bad comment objects and the tendency of comment contents, and the invention adopts the following technical scheme:
the invention provides a system for detecting the trend inconsistency of scores and comment contents, which is used for detecting an online comment text with inconsistent scores of users from an online comment text of the comment object by the users and the scores of the users, and is characterized by comprising the following steps: an online comment acquisition unit which acquires an online comment text of a comment object and a user score as current comment information; the vectorization processing part sequentially carries out vectorization processing on each online comment text to form a corresponding comment text vector; the emotion tendency score prediction part is used for storing a preset emotion tendency score prediction model and sequentially inputting the online comment texts into the emotion tendency score prediction model so as to predict emotion tendency scores representing the user emotions of the online comment texts, and the value range of the emotion tendency scores is consistent with the value range of the user scores; the inconsistent comment judgment and output part is used for sequentially judging the online comment texts with the user scores and the emotion scores of which the difference values are larger than a preset first threshold value as inconsistent comments and outputting all the inconsistent comments in the current comment information, wherein the emotional tendency score prediction model is obtained by training through the following steps: step S1, acquiring training online comment texts containing a plurality of comment objects and a training set of training user scores; step S2, sequentially carrying out vectorization processing on each piece of online comment text for training to form a corresponding comment text vector for training; step S3, clustering all the comment text vectors for training and forming a plurality of comment clusters corresponding to different subjects; step S4, matching and fusing the training online comment texts through a preset emotional mode word library to obtain the emotional tendency score of each training online comment text; step S5, taking the training online comment text with the difference between the emotional tendency score and the training user score lower than a preset second threshold value as a training sample, and taking the emotional tendency score as a sample label of the training sample; and step S6, finishing the training of the emotional tendency score prediction model through the training samples and the sample labels.
The system for detecting the trend inconsistency of the scores and the comment contents, provided by the invention, can also have the technical characteristics that the matching fusion operation is as follows: carrying out sentence segmentation on the online comment text for training to obtain a comment sentence; segmenting the comment sentences to obtain candidate words; matching the candidate words with each emotional word in an emotional mode word bank in sequence and scoring the emotional score of the comment statement based on the emotional score of the matched emotional word; and carrying out weighted fusion on the emotion scores according to the sentence lengths of the comment sentences in each online comment text for training so as to obtain the content score of the online comment text for training.
The system for detecting the trend inconsistency of the scoring and the comment content, provided by the invention, can also have the technical characteristics that: the online comment storage part stores a plurality of comment objects, online comment texts corresponding to the comment objects and user scores, the picture storage part stores comment object browsing pictures and comment viewing pictures, the input display part displays the comment object browsing pictures and displays the stored comment objects to enable a user to select a to-be-viewed comment object to be viewed, once the user selects the to-be-viewed comment object, the comment judgment control part controls the online comment acquisition part to acquire the corresponding online comment text and user scores from the online comment storage part according to the selected to-be-viewed comment object as current comment information, controls the inconsistent comment judgment output part to output inconsistent comments in the current comment information to the input display part, and once the inconsistent comments are received, the input display part displays a comment viewing picture, displays the current comment information and simultaneously displays the emotional tendency scores of the inconsistent comments for the user to view.
The system for detecting the trend inconsistency of the scoring and the comment content, provided by the invention, can also have the technical characteristics that: the online comment storage unit stores a plurality of comment objects, online comment texts and user scores corresponding to the comment objects, the screen storage unit stores a comment input screen and a comment object browsing screen, the input display unit displays the comment input screen to allow the user to input the online comment texts and the user scores of the comment objects as comment input information, the online comment storage unit stores the comment input information as new online comment texts and the user scores corresponding to the comment objects once the user confirms the input of the comment input information, and the online comment storage unit controls the online comment acquisition unit to acquire all the online comment texts corresponding to the newly added comment objects and to store the new online comment texts and the user scores once the online comment storage unit stores the new online comment texts and the user scores And the user scores are used as current comment information, the inconsistent comment judgment output part is controlled to output inconsistent comments in the current comment information to the inconsistent proportion generation part, once the inconsistent comments are received, the inconsistent proportion generation part generates corresponding inconsistent proportions based on the quantity proportion occupied by the inconsistent comments in all online comment texts of the current comment information, the comprehensive score generation part carries out average value calculation on all the user scores in the current comment information and generates comprehensive scores of corresponding comment objects, and the input display part displays a comment object browsing picture and displays the stored comment objects and the corresponding comprehensive scores and the inconsistent proportions to enable the user to confirm the quality of the comment objects.
Action and Effect of the invention
According to the system for detecting the tendency inconsistency of the scores and the comment contents, the online comment acquisition part, the vectorization processing part and the emotional tendency score prediction part are arranged, so that all online comment texts and user scores of comment objects can be acquired, and the quantitative online comment texts are predicted through the emotional tendency score prediction model stored in the emotional tendency score prediction part, so that emotional tendency scores capable of accurately reflecting the actual emotions of the online comment texts are predicted, and the comment tendency of the online comment texts is objectively obtained. And the inconsistent comment judgment output part is arranged, so that inconsistent comments, which do not accord with the comments, in the online comment text can be screened out, and a more objective and real information reference basis is provided for merchants and users.
In addition, when the emotional tendency score prediction model is trained, the online comment texts are divided into different semantic subjects through clustering, so that the emotional deviation caused by the semantic subjects is reduced, and the emotional tendency score prediction model obtained through final training can accurately complete prediction. Furthermore, because the samples and the sample labels for training are constructed in the steps S4 and S5, the emotional tendency score prediction model does not need to be manually labeled with consistency samples, and has higher model usability and generalization performance.
Drawings
FIG. 1 is a block diagram of a trend inconsistency detection system for scoring and reviewing content in an embodiment of the present invention;
FIG. 2 is a flow chart of the training process of the emotional tendency score prediction model in the embodiment of the invention;
FIG. 3 is a flow chart of a review viewing process in an embodiment of the present invention; and
fig. 4 is a flowchart of a comment input process in the embodiment of the present invention.
Detailed Description
In order to make the technical means, creation features, achievement purposes and effects of the invention easy to understand, the trend inconsistency detection system for scoring and commenting contents of the invention is specifically described below with reference to the embodiments and the drawings.
< example >
In this embodiment, taking the comment object as an online shopping commodity as an example, the system for detecting the trend inconsistency of the scores and the comment contents is a part of an online shopping platform, and is used for detecting the inconsistency of the online comments of the online shopping commodity by the user and the scores of the user.
Fig. 1 is a block diagram of the configuration of a tendency inconsistency detection system of scoring and commenting contents in the embodiment of the present invention.
As shown in fig. 1, the tendency inconsistency detecting system 100 for scores and comment contents includes an online comment storage unit 101, a comment determination control unit 102, an online comment acquisition unit 103, a vectorization processing unit 104, an emotional tendency score prediction unit 105, an inconsistent comment determination output unit 106, a composite score generation unit 107, an inconsistent proportion generation unit 108, a screen storage unit 109, an input display unit 110, and a system control unit 111 for controlling the above units.
The system control unit 111 stores therein a computer program for controlling the operations of the respective components of the detection system 100.
The online comment storage unit 101 stores a plurality of comment objects, online comment texts and user scores, which correspond to the comment objects, respectively.
In the present embodiment, online comment texts and user scores of items uploaded by all users in the past are stored in the online comment storage unit 101. In the online comment storage unit 101, online comment texts correspond to user scores one by one, and each comment object corresponds to a plurality of sets of online comment texts and user scores.
The comment determination control unit 102 controls the operation of components involved in the comment inconsistency determination process, and more specifically, controls the operation of components of the online comment acquisition unit 103, the vectorization processing unit 104, the emotional tendency score prediction unit 105, and the inconsistent comment determination output unit 106.
The online comment acquisition unit 103 is configured to acquire an online comment text and a user score as current comment information.
In this embodiment, the online comment acquisition unit 103 acquires current comment information under the control of the comment determination control unit 102, specifically:
(1) once the new online comment text and the user score are stored in the online comment storage unit 101, that is, the user uploads a new online comment, the comment determination control unit 102 controls the online comment acquisition unit 103 to acquire all online comment texts and user scores of comment objects corresponding to the new online comment text from the online comment storage unit 101 as current comment information.
(2) Once the user selects a comment object to be viewed, for which a comment needs to be viewed, the comment determination control unit 102 controls the online comment acquisition unit 103 to acquire all online comment texts and user scores corresponding to the comment object to be viewed as current comment information from the online comment storage unit 101.
In another embodiment of the present invention, the online comment acquisition unit 103 may acquire an online comment, which is input by a user or input by another system, as the current comment information.
In the present embodiment, when the online comment acquisition unit 103 acquires the current comment information, the comment determination control unit 102 sequentially controls the vectorization processing unit 104, the emotional tendency score prediction unit 105, and the inconsistent comment determination output unit 106 to perform inconsistency determination on the current comment information.
The vectorization processing portion 104 is configured to perform vectorization processing on each piece of online comment text acquired by the online comment acquisition portion 101, and form a corresponding comment text vector.
In the present embodiment, the vectorization processing of the vectorization processing portion is a conventional vectorization method, for example, vectorizing the online comment text by using a word2vec model and an LSTM model.
The emotional tendency score prediction unit 105 is configured to predict an emotional tendency score for each online comment text.
In this embodiment, the emotional tendency score prediction unit 105 stores one emotional tendency score prediction model, and the emotional tendency score prediction unit 105 inputs the online comment text into the emotional tendency score prediction model to complete the prediction of the emotional tendency score.
FIG. 2 is a flowchart of the training process of the emotional tendency score prediction model in the embodiment of the invention.
As shown in FIG. 2, the emotional tendency score prediction model is obtained by training the following processes:
step S1-1, obtaining training online comment texts for training containing a plurality of comment objects and a training set of user scores for training.
In this embodiment, the training set is obtained by the online comment acquisition unit 103 from the online comment storage unit 101, and is the online comment text and the user score stored in the online comment storage unit 101. When acquiring the training set, the online comment acquisition unit 103 acquires all online comment texts and user scores of a plurality of comment subjects, and the number of the comment subjects may be set in advance.
And step S1-2, sequentially carrying out vectorization processing on each piece of online comment text for training to form a corresponding comment text vector for training.
In the present embodiment, the on-line comment text for training in step S1-2 completes the vectorization processing by the vectorization processing section 104.
And step S1-3, clustering all the comment text vectors for training and forming a plurality of comment clusters corresponding to different semantic subjects.
In this embodiment, because it is considered that the number of training online comment texts obtained during training is generally large, a MinBatch KmeanS 1-clustering method is adopted for clustering.
After clustering is completed, according to the similarity of comment texts corresponding to the comment text vectors in each clustering cluster in the aspect of semantic expression, a corresponding comment mode (semantic topic) is formed, namely, the comment text vectors of the same semantic topic and the corresponding online comment texts form the comment clusters of the corresponding semantic topic.
And step S1-4, matching and fusing the training online comment texts through a preset emotional mode word library to obtain the emotional tendency score of each training online comment text.
In this embodiment, each comment cluster clustered in step S1-3 represents a different semantic topic of the comment content, and each comment of each semantic topic is subjected to sentiment scoring in step S1-4. The emotion scoring method specifically comprises the following steps:
firstly, each comment is subjected to sentence segmentation (if only one sentence is available, the sentence is not segmented) to obtain comment sentences, and each sentence of comment sentences is subjected to word segmentation to obtain a plurality of corresponding candidate words;
and then, matching various types of emotional words in the emotional mode word library with each candidate word by utilizing an emotional mode word library (positive, negative, uncertain, negative, positive, strong mode and weak mode) and calculating a matching rate. For example, a sentence has ten words, three of which can correspond to the word stock of the forward emotion dictionary, and the matching rate of the sentence under the forward emotion index is 0.3, and other indexes are the same;
and then, carrying out weighted fusion on all sentence scores in a single comment according to the sentence length of the comment sentence to obtain the content score of the single online comment text. The score value range is consistent with the user score value range, and the score is 1-5 points by taking the hotel industry score rule as an example.
The emotional modal lexicon comprises the following components: positive (+1), neutral or neutral (0), negative (-1). The degree word reconciles the scores of the emotional words. For example, if the positive word is preceded by the negative word, then x (-1), positive x 2, strong mode x 1.5, weak mode x 0.5.
And step S1-5, taking the training online comment text with the emotional tendency score and the training user score difference lower than a preset second threshold value as a training sample, and taking the emotional tendency score as a sample label of the training sample.
For each cluster of comments, comments with small gaps between emotional tendency scores and real scores (i.e., training user scores) are selected as training samples of "score-comment tendency consistency".
And step S1-6, finishing the training of the emotional tendency score prediction model through the training samples and the sample labels.
In step S1-6 of this embodiment, the emotional tendency score prediction model includes a plurality of predictor models, and the comment clusters of different semantic topics have different predictor models for training. When training is performed, the input of each predictor model is the comment text feature with consistent score-comment content emotional tendency (namely, the training sample obtained in step S1-5), and the training label is the score corresponding to the consistent comment text. For example, in the portable platform, each comment corresponds to one user score (1-5), and the trained model is a 5-class model.
The emotional tendency score prediction model is a conventional neural network model or a classifier, and the adopted training method is also a conventional training method, which is not described herein again.
Through the training process, the training of the emotional tendency score prediction model is completed, and the emotional tendency score prediction model can be stored in the emotional tendency score prediction unit 105 to predict the emotional tendency score. Meanwhile, since the emotion tendency score prediction model of the present embodiment includes a plurality of prediction submodels corresponding to different semantic topics, before predicting the online comment text, the emotion tendency score prediction unit 105 calculates a comment cluster belonging to which semantic topic the comment text belongs according to the similarity index (the distance between the comment text and each cluster center), and then performs score prediction using the prediction submodel corresponding to the semantic topic.
In addition, in the embodiment, since the training sample and the sample label are automatically constructed through the steps S1-4 and S1-5 in the training process, the training process can be programmed into a corresponding computer program in advance, and the training is automatically performed in the detection system 100 at regular intervals.
The inconsistent comment judgment output unit 106 is configured to judge whether the online comment text is an inconsistent comment or not and output the judged inconsistent comment.
In this embodiment, the inconsistent comment judgment output unit 106 sequentially judges whether the difference between the user score and the emotion score of each online comment text is greater than a preset first threshold, and judges the online comment text as an inconsistent comment when the difference is greater than the first threshold. The first threshold value can be set correspondingly according to actual conditions.
Once the inconsistent comment judgment output unit 106 finishes the judgment of the online comment text to be judged in the current comment information, the comment judgment control unit 102 controls the comment judgment output unit 106 to output all the online comment texts judged to be inconsistent comments. Specifically, the method comprises the following steps:
(1) when the comment determination control unit 102 controls the online comment acquisition unit 103 to acquire all online comment texts and user scores of comment targets corresponding to the newly added online comment texts from the online comment storage unit 101 as current comment information, the control comment determination output unit 106 outputs the inconsistent comments to the inconsistent proportion generation unit 108.
(2) When the comment determination control portion 102 controls the online comment acquisition portion 103 to acquire all online comment texts corresponding to the objects of comments to be viewed and the user scores from the online comment storage portion 101 as current comment information, the comment determination output portion 106 is controlled to output inconsistent comments to the input display portion 110.
The comprehensive score generating unit 107 is configured to, when the online comment acquiring unit 103 acquires all online comment texts and user scores of comment objects corresponding to the newly added online comment texts from the online comment storage unit 101 as current comment information, perform average calculation on all user scores in the current comment information, and generate a comprehensive score corresponding to the comment objects.
In this embodiment, the comprehensive score is an average score of user scores uploaded by all users of one comment object, and is used for enabling the user to visually confirm the quality evaluation of the comment object by all other users.
The inconsistent proportion generating part 108 is used for generating corresponding inconsistent proportions based on the quantity proportion occupied by the inconsistent comments in all online comment texts of the current comment information when the inconsistent comments output by the comment judgment output part 106 are received.
In this embodiment, the inconsistent proportion is a proportion occupied by inconsistent comments in all online comments of one comment object, and is used for enabling a user to visually confirm how many user scores in the comprehensive scores are scores inconsistent with the online comment text.
The screen storage unit 109 stores a comment subject browsing screen, a comment viewing screen, and a comment input screen.
The comment object browsing picture is used for displaying when the user selects a browsing comment object, and displaying all comment objects stored in the online comment storage part 101, corresponding comprehensive scores and inconsistent proportions in the picture, so that the user can select the comment object to be viewed as the comment object to be viewed.
In this embodiment, the review object browsing screen is a product browsing page (e.g., a web shopping page, etc.), the product names of the products (i.e., the review objects) are displayed, and the comprehensive scores and the inconsistent proportions of the products are simultaneously displayed beside each product name, so that a user can determine whether the products need to be selected for viewing according to the comprehensive scores and the inconsistent proportions.
The comment viewing picture is used for displaying when the user selects the comment object to be viewed, and displaying all online comment texts, user scores and emotional tendency scores corresponding to the comment object to be viewed in the picture.
In the embodiment, all online comment texts of the current commodity are displayed in the comment viewing picture, a user score corresponding to each online comment text is displayed beside each online comment text, and meanwhile, an emotional tendency score is also displayed beside each online comment text judged to be inconsistent with each comment, so that the user can visually view the evaluation degree of each comment on the quality of the commodity.
The comment input picture is used for displaying when a user selects a comment object and selects a comment input operation, and allowing the user to input a new online comment text and a user score.
In this embodiment, the comment input picture includes two input boxes, which are respectively used for allowing the user to comment online and for allowing the user to score. Once the user confirms the input of the comment (for example, clicks a confirmation button), the input display unit 110 uses the online comment text and the user score as comment input information, and the online comment storage unit 101 stores the comment input information as a newly added online comment text and user score in association with the corresponding comment object.
In another aspect of the present invention, the comment viewing screen and the comment input screen may be displayed simultaneously as two parts of the same screen.
The input display unit 110 is used for displaying the above-mentioned screens, so that the user can complete the corresponding human-computer interaction through the screens.
FIG. 3 is a flow chart of a review viewing process in an embodiment of the invention.
As shown in fig. 3, when the user selects to browse the comment object and view a specific online comment of the comment object, the following steps are started:
step S2-1, the input display section 110 displays a comment object browsing screen to let the user select a comment object to be viewed, and once the user confirms the selection, the selected comment object is taken as a comment object to be viewed and the process proceeds to step S2-2;
step S2-2, the comment determination control part 102 controls the online comment acquisition part 103 to acquire all online comment texts and user scores corresponding to the comment objects to be viewed from the online comment storage part 101 as current comment information, and then the step S2-3 is performed;
step S2-3, the comment determination control unit 102 controls the vectorization processing unit 104 to perform vectorization processing on each piece of online comment text acquired in step S2-2, and form a corresponding comment text vector, and then proceeds to step S2-4;
step S2-4, the comment judgment control part 102 controls the emotional tendency score prediction part 105 to input each online comment text into the emotional tendency score prediction model in sequence, so that the emotional tendency score of the online comment text is predicted, and then the step S2-5 is carried out;
step S2-5, the comment determination control section 102 controls the inconsistent comment determination output section 106 to determine the inconsistency of each online comment text based on the user score acquired in step S2-2 and the emotional tendency score predicted in step S2-4, and outputs the determined inconsistent comment to the input display section 110, and then proceeds to step S2-6;
in step S2-6, the input display unit 110 displays a comment viewing screen, and displays the online comment text and the user score acquired in step S2-2, and the emotional tendency score corresponding to each inconsistent comment output in step S2-5 on the screen, and then enters an end state.
Fig. 4 is a flow chart of a score input process in an embodiment of the invention.
As shown in fig. 4, once the user selects the comment input operation, the following steps are started:
step S3-1, the input display unit 110 displays a comment input screen for the user to input the online comment text and the user score of the comment target, and upon the user confirming the input, the input online comment text and the user score are taken as comment input information and the process proceeds to step S3-2;
step S3-2, the online comment storage part 101 takes the comment input information as the newly added online comment text and the corresponding comment object for corresponding storage, and then the step S3-3 is entered;
step S3-3, the comment determination control part 102 controls the online comment acquisition part 103 to acquire all online comment texts and user scores of comment objects corresponding to the online comment texts newly added in step S3-2 from the online comment storage part 101 as current comment information, and then the step S3-4 is carried out;
step S3-4, the comment judgment control part 102 controls the vectorization processing part 104 to carry out vectorization processing on each piece of online comment text acquired in step S3-3, form a corresponding comment text vector, and then go to step S3-5;
step S3-5, the comment judgment control part 102 controls the emotional tendency score prediction part 105 to input each online comment text into the emotional tendency score prediction model in sequence, so as to predict the emotional tendency score of the online comment text, and then the step S3-6 is carried out;
step S3-6, the comment determination control section 102 controls the inconsistent comment determination output section 106 to determine the inconsistency of each online comment text based on the user score acquired in step S3-3 and the emotional tendency score predicted in step S3-5, and outputs the determined inconsistent comment to the input display section 110, and then proceeds to step S3-7;
step S3-7, the inconsistent proportion generating part 108 generates corresponding inconsistent proportions based on the quantity proportion occupied by the inconsistent comments output in the step S3-6 in all online comment texts of the current comment information, and then the step S3-8 is carried out;
in step S3-8, the composite score generation unit 107 calculates the average value of all the user scores acquired in step S3-3, generates a composite score corresponding to the review object, and enters an end state.
In this embodiment, after the comment viewing process enters the end state, the user may also enter the comment generating process by selecting a comment input operation. In addition, in the comment object browsing screen displayed by the input display unit 110 when the user performs the comment viewing process, the composite score and the disagreement ratio generated by each comment object in the previous comment generation process are displayed.
Examples effects and effects
According to the system for detecting the tendency inconsistency between the score and the comment content provided by the embodiment, the online comment acquisition unit, the vectorization processing unit and the emotional tendency score prediction unit are provided, so that all online comment texts and user scores of comment objects can be acquired, and the quantitative online comment texts are predicted by the emotional tendency score prediction model stored in the emotional tendency score prediction unit, so that emotional tendency scores capable of accurately reflecting the actual emotion of the online comment texts are predicted, and the comment tendency of the online comment texts is more objectively obtained. And the inconsistent comment judgment output part is arranged, so that inconsistent comments, which do not accord with the comments, in the online comment text can be screened out, and a more objective and real information reference basis is provided for merchants and users.
In addition, when the emotional tendency score prediction model is trained, the online comment texts are divided into different semantic subjects through clustering, so that the emotional deviation caused by comment semantics is reduced, and the emotional tendency score prediction model obtained through final training can accurately complete prediction. Furthermore, because the samples and the sample labels for training are constructed in the steps S4 and S5, the emotional tendency score prediction model does not need to be manually labeled with consistency samples, and has higher model usability and generalization performance.
In addition, because the emotion tendency scoring prediction model comprises a plurality of prediction submodels corresponding to different semantic topics, the problems that the emotion tendencies of different topics existing in online comments have deviation, and scoring standards of clients focusing on different topics are different when the clients write comment contents can be solved. For example, hotel scenarios, customers are more sensitive to conditions in terms of hotel hygiene, traffic, etc., than hotel conditions (television size, shade color, etc.), which are more likely to be badly rated. For example, for a hotel, the sanitation evaluation of the client is good overall, the bad evaluation content relates to service attitudes more, at this time, the sanitation score tends to be high, the service attitudes tend to be low, and if the theme is divided, the consistency judgment is more accurate.
The above-described embodiments are merely illustrative of specific embodiments of the present invention, and the present invention is not limited to the description of the above-described embodiments.
In the above embodiment, the comment objects are respective commodities of an online shopping platform. In other schemes of the invention, the comment objects can also be various comment objects such as hotels, shops, music and the like, and the comment inconsistency detection is carried out on various comment objects.
Claims (5)
1. A system for detecting a tendency inconsistency of a review with respect to a review content, which is used for detecting an online review text of a review object by a user and an online review text of the review object by the user, the online review text being inconsistent with a user score, the system comprising:
an online comment acquisition unit configured to acquire the online comment text of the comment object and the user score as current comment information;
the vectorization processing part sequentially carries out vectorization processing on each online comment text to form a corresponding comment text vector;
the emotion tendency score predicting part is used for storing a preset emotion tendency score predicting model, sequentially inputting the online comment texts into the emotion tendency score predicting model so as to predict emotion tendency scores representing the user emotions of the online comment texts, wherein the value range of the emotion tendency scores is consistent with the value range of the user scores;
an inconsistent comment judgment output unit which judges the online comment text in which the difference between the user score and the emotion score is larger than a preset first threshold as inconsistent comments and outputs all the inconsistent comments in the current comment information,
the emotional tendency score prediction model is obtained by training through the following steps:
step S1, acquiring training online comment texts for training containing a plurality of comment objects and a training set of user scores for training;
step S2, sequentially carrying out vectorization processing on each training comment text for online use to form a corresponding training comment text vector;
step S3, clustering all the training comment text vectors, and forming a plurality of comment clusters corresponding to different semantic subjects by the training comment text vectors and the corresponding training comment texts on line according to the clustering;
step S4, matching and fusing words in the online comment texts for training through a preset emotional mode word library to obtain emotional tendency scores of each online comment text for training;
step S5, taking the training online comment text with the emotional tendency score and the training user score difference lower than a preset second threshold value as a training sample, and taking the emotional tendency score as a sample label of the training sample;
and step S6, finishing the training of the emotional tendency score prediction model through the training sample and the sample label.
2. The system for detecting a trend inconsistency of scoring and reviewing content according to claim 1, wherein:
wherein the matching fusion operation is:
sentence dividing is carried out on the training online comment text to obtain a comment sentence;
segmenting the comment sentences to obtain candidate words;
matching the candidate words with all the emotional words in the emotional mode word bank in sequence and scoring the emotional scores of the comment sentences based on the matched emotional scores of the emotional words;
and weighting and fusing the emotion scores according to the sentence lengths of the comment sentences in each online comment text for training so as to obtain the content scores of the online comment text for training.
3. The system for detecting a trend inconsistency of scoring and reviewing content according to claim 1, further comprising:
an online comment storage unit, a comment determination control unit, a screen storage unit, and an input display unit,
wherein the online comment storage section stores a plurality of the comment objects and the online comment texts and the user scores corresponding to the comment objects,
the screen storage section stores a comment subject browsing screen and a comment viewing screen,
the input display part displays the comment object browsing screen and displays the stored comment object to allow the user to select a comment object to be viewed,
upon the user selecting the comment object to be viewed, the comment determination control section controls the online comment acquisition section to acquire the corresponding online comment text and the user score from the online comment storage section as the current comment information in accordance with the selected comment object to be viewed, and controls the inconsistent comment determination output section to output the inconsistent comment in the current comment information to the input display section,
upon receiving the inconsistent comments, the input display portion displays the comment viewing screen and displays the emotional tendency score of each of the inconsistent comments for the user to view while displaying the current comment information.
4. The system for detecting a trend inconsistency of scoring and reviewing content according to claim 1, further comprising:
an online comment storage unit, a comment determination control unit, an inconsistency ratio generation unit, a composite score generation unit, a screen storage unit, and an input display unit,
wherein the online comment storage section stores a plurality of the comment objects and the online comment texts and the user scores corresponding to the comment objects,
the screen storage section stores a comment input screen and a comment object browsing screen,
the input display section displays a comment input screen for the user to input the online comment text of the comment object and the user score as comment input information,
the online comment storing section stores the comment input information as a newly added online comment text and user comments in correspondence with the respective comment objects once the user confirms the input of the comment input information,
once the online comment storage unit stores the newly added online comment text and the user score, the comment determination control unit controls the online comment acquisition unit to acquire all the online comment texts and the user score corresponding to the newly added comment subjects as the current comment information, and controls the inconsistent comment determination output unit to output the inconsistent comments in the current comment information to the inconsistent proportion generation unit,
upon receiving the inconsistent comments, the inconsistent proportion generation section generates corresponding inconsistent proportions based on a proportion of the number of the inconsistent comments occupied in all of the online comment texts of the current comment information,
the composite score generation unit calculates an average value of all the user scores in the current comment information and generates a composite score corresponding to the comment object,
the input display unit displays the review object browsing screen and displays the stored review object and the corresponding composite score and the inconsistency rate to allow the user to confirm whether the review object is good or bad.
5. The system for detecting a trend inconsistency of scoring and reviewing content according to claim 1, wherein:
wherein the emotional tendency score prediction model comprises a plurality of predictor models corresponding to different semantic topics,
when the emotional tendency score prediction model is trained in the step S6, the corresponding predictor model is trained according to the semantic topic corresponding to the comment cluster corresponding to the training sample.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010485096.0A CN111639184A (en) | 2020-06-01 | 2020-06-01 | Detection system for tendency inconsistency of scores and comment contents |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010485096.0A CN111639184A (en) | 2020-06-01 | 2020-06-01 | Detection system for tendency inconsistency of scores and comment contents |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN111639184A true CN111639184A (en) | 2020-09-08 |
Family
ID=72332123
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010485096.0A Pending CN111639184A (en) | 2020-06-01 | 2020-06-01 | Detection system for tendency inconsistency of scores and comment contents |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111639184A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112785331A (en) * | 2021-01-07 | 2021-05-11 | 之江实验室 | Injection attack resistant robust recommendation method and system combining evaluation text |
| CN115169996A (en) * | 2022-09-06 | 2022-10-11 | 天津所托瑞安汽车科技有限公司 | Road risk determination method, apparatus and storage medium |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107180021A (en) * | 2016-03-09 | 2017-09-19 | 北京京东尚科信息技术有限公司 | A kind of data processing method, system and its server |
| US20180041458A1 (en) * | 2016-08-08 | 2018-02-08 | Flipboard, Inc. | Adaptive presentation of comments based on sentiment |
| CN108446813A (en) * | 2017-12-19 | 2018-08-24 | 清华大学 | A kind of method of electric business service quality overall merit |
| CN108573411A (en) * | 2018-04-17 | 2018-09-25 | 重庆理工大学 | Depth sentiment analysis and multi-source based on user comment recommend the mixing of view fusion to recommend method |
| CN109635291A (en) * | 2018-12-04 | 2019-04-16 | 重庆理工大学 | A kind of recommended method of fusion score information and item contents based on coorinated training |
| CN110489553A (en) * | 2019-07-26 | 2019-11-22 | 湖南大学 | A kind of sensibility classification method based on Multi-source Information Fusion |
-
2020
- 2020-06-01 CN CN202010485096.0A patent/CN111639184A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107180021A (en) * | 2016-03-09 | 2017-09-19 | 北京京东尚科信息技术有限公司 | A kind of data processing method, system and its server |
| US20180041458A1 (en) * | 2016-08-08 | 2018-02-08 | Flipboard, Inc. | Adaptive presentation of comments based on sentiment |
| CN108446813A (en) * | 2017-12-19 | 2018-08-24 | 清华大学 | A kind of method of electric business service quality overall merit |
| CN108573411A (en) * | 2018-04-17 | 2018-09-25 | 重庆理工大学 | Depth sentiment analysis and multi-source based on user comment recommend the mixing of view fusion to recommend method |
| CN109635291A (en) * | 2018-12-04 | 2019-04-16 | 重庆理工大学 | A kind of recommended method of fusion score information and item contents based on coorinated training |
| CN110489553A (en) * | 2019-07-26 | 2019-11-22 | 湖南大学 | A kind of sensibility classification method based on Multi-source Information Fusion |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112785331A (en) * | 2021-01-07 | 2021-05-11 | 之江实验室 | Injection attack resistant robust recommendation method and system combining evaluation text |
| CN115169996A (en) * | 2022-09-06 | 2022-10-11 | 天津所托瑞安汽车科技有限公司 | Road risk determination method, apparatus and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108959603B (en) | Personalized recommendation system and method based on deep neural network | |
| CN109034973B (en) | Commodity recommendation method, commodity recommendation device, commodity recommendation system and computer-readable storage medium | |
| CN107391493B (en) | Public opinion information extraction method and device, terminal equipment and storage medium | |
| CN107944911B (en) | Recommendation method of recommendation system based on text analysis | |
| KR101448228B1 (en) | Apparatus and Method for social data analysis | |
| CN108230009B (en) | User preference prediction method and device and electronic equipment | |
| Chen et al. | Exploring determinants of attraction and helpfulness of online product review: A consumer behaviour perspective | |
| CN111667337A (en) | Commodity evaluation ordering method and system | |
| CN109902229B (en) | Comment-based interpretable recommendation method | |
| US20220084102A1 (en) | Commodity recommendation method, server, shopping cart and shopping system | |
| CN111260437A (en) | Product recommendation method based on commodity aspect level emotion mining and fuzzy decision | |
| CN111460819B (en) | Personalized comment text recommendation system and recommendation method based on fine granularity emotion analysis | |
| CN111639184A (en) | Detection system for tendency inconsistency of scores and comment contents | |
| CN113946754A (en) | User portrait based rights and interests recommendation method, device, equipment and storage medium | |
| CN116894711A (en) | Commodity recommendation reason generation method and device and electronic equipment | |
| CN111966888A (en) | External data fused interpretable recommendation method and system based on aspect categories | |
| CN113989476A (en) | Object identification method and electronic equipment | |
| CN110851694A (en) | Personalized recommendation system based on user memory network and tree structure depth model | |
| CN112434173A (en) | Search content output method and device, computer equipment and readable storage medium | |
| CN111723302A (en) | Recommendation method based on collaborative dual-model deep representation learning | |
| CN110020195B (en) | Article recommendation method and device, storage medium and electronic equipment | |
| CN116703506A (en) | Multi-feature fusion-based E-commerce commodity recommendation method and system | |
| Seroussi | Utilising user texts to improve recommendations | |
| CN114519100A (en) | Catering data analysis method and device, electronic equipment and storage medium | |
| CN114358871A (en) | Commodity recommendation method and electronic equipment |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20200908 |
|
| RJ01 | Rejection of invention patent application after publication |