CN108446813A - A kind of method of electric business service quality overall merit - Google Patents
A kind of method of electric business service quality overall merit Download PDFInfo
- Publication number
- CN108446813A CN108446813A CN201711372805.9A CN201711372805A CN108446813A CN 108446813 A CN108446813 A CN 108446813A CN 201711372805 A CN201711372805 A CN 201711372805A CN 108446813 A CN108446813 A CN 108446813A
- Authority
- CN
- China
- Prior art keywords
- commodity
- comment
- text
- comment text
- theme
- 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
Classifications
-
- 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
- G06Q10/06395—Quality analysis or management
-
- 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
- G06Q30/00—Commerce
- G06Q30/06—Buying, selling or leasing transactions
- G06Q30/0601—Electronic shopping [e-shopping]
Landscapes
- Business, Economics & Management (AREA)
- Human Resources & Organizations (AREA)
- Engineering & Computer Science (AREA)
- Strategic Management (AREA)
- Development Economics (AREA)
- Economics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Entrepreneurship & Innovation (AREA)
- Marketing (AREA)
- Finance (AREA)
- Accounting & Taxation (AREA)
- Educational Administration (AREA)
- Physics & Mathematics (AREA)
- General Business, Economics & Management (AREA)
- Tourism & Hospitality (AREA)
- Game Theory and Decision Science (AREA)
- Quality & Reliability (AREA)
- Operations Research (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses the method for electric business service quality overall merit, this method is clustered by each theme of evaluative dimension discovery technique Mining Interesting commodity and to the comment short text of each theme；Sentiment analysis is carried out to each comment short text of each theme using emotion classifiers, obtains the Sentiment orientation information of each comment short text of each theme；The Sentiment orientation information of each comment short text of each theme is counted, determines the satisfaction rate of each theme, and then realize the service quality overall merit to the multi objective, various dimensions of paying close attention to commodity.
Description
Technical field
The present invention relates to electric business technical field more particularly to a kind of methods of electric business service quality overall merit.
Background technology
As e-commerce flourishes, shopping at network has become common one of shopping way.But user carries out net
When network is done shopping, user can not find suitable mode easily to obtain the accurate information in relation to businessman's Integrated Services Quality, cause
The reason of this situation approximately as：First, user can not contact actuals, therefore have very much to the perception of commercial quality
Limit；Secondly as the remoteness and retardance of businessman's service provided, usually efficiency is more low for the process of user's acquisition information
Under, and the information got is difficult to tell truth from falsehood sometimes；Third, National Commerce standardization effort also relatively lags behind, each
Businessman and commodity usually do not obtain the Standardization Quality certification of authority.
Currently, more traditional evaluation method is to participate in investigation and evaluation in the form of questionnaire, usually with the collection of certain time
Based on middle investigation, data and information collection lack continuity, simultaneously because the interval with service time, the timeliness of information lacks
It loses, can then influence the accuracy of feedback.Traditional evaluation method is based on credit appraisal, during consumer feedback is only evaluated
The consumer feedback data of one link, magnanimity are not fully utilized.Large-scale transmitting-receiving, statistics questionnaire need very
More manual operations, human cost are high.
Therefore, a kind of method preferably evaluated the commodity of electric business platform becomes technical problem urgently to be resolved hurrily.
Invention content
The purpose of the present invention is intended to solve above-mentioned one of technical problem at least to a certain extent.
For this purpose, first purpose of the present invention is that the method for the electric business service quality overall merit proposed, this method are logical
It crosses each theme of evaluative dimension discovery technique Mining Interesting commodity and the comment short text of each theme is clustered；Profit
Sentiment analysis is carried out to each comment short text of each theme with emotion classifiers, each comment for obtaining each theme is short
The Sentiment orientation information of text；The Sentiment orientation information of each comment short text of each theme is counted, is determined each
The satisfaction rate of theme, and then realize the service quality overall merit to the multi objective, various dimensions of paying close attention to commodity.It is this to be based on greatly
The E-business service quality evaluating method of data has certain advantage compared with conventional thought：First, the clothes of data-driven
Business quality evaluating method is that the service of goods experience based on true user is authenticated, therefore can be more objectively anti-
Answer the psychology of user in true buying experience；Secondly, the mode of automation collection analysis data is saving a large amount of manpower moneys
Source can more efficiently carry out extensive statistical analysis；Third, compared with questionnaire is provided in sampling, the user's evaluation data of magnanimity
Covering surface is wider, can more react the truth of entire user group.
To achieve the goals above, the method for the electric business service quality overall merit of first aspect present invention embodiment, packet
It includes：
According to the commodity data warehouse that the querying commodity information of concern commodity is built in advance, to obtain the concern commodity
At least one comment text, wherein the commodity data warehouse includes the merchandise news and field feedback of each commodity
Correspondence, the field feedback include comment text and scoring of the user to commodity；
Obtain at least one comment text after the data prediction of the concern commodity；
Using the corresponding topic model of the concern commodity at least one of the concern commodity after data prediction
Comment text carries out topic identification, determines at least one theme of the concern commodity and the short text set of each theme,
Wherein, the short text set includes at least one comment short text；
Each item of each theme short text set of the concern commodity is commented on using the emotion classifiers built in advance
Short text carries out sentiment analysis, obtains the Sentiment orientation information of each theme of the concern commodity, and each theme of statistics
Sentiment analysis result with obtain it is described concern commodity each theme satisfaction rate.
Method as described above further includes：
Utilize the merchandise news and field feedback of each commodity on web crawlers crawl electric business platform, wherein institute
It includes comment text and scoring of the user to commodity to state field feedback；
For each commodity, the merchandise news of current commodity and the correspondence of field feedback are established；
Commodity data warehouse, the commodity data warehouse are established according to the merchandise news of each commodity and field feedback
The correspondence of merchandise news and field feedback including each commodity.
Method as described above further includes：
The comment text collection of all kinds of commodity is obtained from the commodity data warehouse, wherein include at least one per class commodity
A commodity, each commodity include at least one comment text；
The each comment text for handling all kinds of commodity obtains whole comment short texts of all kinds of commodity；
Sentence-LDA models are trained using whole comment short texts of every class commodity, to build the master per class commodity
Inscribe model.
Method as described above, each comment text of all kinds of commodity of processing obtain the whole of all kinds of commodity and comment
By short text, including：
For each comment text of every class commodity, the client type for submitting each comment text and each comment are determined
The sentence of text is long；
According to the Chinese subordinate sentence rule of the long each comment text of selection of the corresponding client type of each comment text and sentence；
According to the corresponding Chinese subordinate sentence rule of each comment text to the Chinese subordinate sentence of each comment text progress and at least
Carry out following data pretreatment operation：Chinese word segmentation, word frequency statistics, word filtering, it is corresponding extremely to obtain each comment text
Few comment short text.
Method as described above further includes：
Determine the clear Sentiment orientation comment text collection for model training per class commodity, wherein clear Sentiment orientation
Comment text collection includes at least one clear Sentiment orientation comment text；
It converts the clear Sentiment orientation comment text of each item to corresponding each word using the term vector model built in advance
Sequence vector；
It will be trained in each term vector sequence inputting to deep neural network, to build the emotional semantic classification per class commodity
Device.
Method as described above, clear Sentiment orientation comment text collection for model training of the determination per class commodity
Including：
Whole clear Sentiment orientation comment datas per class commodity are obtained from the commodity data warehouse, wherein bright
True Sentiment orientation comment data includes clear Sentiment orientation scoring and clear Sentiment orientation corresponding with the scoring of clear Sentiment orientation
Comment text；
It is chosen from whole clear Sentiment orientation comment datas according to default selection rule and determines being used for for every class commodity
The clear Sentiment orientation comment text collection of model training.
Method as described above, the whole clear emotions obtained from the commodity data warehouse per class commodity are inclined
To comment data, including：
To obtaining whole field feedbacks progress down-samplings per class commodity from the commodity data warehouse；
Following data pretreatment operation is at least carried out to the whole field feedbacks obtained by down-sampling：Data are clear
Wash, the filtering of Chinese subordinate sentence, Chinese word segmentation, word frequency statistics, word, the long operation of unified sentence, to obtain whole clear Sentiment orientations
Comment data.
Method as described above, the deep neural network are any one of CNN, LSTM, GRU.
Method as described above further includes：
Storewide comment text collection is obtained from the commodity data warehouse, handles the comment text collection to obtain
Corpus；
Using training Skip-Gram models, to build term vector model.
Method as described above, it is described to handle the comment text collection to obtain corpus, including：
Following data pretreatment operation is at least carried out to whole comment text collection：Data cleansing, Chinese subordinate sentence, Chinese point
Word, word frequency statistics, word filtering, to obtain corpus.
The additional aspect of the present invention and advantage will be set forth in part in the description, and will partly become from the following description
Obviously, or practice through the invention is recognized.
Description of the drawings
Above-mentioned and/or additional aspect and advantage of the invention will become from the following description of the accompanying drawings of embodiments
Obviously and it is readily appreciated that, wherein
Fig. 1 is the flow diagram of the method for building up in the illustrative commodity data warehouse of the embodiment of the present invention；
Fig. 2 is the flow diagram of the method for building up of the illustrative topic identification model of the embodiment of the present invention；
Fig. 3 is illustrative word cloud；
Fig. 4 is the flow diagram of the method for building up of the illustrative emotion classifiers of the embodiment of the present invention；
Fig. 5 is the comment text distributed number figure of each star on certain illustrative electric business platform；
Fig. 6 is the flow diagram of the method for building up of the illustrative term vector model of the embodiment of the present invention；
Fig. 7 is the flow diagram of the method for the electric business service quality overall merit of one embodiment of the invention；
Fig. 8 is electric business service authentication Review Workflow figure；
Fig. 9 is the schematic diagram of the interactive mode of service quality evaluation system.
Specific implementation mode
The embodiment of the present invention is described below in detail, examples of the embodiments are shown in the accompanying drawings, wherein from beginning to end
Same or similar label indicates same or similar element or element with the same or similar functions.Below with reference to
The embodiment of attached drawing description is exemplary, it is intended to for explaining the present invention, and is not considered as limiting the invention.
Below with reference to the accompanying drawings the method for describing the electric business service quality overall merit of the embodiment of the present invention.
In information technology highly developed today, user has the evaluation data of merchant service quality and commercial quality
Higher availability --- nearly all electric business platform is each provided with the functions such as user's evaluation, user's scoring, consultation and advices.But
Since user's evaluation is usually from short text and normative poor, often information content is smaller for every comment, and the comment of different user is very
Contradiction extremely can mutually occur, add a series of excessively huge etc. reasons of data volume, for many commodity, user is often
All history evaluations can not possibly be read.However utilize big data analysis technology extensive to user's scoring, comment text etc.
Data carry out depth mining analysis, then can obtain that merchant service quality is relevant to enrich one's knowledge.For example, being sent out by evaluative dimension
Existing (aspect discovery) technology, can extract the dimension that user evaluates goods and services, excavate user and be directed to
Product and the focus of service；By sentiment analysis (sentiment analysis) technology, can analyze user to product and
The taste of service, and it is for statistical analysis to the favorable comment degree of goods and services based on comment text；And combine the two, then may be used
Each evaluative dimension and emotion corresponding with the dimension for occurring in text sentence are identified, and then realizes and refers to more
The service quality guarantee of mark, various dimensions.
The present invention devises the interactive system of an E-business service quality evaluation.The core solved in the present invention is asked
Topic is the quality evaluation dimension excavated user in class I goods and be more concerned about, and by the sentence in comment text according to above-mentioned
Dimension is clustered, and analyzes the Sentiment orientation of every comment sentence in each dimension, and then carry out to the satisfaction rate of each dimension
Statistical analysis is obtained to the overall merit of businessman or commodity in electric business platform.Here it is a kind of new text by this problem definition
This Mining Problems-dimension emotion excavates (Aspect Sentiment Mining, ASM).
It includes two sub-problems that dimension emotion, which is excavated,.First subproblem is that evaluative dimension excavates and comments on sentence cluster,
By taking electronic product as an example, the evaluative dimension excavated may include effect screen, sound effect, dispatching speed, price etc.；
And in the cluster of comment sentence, it is necessary first to every comment text is disassembled according to the dimension of evaluation, such as
" audio is praised very much, and is dispensed quickly！" this comment text, it should be disassembled first according to the dimension of evaluation as " audio is very
Praise " and " dispatching is quickly " two short sentences, it will followed by decompose two obtained short sentences and be included into " sound effect " and " dispatching respectively
Two evaluation themes of speed ".Second subproblem is to comment on sentence set accordingly to each evaluative dimension to carry out emotion point
Analysis, it is assumed for example that there are 10,000 short sentences to be grouped into " sound effect " this theme, then need short to each by emotion classifiers
Sentence carries out sentiment analysis respectively, the comment text item number of statistics emotion actively, passive it is each how many, finally calculate that " sound is imitated
A certain product or the satisfaction rate of brand on this dimension of fruit ".The models coupling of two sub-problems can be realized to electric business
The a certain businessman of platform or the various dimensions overall merit of a certain product：List the evaluation index and phase that user is more concerned about
The scoring event answered can have corresponding true comment sentence simultaneously for the front evaluation of each index and unfavorable ratings
Illustratively.
It is that this background proposes based on electric business service quality overall merit that dimension emotion, which excavates this problem, it and electric business
The characteristics of platform data, combines closely.First, the label in electric business platform and there is no evaluative dimension, is for each
Indicate this sentence makes evaluation for which dimension to commodity there is no data for comment sentence.Secondly, electric business is flat
The text set for having affective tag can be found in platform, because the scoring of user reflects the emotion of corresponding text to a certain extent
Tendency, and extreme scoring (full marks or minimum point) often can be used as reliable label.Third, electric business are flat
Often every user is only that the commodity of purchase make a call to a whole score to platform, does not have the case where giving a mark respectively to each subitem, because
This initial data only provides each product or the whole score of businessman, can't provide their performance feelings in different aspect
Condition.Therefore can furthermore, the particularity of electric business platform initial data determines that dimension emotion excavates this problem
Particularity；Realize that product and the various dimensions of businessman are evaluated in electric business platform using the thinking that dimension emotion is excavated, it can be with maximum journey
It makes good use of to degree the available data of electric business platform and is marked without additional data.
To realize that the commodity to electric business platform carry out dimension emotion excavation, need early period to establish commodity data warehouse, theme
Model, sentiment analysis device, term vector model, start to introduce below commodity data warehouse, topic model, sentiment analysis device, word to
Measure the method for building up of model.
Fig. 1 is the flow diagram of the method for building up in the illustrative commodity data warehouse of the embodiment of the present invention.This implementation
The mass datas such as merchandise news and field feedback on electric business platform are stored by commodity data warehouse in example, in this way
Depth is carried out to the field feedback of the magnanimity such as user's scoring, comment text for later use big data analysis technology
Mining analysis obtains relevant enrich one's knowledge etc. of merchant service quality and provides feasibility.
As shown in Figure 1, the method for building up in the commodity data warehouse includes the following steps：
S101, the merchandise news and field feedback of each commodity on web crawlers crawl electric business platform are utilized.
Wherein, the field feedback includes comment text and scoring of the user to commodity.
Specifically, electric business platform be each provided with functions, the users such as user's evaluation, user's scoring, consultation and advices can be in electricity
Merchant service quality and commercial quality etc. are evaluated on quotient's platform.For example, company A electric business is obtained by web crawlers
All dependent merchandise information of three categories commodity (being respectively laptop, removal vehicle seat cover, umbrella), user feedback in platform
Information, data scale 1.64G.Merchandise news totally 902, including commodity ID, trade name, commodity price, retail shop ID, use
The number distribution etc. of family number of reviews, different stars；Comment data totally 7146603, including lower single time, user name, user
Province, comment serviceability poll, comment text, client, scoring, user class etc..
S102, each commodity are directed to, establish the merchandise news of current commodity and the correspondence of field feedback；
S103, commodity data warehouse is established according to the merchandise news and field feedback of each commodity.
Wherein, the commodity data warehouse includes the merchandise news of each commodity and the correspondence of field feedback.
Specifically, the correspondence for first establishing the corresponding field feedback of merchandise news of each commodity, by quotient
The correspondence of product information and field feedback is stored to the data warehouse based on Hadoop and Hive to establish data bins
Library.Hadoop is the software platform of an exploitation and operation processing large-scale data, realizes the cluster formed in a large amount of computers
In to mass data carry out Distributed Calculation.Hive be based on Hadoop on data warehouse, can store, inquire and divide
Analysis is stored in the large-scale data in Hadoop.It so subsequently can be according to the querying commodity information data warehouse of concern commodity
Obtain the field feedbacks such as comment text and the scoring of concern commodity.
The method provided in this embodiment for establishing commodity data warehouse obtains magnanimity first with web crawlers from electric business platform
The data of data, magnanimity may include merchandise news, field feedback etc., and mass data is stored in a distributed manner
In the commodity data warehouse based on Hadoop and Hive, to carry out large-scale calculations analysis.
Fig. 2 is the flow diagram of the method for building up of the illustrative topic model of the embodiment of the present invention.In the present embodiment
Topic model dimension that the comment text of commodity is commented on can be excavated.Based on the data set got, user is excavated
The theme of comment on commodity is carried out, obtains the goods and services evaluative dimension of user's concern, and according to excavating obtained theme to each
The comment text of a commodity carries out topic identification, namely identifies that the evaluation of its description is tieed up for each short sentence of comment text
Degree, to carry out short sentence fractionation to the comment text of the commodity and sort out.
As shown in Fig. 2, the method for building up of the topic model includes the following steps：
S201, the comment text collection that all kinds of commodity are obtained from the commodity data warehouse.
In the present embodiment, comment text collection can be understood as the set of whole comment text compositions, per class commodity bundle
At least one commodity are included, each commodity include at least one comment text.Type of merchandize on electric business platform is various, in this reality
It applies in example, needs the comment text collection for categorizedly obtaining all kinds of commodity, subsequently built according to the comment text collection of all kinds of commodity
Found the topic model of all kinds of commodity.
Each comment text of S202, all kinds of commodity of processing obtain whole comment short texts of all kinds of commodity.
Specifically, since the punctuate of comment text in electric business platform is extremely lack of standardization, carry out subordinate sentence rule also compared with
For complexity.Space and "~", "..." etc. symbols be likely to be sentence decollator.In general, shorter text is often
It is to be submitted from mobile terminal, and often will appear the case where replacing punctuate with space in the input of mobile terminal, such case is longer
Text in will not then occur.Therefore, it needs to judge the provision according to the length of sentence and the type of client in subordinate sentence
What classification policy or rule should be used.Notice that each subordinate sentence of a long sentence is possible to describe different themes,
Therefore each long sentence has also been cut into multiple short sentences, to ensure that the theme of every words has higher purity.
In one possible implementation, the specific implementation of step S202 includes：
Step S21, for each comment text of every class commodity, determine the client type for submitting each comment text and
The sentence of each comment text is long.
Step S22, according to the Chinese of the long each comment text of selection of the corresponding client type of each comment text and sentence
Subordinate sentence rule.
Step S23, Chinese subordinate sentence is carried out to each comment text according to the corresponding Chinese subordinate sentence rule of each comment text
And at least carry out following data pretreatment operation：Chinese word segmentation, word frequency statistics, word filtering, to obtain each comment text
Corresponding at least one comment short text.
Specifically, after completing Chinese subordinate sentence to comment text, remainder data pretreatment work, including Chinese point are proceeded by
Word, word frequency statistics, word filtering etc., with remove the occurrence number in original comment short text be 3 times or lower low-frequency word,
Punctuate and stop words etc., and then obtain comment short text.
The present embodiment before training topic model, to from comment file that electric business platform acquire carry out Chinese subordinate sentence, in
Operations, the acquired comment short texts such as literary participle, word frequency statistics, word filtering have the theme of higher degree, can ensure institute
The topic model of foundation has good topic identification ability.
S203, Sentence-LDA models are trained using whole comment short texts of every class commodity, to build per class quotient
The topic model of product.
It should be pointed out that in the present embodiment, for different classes of commodity, constructing different theme moulds respectively
Type subsequently determines concern commodity generic according to the merchandise news of concern commodity, utilizes theme mould corresponding with concern commodity
Multiple themes (i.e. each evaluative dimension) of type identification concern commodity and the comment short text of each theme.
Table 1 lists LDA (Latent Dirichlet Allocation) distribution subjects and the larger word of respective weights
Language.By taking merchandise classification is laptop as an example, theme that table 1 is listed, such as price, dispatching speed, starting up speed etc., base
Strictly consumer buys the assessment dimension more paid close attention to when laptop from the point of view of visual impression.
The distribution subject and corresponding word that table 1LDA models obtain
Table 2 list Sentence-LDA (Sentence-Latent Dirichlet Allocation) distribution subjects with
And the word that respective weights are larger.It is worth noting that, Sentence-LDA will be in comment text with each in a word
A word imparts identical theme, that is to say using a sentence rather than an entire document as a bag of words, therefore certain journey
The information in inside documents structure is utilized on degree.And from table 2, it has also been discovered that, it compares and LDA models, sentence-
LDA models, which can excavate some, has more fine-grained theme, such as " game effect ", " appearance " etc..
The distribution subject and corresponding word that table 2Sentence-LDA models obtain
Sentence-LDA models are that each short sentence is assigned with theme, therefore can be carried out to short sentence according to distributed theme
Subject Clustering.In this way, for each theme, all there can be a large amount of short sentences as the example collection belonging to the theme.Here " to match
Send speed ", " after-sale service ", for " price " three themes, table lists several sentences belonging to above three theme at random
(Business Name in sentence has been replaced by " company A ").It can be seen that the sentence set belonging to each theme really can be with
The theme quite well.Such as " so evening company A master worker returns me and delivers goods to the customers ", " double 11 the second arrivals "
All it is the real user comment of typical description dispatching speed.
The affiliated sentence example of distribution subject in table 3Sentence-LDA models
Fig. 3 is illustrative word cloud.Fig. 3 has more intuitively reacted " dispatching speed ", " after-sale service ", " price " three samples
The overall condition of example the included text data of theme.In three word clouds in figure 3, the area shared by word is with word frequency at just
Than, namely bigger word occur it is about frequent.It is not difficult to find out that the high frequency words under each theme can be protected with corresponding theme
It holds unanimously, such as in " dispatching speed " this theme, the words such as " receiving ", " logistics ", " second day " are all close with the theme
It is relevant.
The method for building up of topic model provided in this embodiment excavates commenting for commodity by Sentence-LDA model depths
Valence dimension, and Subject Clustering is carried out to the comment short text of commodity according to evaluative dimension, to realize pass of the user to commodity
Note the excavation of point.
Fig. 4 is the flow diagram of the method for building up of the illustrative emotion classifiers of the embodiment of the present invention.The present embodiment
In the emotion classifiers trained can be used for on each evaluative dimension each item comment short text carry out sentiment analysis.
As shown in figure 4, the method for building up of the emotion classifiers includes the following steps：
S301, the clear Sentiment orientation comment text collection for model training for determining every class commodity.
Wherein, it includes at least one clear Sentiment orientation comment text to specify Sentiment orientation comment text collection.
In the present embodiment, it includes that positive emotion is inclined to comment text to specify Sentiment orientation comment text i.e., also includes disappearing
Pole Sentiment orientation comment text.For example, company A platform provides the Star rating to commodity and its service for user
Window, by Star rating window, user can provide commodity and its service the scoring in one star to five constellations.
One star scoring illustrates user to very dissatisfied to commodity and its service, it can be understood as Negative Affect tendency scoring, user give
It is considered as being inclined to Negative Affect to go out the comment write out when Negative Affect tendency scoring.Five star scorings illustrate user
Commodity and its service are felt quite pleased, it can be understood as positive emotion tendency scoring, user write when providing positive emotion tendency scoring
The comment gone out is considered as being inclined to Negative Affect.Certainly, the scoring of Negative Affect tendency and positive emotion tendency scoring basis
Practical situation is set, and is not limited herein.
In one possible implementation, the specific implementation of step S301 is：
Step S31, whole clear Sentiment orientation comment datas per class commodity are obtained from the commodity data warehouse.
Wherein, it includes the scoring of clear Sentiment orientation and corresponding with the scoring of clear Sentiment orientation to specify Sentiment orientation comment data
Clear Sentiment orientation comment text.It includes that the scoring of positive emotion tendency and Negative Affect tendency are commented to specify Sentiment orientation scoring
Point；It includes positive emotion tendency comment text and Negative Affect tendency comment text to specify Sentiment orientation comment text.
Specifically, the field feedbacks that the whole per class commodity is obtained from the commodity data warehouse adopt
Sample；Following data pretreatment operation is at least carried out to the whole field feedbacks obtained by down-sampling：Data cleansing,
Chinese subordinate sentence, Chinese word segmentation, word frequency statistics, word filtering, the long operation of unified sentence, are commented with the clear Sentiment orientation for obtaining whole
By data.
Fig. 5 is the comment text distributed number figure of each star on certain illustrative electric business platform.As shown in figure 5, certain is electric
On quotient's platform, the comment text quantity of five-pointed star far more than a star comment text quantity, data there is stronger excess kurtosis,
The entry number of front comment is negative 37 times.After carrying out data cleansing and Chinese subordinate sentence, discovery obtains front comment sentence
8,371,347 and negative reviews sentence 444,781, the two ratio is reduced to 18.8 times.Ratio is reduced since difference is commented
Text often tends to that length is longer, sentence is more, and this point is easier to understand, however front comment and front comment number
Amount difference is still more greatly different.For this data nonbalance (data imbalances) problem, the present embodiment passes through down-sampling
(undersampling) overcome.
Down-sampling, data cleansing, Chinese subordinate sentence, Chinese word segmentation, word frequency system are carried out successively to whole field feedbacks
Data preprocessing operations, removal low-frequency word, filtering punctuate and the stop words such as meter, word filtering, the long operation of unified sentence obtain bright
True Sentiment orientation comment data.
It should be pointed out that by data cleansing, it is indefinite Sentiment orientation can be rejected from whole field feedbacks
Field feedback, leave the specific field feedback of Sentiment orientation.For example, a star scoring indicates passive feelings
Sense tendency scoring, five star scorings indicate positive emotion tendency scoring, and two star scorings, three star scorings, four star scorings are then
It is understood to the indefinite scoring of Sentiment orientation.In data cleansing, reject Sentiment orientation it is indefinite scoring and its it is corresponding
Comment text.
The long operation of unified sentence is briefly introduced herein.Since comment text much has mobile terminal submission, have a large amount of
Ultrashort text, and it includes a word, such as the words such as " good ", " good ", " praising very much ", " rubbish " to lead to many sentences only
It usually can independently constitute the even comment of a sentence.It, can be right in view of most sentence sequence lengths within 20
Sentence blocked with zero padding (zero padding), to which sentence length is unified into 20.Certainly, according to actual needs will
Sentence length is unified into design length.
Step S32, it chooses and determines per class quotient from whole clear Sentiment orientation comment datas according to default selection rule
The clear Sentiment orientation comment text collection for model training of product.
Specifically, whole clear emotions is got to the data preprocessing operation of field feedback by step S31
It is inclined to comment data.Default selection rule can be chosen according to a certain percentage from whole clear Sentiment orientation comment datas
Partial clear Sentiment orientation comment data, as the clear Sentiment orientation comment text collection for model training.Certainly, after
It is continuous can also from the clear Sentiment orientation comment data of whole clear Sentiment orientation comment data selected parts, as
Model verification clear Sentiment orientation comment text collection and as the clear Sentiment orientation comment text collection for model measurement.
Default selection rule is set with specific reference to practical situation.
In the present embodiment, specifying Sentiment orientation comment text can reflect that user is more reliable to commodity and its service
Experience Degree, as the training data of sentiment analysis device, it is ensured that the sentiment analysis device trained can accurately into
Row sentiment analysis.
S302, converted the clear Sentiment orientation comment text of each item to using the term vector model built in advance it is corresponding each
A term vector sequence.
S303, it will be trained in each term vector sequence inputting to deep neural network, to build the feelings per class commodity
Feel grader.
Since term vector can farthest understand the semantic relation between word, using term vector as such as depth nerve
The input of network even depth learning model helps to promote classifying quality.In the present embodiment, deep neural network can be
CNN (Convolutional Neural Network, convolutional neural networks), LSTM (Long Short-Term Memory,
Shot and long term memory network), any one of GRU (Gated Recurrent Unit, thresholding cycling element), but not with this
It is limited.
It should be pointed out that after building emotion classifiers, the clear emotion verified for model can also be inclined
It is input to emotion classifiers to comment text collection to be verified, and will be commented as the clear Sentiment orientation for model measurement
Emotion classifiers are input to by text set to be tested, by the performance for verifying and testing Continuous optimization emotion classifiers.
The method for building up of emotion classifiers provided in this embodiment is used for the bright of model training by determining per class commodity
True Sentiment orientation comment text collection；The clear Sentiment orientation comment text of each item is converted using the term vector model built in advance
For corresponding each term vector sequence；It will be trained in each term vector sequence inputting to deep neural network, it is every to build
The emotion classifiers of class commodity.This method trains deep neural network, structure using with clear Sentiment orientation comment text
Emotion classifiers, the emotion classifiers can be used for carrying out accurate feelings to each item comment short text on each evaluative dimension
Sense analysis.
Fig. 6 is the flow diagram of the method for building up of the illustrative term vector model of the embodiment of the present invention.The present embodiment
Established term vector model makes the input quantity being input in emotion classifiers be term vector sequence, helps to promote classifying quality.
As shown in fig. 6, the method for building up of term vector model provided in this embodiment, including：
S401, storewide comment text collection is obtained from the commodity data warehouse, handle the comment text collection
To obtain corpus.
In one possible implementation, the specific implementation of step S401 is：Extremely to whole comment text collection
Following data pretreatment operation is carried out less：Data cleansing, Chinese subordinate sentence, Chinese word segmentation, word frequency statistics, word filtering, to obtain
Corpus.
The present embodiment carries out data cleansing, Chinese subordinate sentence, Chinese word segmentation, word by whole comment text collection to commodity
Frequency statistics, word filtering etc. a series of data preprocessing operation, remove in the corpus of acquisition there is no low-frequency word, punctuate,
Stop words etc. obtains the corpus of commodity.
S402, using training Skip-Gram models, to build term vector model.
In the present embodiment, the term vector model based on Skip-Gram models can be better understood from the semanteme between word
Relationship.
Table 4 lists the best term vector and included angle cosine of example term.By taking corpus is notebook corpus as an example, choosing
Example of the partial words (partial words are respectively association, heroic alliance, bad luck, two minutes, Beijing) as verification is taken, respectively
It seeks being listed in Table 4 below together with highest 10 words of their semantic similarities and included angle cosine.As can be seen from Table 4, with
Word " A brand computers " immediate word includes " B brand computers ", " C brand computers ", " D brand computers " etc., same equal
For computer brand name；With the term vector corresponding respectively " lol " of " heroic alliance " word angle minimum, " LOL ", " alliance " etc.,
In most similar two words all be play " heroic alliance " english abbreviation；It is then " poor with the most similar word of " bad luck " word
The vocabulary of the equally expression negative emotions such as strength ", " poor "；With corresponding respectively " one point of the term vector of " two minutes " word angle minimum
Clock ", " three minutes " etc. indicate the word of minute or second rank time span；And this place name semanteme is closest with " Beijing "
Be similarly the place names such as " Guangzhou ", " Shanghai ".It is worth noting that, most similar in this phrase semantic with " A brand computers "
Word further includes " D1 brand computers ", and actually this word is the malapropism version of computer brand " D brand computers " word,
I.e. term vector actually identifies wrong word common in text, and can identify the semanteme for including wrong word word.With such
It pushes away, the B1 brand computers, B2 brand computers in table 4 are the malapropism version of B brand computers, C1 brand computers, C2 in table 4
Brand computer is the malapropism version of C brand computers.By the verification of above-mentioned example it is found that being based on entire notebook corpus
The term vector model constructed can largely understand the semantic relation between word, the word that term vector model is exported to
It measures and will be helpful to promote classifying quality as the input of deep learning model.
The most close term vector and included angle cosine of 4 example term of table
Fig. 7 is the flow diagram of the method for the electric business service quality overall merit of one embodiment of the invention.The present embodiment
To be illustrated to how user's commodity of interest carry out dimension sentiment analysis.
As shown in fig. 7, the method for the electric business service quality overall merit includes the following steps：
S501, the commodity data warehouse built in advance according to the querying commodity information of concern commodity, to obtain the concern
At least one comment text of commodity.
Wherein, the commodity data warehouse includes the merchandise news of each commodity and the correspondence of field feedback,
The field feedback includes comment text and scoring of the user to commodity.
For paying close attention to commodity and be certain brand laptop, a lot of other users take down notes certain brand on electric business platform
This into
S502, at least one comment text after the data prediction for paying close attention to commodity is obtained.
Specifically, come from different user to the comment text of concern commodity, different in size, some is write very professional, has
Write very arbitrarily, need to carry out a series of data prediction to it, can be completed to electricity when building topic model
The data prediction of each commodity on quotient's platform.
The flow of data prediction is as follows：
Step S51, for each comment text of each commodity, determine the client type for submitting each comment text and
The sentence of each comment text is long.
Step S52, according to the Chinese of the long each comment text of selection of the corresponding client type of each comment text and sentence
Subordinate sentence rule.
Step S53, Chinese subordinate sentence is carried out to each comment text according to the corresponding Chinese subordinate sentence rule of each comment text
And at least carry out following data pretreatment operation：Chinese word segmentation, word frequency statistics, word filtering, to obtain each comment text
Corresponding at least one comment short text.
After above-mentioned steps, each comment text is divided into multiple comment short texts.The comment short text will not
There is various low-frequency words, punctuate, stop words etc., and the theme with higher degree.
S503, using the corresponding topic model of the concern commodity to the concern commodity by data prediction extremely
A few comment text carries out topic identification, determines at least one theme of the concern commodity and the short essay of each theme
This set, wherein the short text set includes at least one comment short text.
For paying close attention to commodity and be certain brand laptop, user distinguishes the evaluative dimension of certain brand laptop
For promotional price, after-sale service, logistics speed, hardware configuration, body appearance, each evaluative dimension corresponds to a theme.This reality
Topic identification and Subject Clustering can be carried out by whole evaluation texts of certain brand laptop by applying the topic model in example,
It exports each theme and corresponds to short text set, short text set includes at least one comment short text.
S504, using the emotion classifiers built in advance to it is described concern commodity each theme short text set each item
It comments on short text and carries out sentiment analysis, obtain the Sentiment orientation information of each theme of the concern commodity, and count each
The sentiment analysis result of theme is to obtain the satisfaction rate of each theme of the concern commodity.
By step S503, the Topics Crawling to paying close attention to commodity and the short text clustering based on comment dimension are completed.With
It pays close attention to for commodity are certain brand laptop, excavates promotional price respectively, after-sale service, logistics speed, hardware are matched
It sets, 5 themes such as body appearance, each theme is to by comment short text.Sentiment analysis device is short to the comment of each theme one by one
Text carries out sentiment analysis, determines the Sentiment orientation information of each comment short text.
For example, for this evaluative dimension of promotional price, 100 evaluation short texts, the emotion of sentiment analysis device are corresponded to altogether
Trend information is：The Sentiment orientation information of 60 evaluation short texts scores for five-pointed star, the Sentiment orientation letter of 30 evaluation short texts
Breath is that four stars score, and the Sentiment orientation information of 10 evaluation short texts scores for a star.According to the statistical method of setting to promotion
The Sentiment orientation information of this evaluative dimension of price is counted, and determines the satisfaction rate 80% of this evaluative dimension of promotional price.
In the present embodiment, statistical method is set according to practical situation, specific unlimited.
The method of electric business service quality overall merit provided in an embodiment of the present invention, is excavated by evaluative dimension discovery technique
It pays close attention to each theme of commodity and the comment short text of each theme is clustered；Using emotion classifiers to each theme
Each comment short text carry out sentiment analysis, obtain the Sentiment orientation information of each comment short text of each theme；To every
The Sentiment orientation information of each comment short text of one theme is counted, and determines the satisfaction rate of each theme, and then realization pair
Pay close attention to the multi objective of commodity, the service quality overall merit of various dimensions.This E-business service quality based on big data is commented
Valence method has certain advantage compared with conventional method：First, the QoS evaluating method of data-driven is based on true
What the service of goods experience of user was authenticated, therefore more can objectively react the heart of user in true buying experience
Reason；Secondly, the mode of automation collection analysis data is saving a large amount of human resources, can more efficiently carry out extensive
Statistical analysis；Third, compared with questionnaire is provided in sampling, the user's evaluation data cover face of magnanimity is wider, can more react entire use
The truth of family group.
The overall architecture of electric business service authentication review system is illustrated below.Fig. 8 is electric business service authentication evaluation stream
Cheng Tu, Fig. 9 are the schematic diagram of the interactive mode of service quality evaluation system.Service quality evaluation system in Fig. 8 is appreciated that
For the executive agent of the method for executing electric business service quality overall merit.
Referring to such as 8, the data-driven function that web crawlers provides for electric business service authentication review system is used for from electric business
Data are captured in platform；The model of core be service quality evaluation system in based on the topic model of Sentence-LDA with
And the emotion classifiers based on deep learning.Trained topic model is first carried out evaluative dimension and excavates to be gathered with comment short sentence
Class, the satisfaction rate that each comment dimension is next carried out with emotion classifiers count, and finally provide each commodity in different indexs
On comprehensive score, big data analysis report and syndic the staff of service quality certification (artificial carry out) generated.
Its main function has two：First, with suggestiveness can expand existing evaluation index；Second is that being provided at the beginning of one for syndic
The service quality review result of step.Syndic continues subsequent evaluation link after obtaining above-mentioned support.
It should be noted that user comment text and to containing bulk information in the scoring of commodity, but still can not capsule
Include whole dimensions that service quality overall merit is carried out to electric business.In e-commerce platform service quality evaluation index, such as logistics
Delivery service ensures that after sale service, technical performance etc. are usually referred to by user in comment text, and the main body of businessman
Qualification, management system, the disclosure of system measure content etc. can only then be audited by auditor's by other means, from user
Evaluation data in can not excavate enough relevant knowledges.Therefore, the cooperation interaction formula of algorithm and auditor evaluation should be this
The operational mode of system.Certificate scheme in Fig. 8 and interpretational criteria by《E-commerce platform service quality evaluation is drawn with grade
Point》Etc. national standards provide, the index and score accounting of service quality evaluation are defined in these national standards.It is in this
For system, the evaluation index in scheme and criterion can be used as the comment theme paid close attention to apply in evaluative dimension excavation；
And the theme that model is excavated equally can be used for suggestiveness the expansion of evaluation index in existing standard.
In addition, this system constantly promotes the performance of itself by Active Learning and on-line study.As shown, at this
In interactive evaluation procedure, when evaluating certain class I goods, a series of commented by what algorithm identified that user is more concerned about first
Valence dimension and index；For the commodity or businessman for needing to be authenticated evaluation, calculate in corresponding evaluation sentence above-mentioned each
The accounting of positive emotion in dimension, you can the positive rating that each commodity or businessman obtain in above-mentioned each dimension respectively is obtained,
Each merchant service qualitative data analysis report of final output.Auditor is based on the selected needs of data analysis report that system generates
The evaluative dimension further appreciated that, system returns to the abstract of all related commentaries of the dimension, while extracting and this topic relativity
Higher comment text carries out further subject matter analysis.
For the text extracted, auditor can be audited by marking a certain amount of data update sorter model help
Member further carries out text the classification of higher granularity, this process is by Active Learning (active learning) and online
It is accomplished to learn (online learning) two kinds of machine learning methods.Wherein, Active Learning is during data mark
The sample for needing most to obtain domain expert's mark is selectively selected to be putd question to data mark person from a large amount of Unlabeled datas,
Namely the efficiency of data mark and model training is improved by " the high sample of preferential mark weight "；And on-line study is then led to
It crosses new data and is continually updated model so that model performance is constantly promoted.Emerging data, auditor newly mark on electric business platform
The data of note can all so that this system is optimized.
In the description of this specification, reference term " one embodiment ", " some embodiments ", " example ", " specifically show
The description of example " or " some examples " etc. means specific features, structure, material or spy described in conjunction with this embodiment or example
Point is included at least one embodiment or example of the invention.In the present specification, schematic expression of the above terms are not
It must be directed to identical embodiment or example.Moreover, particular features, structures, materials, or characteristics described can be in office
It can be combined in any suitable manner in one or more embodiments or example.In addition, without conflicting with each other, this field
Technical staff can carry out the feature of different embodiments or examples described in this specification and different embodiments or examples
In conjunction with and combination.
In addition, term " first ", " second " are used for description purposes only, it is not understood to indicate or imply relative importance
Or implicitly indicate the quantity of indicated technical characteristic.Define " first " as a result, the feature of " second " can be expressed or
Person implicitly includes at least one this feature.In the description of the present invention, the meaning of " plurality " is at least two, such as two,
Three etc., unless otherwise specifically defined.
Any process described otherwise above or method description are construed as in flow chart or herein, and expression includes
It is one or more for realizing custom logic function or process the step of the module of code of executable instruction, segment or
Part, and the range of the preferred embodiment of the present invention includes other realization, wherein can not press shown or discussion
Sequentially, include according to involved function by it is basic simultaneously in the way of or in the opposite order, to execute function, this should be by this
The embodiment person of ordinary skill in the field of invention is understood.
Expression or logic and/or step described otherwise above herein in flow charts, for example, being considered use
In the order list for the executable instruction for realizing logic function, may be embodied in any computer-readable medium, for
Instruction execution system, device or equipment (system of such as computer based system including processor or other can be from instruction
Execute system, device or equipment instruction fetch and the system that executes instruction) use, or combine these instruction execution systems, device or
Equipment and use.For the purpose of this specification, " computer-readable medium " can any can be included, store, communicating, propagating
Or transmission program uses for instruction execution system, device or equipment or in conjunction with these instruction execution systems, device or equipment
Device.The more specific example (non-exhaustive list) of computer-readable medium includes following：It is connected up with one or more
Electrical connection section (electronic device), portable computer diskette box (magnetic device), random access memory (RAM), read-only memory
(ROM), erasable edit read-only storage (EPROM or flash memory), fiber device and portable optic disk are read-only
Memory (CDROM).In addition, can even is that can the paper of print routine or other are suitable on it for computer-readable medium
Medium because can for example by carrying out optical scanner to paper or other media, then into edlin, interpretation or when necessary with
Other suitable methods are handled electronically to obtain program, are then stored in computer storage.
It should be appreciated that each section of the present invention can be realized with hardware, software, firmware or combination thereof.Above-mentioned
In embodiment, multiple steps or method can in memory and by suitable instruction execution system be executed soft with storage
Part or firmware are realized.Such as, if with hardware come realize in another embodiment, can be under well known in the art
Any one of row technology or their combination are realized：With the logic gate electricity for realizing logic function to data-signal
The discrete logic on road, the application-specific integrated circuit with suitable combinational logic gate circuit, programmable gate array (PGA) are existing
Field programmable gate array (FPGA) etc..
Those skilled in the art are appreciated that realize all or part of step that above-described embodiment method carries
Suddenly be that relevant hardware can be instructed to complete by program, program can be stored in a kind of computer readable storage medium
In, which includes the steps that one or a combination set of embodiment of the method when being executed.
In addition, each functional unit in each embodiment of the present invention can be integrated in a processing module, it can also
That each unit physically exists alone, can also two or more units be integrated in a module.Above-mentioned integrated mould
The form that hardware had both may be used in block is realized, can also be realized in the form of software function module.If integrated module with
The form of software function module is realized and when sold or used as an independent product, can also be stored in a computer can
It reads in storage medium.
Storage medium mentioned above can be read-only memory, disk or CD etc..Although having been shown and retouching above
The embodiment of the present invention is stated, it is to be understood that above-described embodiment is exemplary, and should not be understood as the limit to the present invention
System, those skilled in the art above-described embodiment can be changed, be changed within the scope of the invention, replaced and
Modification.
Claims (10)
1. a kind of method of electric business service quality overall merit, which is characterized in that including：
According to the commodity data warehouse that the querying commodity information of concern commodity is built in advance, to obtain the concern commodity at least
One comment text, wherein the commodity data warehouse includes that the merchandise news of each commodity is corresponding with field feedback
Relationship, the field feedback include comment text and scoring of the user to commodity；
Obtain at least one comment text after the data prediction of the concern commodity；
At least one comment using the corresponding topic model of the concern commodity to the concern commodity after data prediction
Text carries out topic identification, determines at least one theme of the concern commodity and the short text set of each theme, wherein
The short text set includes at least one comment short text；
Short essay is commented on to each item of each theme short text set of the concern commodity using the emotion classifiers built in advance
This progress sentiment analysis obtains the Sentiment orientation information of each theme of the concern commodity, and the feelings of each theme of statistics
Sense analysis result is to obtain the satisfaction rate of each theme of the concern commodity.
2. the method as described in claim 1, which is characterized in that further include：
Utilize the merchandise news and field feedback of each commodity on web crawlers crawl electric business platform, wherein the use
Family feedback information includes comment text and scoring of the user to commodity；
For each commodity, the merchandise news of current commodity and the correspondence of field feedback are established；
Commodity data warehouse is established according to the merchandise news of each commodity and field feedback, the commodity data warehouse includes
The merchandise news of each commodity and the correspondence of field feedback.
3. the method as described in claim 1, which is characterized in that further include：
The comment text collection of all kinds of commodity is obtained from the commodity data warehouse, wherein every class commodity include at least one quotient
Product, each commodity include at least one comment text；
The each comment text for handling all kinds of commodity obtains whole comment short texts of all kinds of commodity；
Sentence-LDA models are trained using whole comment short texts of every class commodity, to build the theme mould per class commodity
Type.
4. method as claimed in claim 3, which is characterized in that each comment text of all kinds of commodity of processing obtains each
Whole comment short texts of class commodity, including：
For each comment text of every class commodity, the client type for submitting each comment text and each comment text are determined
Sentence it is long；
According to the Chinese subordinate sentence rule of the long each comment text of selection of the corresponding client type of each comment text and sentence；
Chinese subordinate sentence is carried out to each comment text according to the corresponding Chinese subordinate sentence rule of each comment text and is at least carried out
Following data pretreatment operation：Chinese word segmentation, word frequency statistics, word filtering, to obtain each comment text corresponding at least one
Item comments on short text.
5. the method as described in claim 1, which is characterized in that further include：
Determine the clear Sentiment orientation comment text collection for model training per class commodity, wherein clear Sentiment orientation comment
Text set includes at least one clear Sentiment orientation comment text；
It converts the clear Sentiment orientation comment text of each item to corresponding each term vector using the term vector model built in advance
Sequence；
It will be trained in each term vector sequence inputting to deep neural network, to build the emotion classifiers per class commodity.
6. method as claimed in claim 5, which is characterized in that clear feelings for model training of the determination per class commodity
Comment text collection is inclined in sense：
Whole clear Sentiment orientation comment datas per class commodity are obtained from the commodity data warehouse, wherein clear feelings
Sense tendency comment data includes clear Sentiment orientation scoring and clear Sentiment orientation comment corresponding with the scoring of clear Sentiment orientation
Text；
It chooses and determines per class commodity for model from whole clear Sentiment orientation comment datas according to default selection rule
Trained clear Sentiment orientation comment text collection.
7. method as claimed in claim 5, which is characterized in that described to be obtained per class commodity from the commodity data warehouse
Whole clear Sentiment orientation comment datas, including：
To obtaining whole field feedbacks progress down-samplings per class commodity from the commodity data warehouse；
Following data pretreatment operation is at least carried out to the whole field feedbacks obtained by down-sampling：Data cleansing,
Chinese subordinate sentence, Chinese word segmentation, word frequency statistics, word filtering, the long operation of unified sentence, are commented on obtaining whole clear Sentiment orientations
Data.
8. method as claimed in claim 5, which is characterized in that the deep neural network is any in CNN, LSTM, GRU
Kind.
9. the method as described in claim 1, which is characterized in that further include：
Storewide comment text collection is obtained from the commodity data warehouse, handles the comment text collection to obtain language material
Library；
Using training Skip-Gram models, to build term vector model.
10. method as claimed in claim 9, which is characterized in that the processing comment text collection is to obtain corpus, packet
It includes：
Following data pretreatment operation is at least carried out to whole comment text collection：Data cleansing, Chinese subordinate sentence, Chinese word segmentation,
Word frequency statistics, word filtering, to obtain corpus.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711372805.9A CN108446813A (en) | 2017-12-19 | 2017-12-19 | A kind of method of electric business service quality overall merit |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711372805.9A CN108446813A (en) | 2017-12-19 | 2017-12-19 | A kind of method of electric business service quality overall merit |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108446813A true CN108446813A (en) | 2018-08-24 |
Family
ID=63190753
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711372805.9A Pending CN108446813A (en) | 2017-12-19 | 2017-12-19 | A kind of method of electric business service quality overall merit |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108446813A (en) |
Cited By (32)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109165387A (en) * | 2018-09-20 | 2019-01-08 | 南京信息工程大学 | A kind of Chinese comment sentiment analysis method based on GRU neural network |
| CN109359246A (en) * | 2018-12-07 | 2019-02-19 | 上海宏原信息科技有限公司 | A kind of brand cohesion calculation method based on forum user speech |
| CN109461037A (en) * | 2018-12-17 | 2019-03-12 | 北京百度网讯科技有限公司 | Comment on viewpoint clustering method, device and terminal |
| CN109871486A (en) * | 2019-02-18 | 2019-06-11 | 合肥工业大学 | The Product Requirement Analysis method and system of perceived social support under social media environment |
| CN109993576A (en) * | 2019-03-22 | 2019-07-09 | 中国质量认证中心 | A kind of method and system of quality of service aware, data acquisition process and analysis |
| CN110046246A (en) * | 2018-12-07 | 2019-07-23 | 阿里巴巴集团控股有限公司 | The analysis method and device of user's evaluation |
| CN110414811A (en) * | 2019-07-18 | 2019-11-05 | 合肥工业大学 | Generate the product Promotion Strategy acquisition methods and system of content online based on user |
| CN110457472A (en) * | 2019-07-16 | 2019-11-15 | 天津大学 | The emotion association analysis method for electric business product review based on SOM clustering algorithm |
| CN110458615A (en) * | 2019-07-31 | 2019-11-15 | 沃民高新科技（北京）股份有限公司 | Pass through the method and device of internet information assessment customer satisfaction |
| CN110555109A (en) * | 2019-06-28 | 2019-12-10 | 西安理工大学 | multi-granularity viewpoint mining method based on personal computer electronic commerce comments |
| CN110795537A (en) * | 2019-10-30 | 2020-02-14 | 秒针信息技术有限公司 | Method, device, equipment and medium for determining improvement strategy of target commodity |
| CN111523914A (en) * | 2019-01-17 | 2020-08-11 | 阿里巴巴集团控股有限公司 | User satisfaction evaluation method, device and system and data display platform |
| CN111523923A (en) * | 2020-04-06 | 2020-08-11 | 北京三快在线科技有限公司 | Merchant comment management system, method, server and storage medium |
| CN111639184A (en) * | 2020-06-01 | 2020-09-08 | 复旦大学 | Detection system for tendency inconsistency of scores and comment contents |
| CN111914086A (en) * | 2020-07-07 | 2020-11-10 | 广西科技大学 | Method and system for analyzing mobile phone comments based on LSTM neural network |
| CN112115703A (en) * | 2020-09-03 | 2020-12-22 | 腾讯科技（深圳）有限公司 | Article evaluation method and device |
| CN112183053A (en) * | 2020-10-10 | 2021-01-05 | 湖南快乐阳光互动娱乐传媒有限公司 | Data processing method and device |
| CN112598427A (en) * | 2020-12-14 | 2021-04-02 | 珠海格力电器股份有限公司 | After-sale service method and device, electronic equipment and storage medium |
| CN112651768A (en) * | 2020-12-04 | 2021-04-13 | 苏州黑云智能科技有限公司 | E-commerce analysis method and system based on block chain |
| CN113139060A (en) * | 2021-05-13 | 2021-07-20 | 杭州网易再顾科技有限公司 | Text analysis model training and text analysis method, medium, device and equipment |
| CN113792145A (en) * | 2021-09-16 | 2021-12-14 | 平安科技（深圳）有限公司 | Method and device for determining object attribute parameters, terminal equipment and storage medium |
| CN113821627A (en) * | 2020-06-17 | 2021-12-21 | 北京瀚海云星科技有限公司 | Neural network, method, device and program for predicting IP work index |
| CN113821626A (en) * | 2020-06-17 | 2021-12-21 | 北京瀚海云星科技有限公司 | Neural network and method for predicting IP work index according to multi-source social short text |
| CN113869065A (en) * | 2021-10-15 | 2021-12-31 | 梧州学院 | Emotion classification method and system based on 'word-phrase' attention mechanism |
| CN114066117A (en) * | 2020-08-05 | 2022-02-18 | 四川大学 | Park multi-scale evaluation method based on comment text |
| US11347944B2 (en) | 2020-01-27 | 2022-05-31 | Walmart Apollo, Llc | Systems and methods for short text identification |
| CN114584841A (en) * | 2022-03-02 | 2022-06-03 | 湖南快乐阳光互动娱乐传媒有限公司 | Comment content generation method and device, storage medium and electronic equipment |
| CN115587178A (en) * | 2022-09-08 | 2023-01-10 | 上海网商电子商务有限公司 | Automobile comment analysis method |
| WO2023015715A1 (en) * | 2021-08-12 | 2023-02-16 | 惠州Tcl云创科技有限公司 | User-comment-based data processing method and apparatus, and device and storage medium |
| CN116433336A (en) * | 2023-06-13 | 2023-07-14 | 北京易享信息技术有限公司 | User experience evaluating method and system based on multiple databases |
| CN116738298A (en) * | 2023-08-16 | 2023-09-12 | 杭州同花顺数据开发有限公司 | Text classification method, system and storage medium |
| CN117788036A (en) * | 2023-12-29 | 2024-03-29 | 广州伯威逊科技有限公司 | Marketing feedback-based production plan management method and system |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105550269A (en) * | 2015-12-10 | 2016-05-04 | 复旦大学 | Product comment analyzing method and system with learning supervising function |
| CN107025284A (en) * | 2017-04-06 | 2017-08-08 | 中南大学 | The recognition methods of network comment text emotion tendency and convolutional neural networks model |
| CN107451116A (en) * | 2017-07-14 | 2017-12-08 | 中国地质大学（武汉） | Raw big data statistical analysis technique in a kind of Mobile solution |
-
2017
- 2017-12-19 CN CN201711372805.9A patent/CN108446813A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105550269A (en) * | 2015-12-10 | 2016-05-04 | 复旦大学 | Product comment analyzing method and system with learning supervising function |
| CN107025284A (en) * | 2017-04-06 | 2017-08-08 | 中南大学 | The recognition methods of network comment text emotion tendency and convolutional neural networks model |
| CN107451116A (en) * | 2017-07-14 | 2017-12-08 | 中国地质大学（武汉） | Raw big data statistical analysis technique in a kind of Mobile solution |
Cited By (42)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109165387A (en) * | 2018-09-20 | 2019-01-08 | 南京信息工程大学 | A kind of Chinese comment sentiment analysis method based on GRU neural network |
| CN109359246A (en) * | 2018-12-07 | 2019-02-19 | 上海宏原信息科技有限公司 | A kind of brand cohesion calculation method based on forum user speech |
| CN110046246A (en) * | 2018-12-07 | 2019-07-23 | 阿里巴巴集团控股有限公司 | The analysis method and device of user's evaluation |
| CN109461037A (en) * | 2018-12-17 | 2019-03-12 | 北京百度网讯科技有限公司 | Comment on viewpoint clustering method, device and terminal |
| CN109461037B (en) * | 2018-12-17 | 2022-10-28 | 北京百度网讯科技有限公司 | Comment viewpoint clustering method and device and terminal |
| CN111523914A (en) * | 2019-01-17 | 2020-08-11 | 阿里巴巴集团控股有限公司 | User satisfaction evaluation method, device and system and data display platform |
| CN111523914B (en) * | 2019-01-17 | 2023-05-02 | 阿里巴巴集团控股有限公司 | User satisfaction evaluation method, device and system and data display platform |
| CN109871486A (en) * | 2019-02-18 | 2019-06-11 | 合肥工业大学 | The Product Requirement Analysis method and system of perceived social support under social media environment |
| CN109871486B (en) * | 2019-02-18 | 2021-04-06 | 合肥工业大学 | Product demand analysis method and system for market-ahead under social media environment |
| CN109993576A (en) * | 2019-03-22 | 2019-07-09 | 中国质量认证中心 | A kind of method and system of quality of service aware, data acquisition process and analysis |
| CN109993576B (en) * | 2019-03-22 | 2022-08-16 | 中国质量认证中心 | Method and system for sensing service quality, acquiring, processing and analyzing data |
| CN110555109A (en) * | 2019-06-28 | 2019-12-10 | 西安理工大学 | multi-granularity viewpoint mining method based on personal computer electronic commerce comments |
| CN110457472A (en) * | 2019-07-16 | 2019-11-15 | 天津大学 | The emotion association analysis method for electric business product review based on SOM clustering algorithm |
| CN110414811A (en) * | 2019-07-18 | 2019-11-05 | 合肥工业大学 | Generate the product Promotion Strategy acquisition methods and system of content online based on user |
| CN110458615A (en) * | 2019-07-31 | 2019-11-15 | 沃民高新科技（北京）股份有限公司 | Pass through the method and device of internet information assessment customer satisfaction |
| CN110795537A (en) * | 2019-10-30 | 2020-02-14 | 秒针信息技术有限公司 | Method, device, equipment and medium for determining improvement strategy of target commodity |
| CN110795537B (en) * | 2019-10-30 | 2022-10-25 | 秒针信息技术有限公司 | Method, device, equipment and medium for determining improvement strategy of target commodity |
| US11347944B2 (en) | 2020-01-27 | 2022-05-31 | Walmart Apollo, Llc | Systems and methods for short text identification |
| CN111523923B (en) * | 2020-04-06 | 2023-09-29 | 北京三快在线科技有限公司 | Merchant comment management system, merchant comment management method, merchant comment management server and storage medium |
| CN111523923A (en) * | 2020-04-06 | 2020-08-11 | 北京三快在线科技有限公司 | Merchant comment management system, method, server and storage medium |
| CN111639184A (en) * | 2020-06-01 | 2020-09-08 | 复旦大学 | Detection system for tendency inconsistency of scores and comment contents |
| CN113821627A (en) * | 2020-06-17 | 2021-12-21 | 北京瀚海云星科技有限公司 | Neural network, method, device and program for predicting IP work index |
| CN113821626A (en) * | 2020-06-17 | 2021-12-21 | 北京瀚海云星科技有限公司 | Neural network and method for predicting IP work index according to multi-source social short text |
| CN111914086A (en) * | 2020-07-07 | 2020-11-10 | 广西科技大学 | Method and system for analyzing mobile phone comments based on LSTM neural network |
| CN114066117A (en) * | 2020-08-05 | 2022-02-18 | 四川大学 | Park multi-scale evaluation method based on comment text |
| CN114066117B (en) * | 2020-08-05 | 2023-04-07 | 四川大学 | Park multi-scale evaluation method based on comment text |
| CN112115703B (en) * | 2020-09-03 | 2023-10-17 | 腾讯科技（深圳）有限公司 | Article evaluation method and device |
| CN112115703A (en) * | 2020-09-03 | 2020-12-22 | 腾讯科技（深圳）有限公司 | Article evaluation method and device |
| CN112183053A (en) * | 2020-10-10 | 2021-01-05 | 湖南快乐阳光互动娱乐传媒有限公司 | Data processing method and device |
| CN112651768A (en) * | 2020-12-04 | 2021-04-13 | 苏州黑云智能科技有限公司 | E-commerce analysis method and system based on block chain |
| CN112598427A (en) * | 2020-12-14 | 2021-04-02 | 珠海格力电器股份有限公司 | After-sale service method and device, electronic equipment and storage medium |
| CN113139060A (en) * | 2021-05-13 | 2021-07-20 | 杭州网易再顾科技有限公司 | Text analysis model training and text analysis method, medium, device and equipment |
| WO2023015715A1 (en) * | 2021-08-12 | 2023-02-16 | 惠州Tcl云创科技有限公司 | User-comment-based data processing method and apparatus, and device and storage medium |
| CN113792145A (en) * | 2021-09-16 | 2021-12-14 | 平安科技（深圳）有限公司 | Method and device for determining object attribute parameters, terminal equipment and storage medium |
| CN113869065A (en) * | 2021-10-15 | 2021-12-31 | 梧州学院 | Emotion classification method and system based on 'word-phrase' attention mechanism |
| CN113869065B (en) * | 2021-10-15 | 2024-04-12 | 梧州学院 | Emotion classification method and system based on 'word-phrase' attention mechanism |
| CN114584841A (en) * | 2022-03-02 | 2022-06-03 | 湖南快乐阳光互动娱乐传媒有限公司 | Comment content generation method and device, storage medium and electronic equipment |
| CN115587178A (en) * | 2022-09-08 | 2023-01-10 | 上海网商电子商务有限公司 | Automobile comment analysis method |
| CN116433336A (en) * | 2023-06-13 | 2023-07-14 | 北京易享信息技术有限公司 | User experience evaluating method and system based on multiple databases |
| CN116738298A (en) * | 2023-08-16 | 2023-09-12 | 杭州同花顺数据开发有限公司 | Text classification method, system and storage medium |
| CN116738298B (en) * | 2023-08-16 | 2023-11-24 | 杭州同花顺数据开发有限公司 | Text classification method, system and storage medium |
| CN117788036A (en) * | 2023-12-29 | 2024-03-29 | 广州伯威逊科技有限公司 | Marketing feedback-based production plan management method and system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108446813A (en) | A kind of method of electric business service quality overall merit | |
| Colladon | The semantic brand score | |
| CN107491531B (en) | Chinese network comment sensibility classification method based on integrated study frame | |
| Cvijikj et al. | Understanding social media marketing: a case study on topics, categories and sentiment on a Facebook brand page | |
| CN108491377A (en) | A kind of electric business product comprehensive score method based on multi-dimension information fusion | |
| CN110134931A (en) | Media title generation method, device, electronic equipment and readable medium | |
| CN108985347A (en) | Training method, the method and device of shop classification of disaggregated model | |
| WO2007017862A2 (en) | Method and system for extracting web data | |
| CN109582875A (en) | A kind of personalized recommendation method and system of online medical education resource | |
| WO2013059487A1 (en) | System and methods for automatically detecting deceptive content | |
| CN104050243B (en) | It is a kind of to search for the network search method combined with social activity and its system | |
| Kumar et al. | Analysis of various machine learning algorithms for enhanced opinion mining using twitter data streams | |
| CN104951807B (en) | The determination method and apparatus of stock market's mood | |
| Zamani et al. | Sentiment analysis: determining people’s emotions in Facebook | |
| Herrmann et al. | # fridaysforfuture–What does Instagram tell us about a social movement? | |
| Awrahman et al. | Sentiment analysis and opinion mining within social networks using konstanz information miner | |
| Husnain et al. | Estimating market trends by clustering social media reviews | |
| CN104951434B (en) | The determination method and apparatus of brand mood | |
| Gyódi et al. | Informing policy with text mining: technological change and social challenges | |
| Li et al. | Capturing and analyzing e-WOM for travel products: A method based on sentiment analysis and stochastic dominance | |
| Harguem et al. | Machine Learning Based Prediction of Stock Exchange on NASDAQ 100: A Twitter Mining Approach | |
| Cao et al. | Big data in marketing & retailing | |
| Alsubari et al. | Rule-based classifiers for identifying fake reviews in e-commerce: a deep learning system | |
| Bertram et al. | How could we start to develop indicators for evidence-informed policy making in public health and health promotion | |
| Tsao et al. | From mining to meaning: How B2B marketers can leverage text to inform strategy |
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: 20180824 |
|
| RJ01 | Rejection of invention patent application after publication |