CN108197274A - Abnormal individual character detection method and device based on dialogue - Google Patents
Abnormal individual character detection method and device based on dialogue Download PDFInfo
- Publication number
- CN108197274A CN108197274A CN201810015868.7A CN201810015868A CN108197274A CN 108197274 A CN108197274 A CN 108197274A CN 201810015868 A CN201810015868 A CN 201810015868A CN 108197274 A CN108197274 A CN 108197274A
- Authority
- CN
- China
- Prior art keywords
- tensor
- emotion
- transference
- public
- user
- 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.)
- Granted
Links
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
- G06F16/33—Querying
- G06F16/332—Query formulation
- G06F16/3329—Natural language query formulation or dialogue systems
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Mathematical Physics (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Computation (AREA)
- Evolutionary Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Human Computer Interaction (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computational Linguistics (AREA)
- Databases & Information Systems (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Machine Translation (AREA)
Abstract
The present invention provides a kind of abnormal individual character detection method and device based on dialogue.This method includes：Obtain the first preset quantity dialogue data in social media；Emotion recognition is carried out to the first preset quantity dialogue data using support vector machines, obtains public's emotion under the second preset quantity kind emotion and each user feeling；Public's emotion under the second preset quantity kind emotion and each user feeling are marked respectively；The transference tensor of the public and each user are determined based on the label of public's emotion under the second preset quantity kind emotion and each user feeling；Probability statistics are carried out to the transference of the second preset quantity item dialogue, obtain the transference tensor of the public and each user；According to the transference tensor computation tensor similarity of each user of transference tensor sum of the public；Abnormal user individual is determined according to the tensor similarity and similarity threshold.As it can be seen that the present embodiment can quickly determine abnormal user individual.
Description
Technical field
The present invention relates to technical field of data processing more particularly to a kind of abnormal individual character detection methods and dress based on dialogue
It puts.
Background technology
Dialog history data mainly include following scheme in the relevant technologies：
First by collecting dialog history data and being labeled, abnormality detection model is carried out using the data marked
Training, is carried out abnormality detection using trained abnormality detection model when receiving real time conversational data and obtains result；Wherein
Abnormality detection model training is referred to:By the way that abnormality detection problem is attributed to two classification problems, (I representatives have exception, and O representatives do not have
Have exception), and utilize machine learning model, such as one two grader of support vector machines (SVM) or neural metwork training, i.e.,
Abnormality detection model.But the above method has ignored the factors such as the difference of each user and the difference of environmental stimuli factor, only
Model is trained according to historical session language material, causes to detect abnormal inaccurate.In addition, testing result is only including abnormal and non-different
Often, it is a kind of rough qualitative description rather than quantitative description.
Invention content
An embodiment of the present invention provides a kind of abnormal individual character detection method and device based on dialogue, to solve the relevant technologies
In deficiency.
In a first aspect, the present invention provides a kind of abnormal individual character detection method based on dialogue, the method includes：
Obtain the first preset quantity dialogue data in social media；
Emotion recognition is carried out to the first preset quantity dialogue data using support vector machines, it is pre- to obtain second
If quantity kind emotion；
Public's emotion under the second preset quantity kind emotion and each user feeling are marked respectively；
Probability statistics are carried out to the transference of the second preset quantity item dialogue, obtain the feelings of the public and each user
Feel transport tensor；
According to the transference tensor computation tensor similarity of each user of transference tensor sum of the public；
Abnormal user individual is determined according to the tensor similarity and similarity threshold.
Optionally, the second preset quantity kind emotion includes neutral, happy, surprised, sad and angry.
Optionally, the transference tensor is represented using following form：
{ initial emotion stimulates emotion, generates emotion, transition probability }.
Optionally, the tensor phase of the transference tensor of each user of transference tensor sum of the public is calculated using following formula
Like degree：
In formula, tensor A and B represent the transference tensor of each user of transference tensor sum of the public respectively；Am::
And Bn::It is tensor A and the piece matrix of B, Sim (Am::, Bn::) it is the similarity of tensor A and B on lower dimensional space.
Second aspect, the present invention provides a kind of abnormal individual character detection device based on dialogue, described device includes：
Dialogue data acquisition module, for obtaining the first preset quantity dialogue data in social media；
Emotion recognition module, for using support vector machines to the first preset quantity dialogue data into market
Perception is other, obtains the second preset quantity kind emotion；
Dialogue data mark module, for marking public's emotion under the second preset quantity kind emotion and each respectively
User feeling；
Transport tensor determining module, the transference for talking with to the second preset quantity item carry out probability statistics,
Obtain the transference tensor of each public and each user；
Similarity calculation module, for the transference tensor computation of each user of transference tensor sum according to the public
Tensor similarity；
Abnormal individuals determining module, for determining abnormal user individual according to the tensor similarity and similarity threshold.
Optionally, the second preset quantity kind emotion includes neutral, happy, surprised, sad and angry.
Optionally, the transference tensor is represented using following form：
{ initial emotion stimulates emotion, generates emotion, probability value }.
Optionally, the similarity calculation module calculates the feelings of each user of transference tensor sum of the public using following formula
Feel the tensor similarity of transport tensor：
In formula, tensor A and B represent the transference tensor of each user of transference tensor sum of the public respectively；Am::
And Bn::It is tensor A and the piece matrix of B, Sim (Am::, Bn::) it is the similarity of tensor A and B on lower dimensional space.
As shown from the above technical solution, it is default to described first first with support vector machines in the embodiment of the present invention
Quantity dialogue data carries out emotion recognition, obtains the second preset quantity kind emotion；Then second present count is marked respectively
Public's emotion and each user feeling under amount kind of emotion, later, to the transference of the second preset quantity item dialogue into
Row probability statistics obtain the transference tensor of each public and each user；Continue the transference tensor sum according to the public
The transference tensor computation tensor similarity of each user；It is finally determined according to the tensor similarity and similarity threshold different
Common family individual.As it can be seen that obtain the transference tensor of the public and user's individual in the present embodiment, by calculate each user with
The tensor similarity of the public may thereby determine that abnormal user individual, and model is simple, and calculating speed is fast.
Description of the drawings
In order to illustrate more clearly about the embodiment of the present invention or technical scheme of the prior art, to embodiment or will show below
There is attached drawing needed in technology description to be briefly described, it should be apparent that, the accompanying drawings in the following description is only this
Some embodiments of invention, for those of ordinary skill in the art, without creative efforts, can be with
Other attached drawings are obtained according to these figures.
Fig. 1 is the method flow schematic diagram of the abnormal individual character detection method based on dialogue that one embodiment of the invention provides；
Fig. 2 is the flow diagram for carrying out emotional semantic classification in one embodiment of the invention to dialogue data using SVM；
Fig. 3 is the schematic diagram that emotion stimulates tensor model in one embodiment of the invention；
Fig. 4 is the schematic diagram of transference tensor model in one embodiment of the invention；
Fig. 5 is the schematic diagram of film lines in one embodiment of the invention；
Fig. 6 is the block diagram of the abnormal individual character detection device based on dialogue that one embodiment of the invention provides.
Specific embodiment
Below in conjunction with the attached drawing in the embodiment of the present invention, the technical solution in the embodiment of the present invention is carried out clear, complete
Site preparation describes, it is clear that described embodiment is only part of the embodiment of the present invention, instead of all the embodiments.It is based on
Embodiment in the present invention, those of ordinary skill in the art are obtained every other without making creative work
Embodiment shall fall within the protection scope of the present invention.
Fig. 1 is the method flow schematic diagram of the abnormal individual character detection method based on dialogue that one embodiment of the invention provides.
Referring to Fig. 1, this method includes：
101, obtain the first preset quantity dialogue data in social media；
102, emotion recognition is carried out to the first preset quantity dialogue data using support vector machines, obtains the
Public's emotion and each user feeling under two preset quantity kind emotions；
103, public's emotion under the second preset quantity kind emotion and each user feeling are marked respectively；
104, probability statistics are carried out to the transference of the second preset quantity item dialogue, obtain the public and each user
Transference tensor；
105, according to the transference tensor computation tensor similarity of each user of transference tensor sum of the public；
106, abnormal user individual is determined according to the tensor similarity and similarity threshold.
As shown from the above technical solution, in the present embodiment public's emotion is obtained by obtaining dialogue data progress emotion recognition
Classification and user feeling classification, then obtain public's emotion tensor sum user feeling tensor, later by calculating public's emotion
The tensor similarity of amount and user feeling tensor obtains abnormal user individual.As it can be seen that pass through public's emotion tensor in the present embodiment
Can have with user feeling tensor to the variation of user personality and more comprehensively and more accurately describe, and tensor similarity can be with
Quantitatively detect abnormal user.
Abnormal each step of individual character detection method provided in an embodiment of the present invention is made with reference to the accompanying drawings and examples detailed
Description.
First, the step of introducing 101, obtaining the first preset quantity dialogue data in social media.
Above-mentioned social media can be microblogging, Twitter, QQ space and wechat space etc. and TV, website etc..
Above-mentioned first preset quantity can be 1000,10000,100000 etc., and those skilled in the art can be according to specific
Scene is configured, and is not limited thereto.
Above-mentioned dialogue data includes question sentence and answer.For example, question sentence " you have had a meal ", answer " I ate ",
This is for a dialogue data.
In the embodiment of the present invention, first is obtained from social medias such as TV, websites using the crawler technology in the relevant technologies
Preset quantity dialogue data.Wherein dialogue data can be with TV play lines, film lines or different language session website
Dialogue etc..
As it can be seen that obtaining dialogue data using crawler technology in the present embodiment has randomness, in addition, the public of social media
Also not for specific user, the dialogue data obtained in this way more can be by the uncertainty of the transference of people for dialogue or lines
It represents in introducing method, facilitates machine learning.In addition, the dialogue data of specific user can also be obtained in the present embodiment, more
Targetedly.
Secondly, 102 are introduced, emotion knowledge is carried out to the first preset quantity dialogue data using support vector machines
Not, it obtains public's emotion under the second preset quantity kind emotion and each user feeling is rapid.
Above-mentioned second preset quantity can be 3,4,5 even more, and those skilled in the art can be according to specifically being set
It puts.In one embodiment, the second preset quantity is 5 kinds, i.e. the second preset quantity kind emotion can be neutral, happy, surprised, wound
The heart and anger.
In one embodiment of the invention using support vector machines (SVM) to above-mentioned first preset quantity dialogue data into market
Perception is other, i.e., every dialogue data corresponds to neutral, happy, surprised, sad or angry.SVM carries out emotion point to dialogue data
The detailed process of class is shown in Fig. 2.The dialogue data of 5 class labels is subjected to text vector, feature selecting is carried out, then calculates each spy
The weight (TF*IDF) of sign, finally carries out model training and prediction obtains the classification results of dialogue data.
It will be appreciated that 5 groups i.e. 5 kind public's feelings are integrally divided into the first preset quantity dialogue data according to 5 kinds of emotions
Sense.Then for each user, it is divided into 5 groups i.e. 5 kind user feeling to the dialogue data of the user according to 5 kinds of emotions.
As it can be seen that the accuracy of emotional semantic classification is may insure in the embodiment of the present invention by support vector machines.
Then, 103 are introduced, marks public's emotion under the second preset quantity kind emotion and each user feeling respectively
The step of.
It is calculated for convenience of subsequent quantitation, substitutes above-mentioned 5 kinds of feelings using label " 0,1,2,3,4 " in an embodiment of the present invention
Sense, i.e., mark " neutral, happy, surprised, sad and angry " respectively using label " 0,1,2,3,4 ".
In the present embodiment, 5 kinds of public's emotions and 5 kinds of user feelings are marked respectively.
4th, introduce 104, to the second preset quantity dialogue data carry out probability statistics, obtain each public or
The transference tensor of each user.
Above-mentioned transference tensor can include three variables and a probability, wherein three variables include initial emotion,
Stimulate emotion and generation emotion.I.e. in the case where initial emotion is certain emotion, emotion is stimulated to be certain emotion, emotion is generated
Probability for certain emotion.Wherein three variables can be understood as an emotion path.
Such as：
{ neutral, neutral, neutral, probability }, i.e., in the case where initial emotion is neutral, emotion is stimulated to be neutrality, generation
Emotion is neutral probability.
{ neutral, neutral, sad, probability }, i.e., in the case where initial emotion is neutral, emotion is stimulated to be neutrality, generation
Emotion is sad probability.
……
In the present embodiment, determine the emotion probability of initial emotion, stimulate the emotion probability of emotion, then can calculate birth
Into the emotion probability of emotion.Probability statistics are carried out to the first data strip dialogue data in one embodiment, then by each number of sessions
According to initial emotion, stimulation emotion and generation emotion is used as successively, second second present count of preset quantity * may finally be obtained
Measure the transference probability tables of the * the second preset quantity.In one embodiment, if the second preset quantity is 5, transference probability
The size of table is 5*5*5.
In the present embodiment by user or the public in the case of initial emotion, by environment stimulation emotion when,
The probability of emotion is generated, so as to consider that the uncertain of the transference of people introduces, i.e., in view of of different user
Property and extraneous stimulus, so as to improve the accuracy of testing result, avoid qualitative description there are the defects of.
In the present embodiment, the emotion in transference tensor is substituted using label, so as to obtain the emotion of digital form
Transport tensor.I.e.：
{ 0,0,0,0.04534 }, i.e., in the case where initial emotion is neutral (0), stimulation emotion is neutral (0), generation
Emotion is that neutral (0) is 0.04534.
{ 0,0,3,0.00234 }, i.e., in the case where initial emotion is neutral (0), stimulation emotion is neutral (0), generation
Emotion is that neutral (3) are 0.00234.
Fig. 3 is emotion stimulation tensor model proposed by the present invention, and the left side is emotion stimulation model, and the emotion in session is not
Disconnected migration, it will be assumed that initial emotion is happy, gives a stimulation emotion, happy class emotion can migrate, and migrate
Happily, surprised to neutrality, sad, angry class emotion is possible to, and only probability is different；It is mapped to the transference on the right
Amount, since initial emotion has 5 kinds, stimulation emotion also has 5 kinds, then the emotion of generation also has 5 kinds, and corresponding is 5*5*5=125
Kind probability value, transference path and corresponding probability value spatially constitute a transference tensor, the institute on the right of Fig. 3
Show, wherein N, H, D, S, A is represented respectively：Neutrality, it is happily, surprised, it is sad, it is angry.
Fig. 4 is further with the careful tensor model for describing and being stimulated based on emotion：We are by initial emotion (initial
Emotion), stimulation emotion (stimulating emotion), generation emotion (transferred emotion) are mapped to three
In dimension space, three dimensions are corresponded to respectively, are represented respectively with i, s, t.And each transfer can all correspond to a Probability p=r
[i, s, t], transference path and p constitute the tensor [i, s, t, p] of three ranks.
The second preset quantity dialogue data is counted in the present embodiment, the probability of each emotion can be obtained.
In the dialogue of second preset quantity item, we have obtained being labelled with the emotion session of emotion, and only need to before session at this time
The transfer of affective tag is counted.Affective tag such as one section of dialogue (12) is followed successively by：[0,1,2,0,1,0,1,1,1,1,
1,1] it is so corresponding just to have transport tensor generation:[0,1,2,0.1],[1,2,0,0.1],[2,0,1,0.1],[0,1,0,
0.1],[1,0,1,0.1],[0,1,1,0.1],[1,1,10.4].Per initial emotion is in short both used as, lower a word is also served as
Stimulation emotion, share 10 conversation groups here.Then the probability occurred to the triple in each different path counts,
If [0,1,2] occurs 1 time, corresponding probability is 1/10=0.1, then [0,1,2,0.1] just constitutes a transference
Amount.
The dialogue data of each user is chosen in the present embodiment, is united according to all kinds of dialogue datas to each user
Meter, obtains the probability of each emotion.
5th, 105 are introduced, according to the transference tensor computation tensor phase of each user of transference tensor sum of the public
The step of seemingly spending.
In the present embodiment, the transference tensor computation tensor with reference to each user of transference tensor sum of the public is similar
Degree, formula are as follows：
Wherein, tensor A and B represent the transference tensor of each user of transference tensor sum of the public respectively；Am::
And Bn::It is tensor A and the piece matrix (slice matrix) of B, Sim (Am::, Bn::) it is tensor A and B on lower dimensional space
Similarity.
It should be noted that due to transference path difference, the similarity of higher dimensional space is directly calculated, result can be caused
Inaccuracy, so needing to carry out dimensionality reduction to former tensor.Dimension reduction method can refer to the relevant technologies, be not illustrated herein.
Finally, 106 are introduced, the step of abnormal user individual is determined according to the tensor similarity and similarity threshold.
In the present embodiment, tensor similarity between zero and one, closer to 1, illustrate user and normally talk with it is closer, this
User personality tends to popular.Tensor similarity illustrates that the emotion of user session is different from public sentiment, also illustrates closer to 0
The user relatively has individual character.In one embodiment, it is 0.2 to choose similarity threshold.That is, in the result of tensor similarity
In, the user data for finally having 20% is marked as exception.
As it can be seen that the transference tensor of the public and user's individual is obtained in the present embodiment, by calculating each user and public affairs
Many tensor similarities may thereby determine that abnormal user individual, and model is simple, and calculating speed is fast.
Embodiment
In the present embodiment, with English movies《Before Sunset》Talk with abnormality detection.First, platform in the film is collected
Word.Subdialogue data are as shown in Figure 5.Then emotional semantic classification is carried out to lines using SVM, it is right using label " 0,1,2,3,4 "
Every dialogue data emotional category marks " neutral, happy, surprised, sad and angry " respectively.
The transference probability tables of affective tag and transference tensor structure film lines are then based on, as shown in table 1.
Table 1, user feeling transition probability table
Later, based on transference probability tables, digital transference tensor is built, as represented shown in 2.Normal tensor
Refer to the session transference tensor of the public residing for individual, and current tensor refers to the session transference tensor of individual.
Table 2, user feeling transport tensor (normal tensor and current tensor)
In the present embodiment, according to table 2, emotion path (Emotionalpath) and normal transition probability
(Normalprobability) normal tensor A is then constituted, then emotion path (Emotional path) and current transfer are general
Rate (Currentprobability) then constitutes current tensor B.
1000 film lines are divided into 10 groups, every one group of 100 conducts, then calculate each group and normal emotion turns
The tensor similarity of tensor is moved, if tensor similarity result is less than similarity threshold (0.2), user is marked and generates abnormal feelings
Thread, as shown in table 3.
Table 3, abnormal session segment mark
Referring to table 3, the 4th group and the 7th group session occur abnormal in the film lines.Verification result is shown in Table shown in 4 and table 5.
Table 4, abnormal session authentication (301-400)
Table 5, abnormal session authentication (601-700)
Fig. 6 is the block diagram of the abnormal individual character detection device based on dialogue that one embodiment of the invention provides.It, should referring to Fig. 6
Device includes：
Dialogue data acquisition module 601, for obtaining the first preset quantity dialogue data in social media；
Emotion recognition module 602, for being carried out using support vector machines to the first preset quantity dialogue data
Emotion recognition obtains the second preset quantity kind emotion；
Dialogue data mark module 603, for mark respectively public's emotion under the second preset quantity kind emotion and
Each user feeling；
Transport tensor determining module 604, for based on public's emotion under the second preset quantity kind emotion and each
The label of user feeling determines the transference tensor of the public and each user；
Similarity calculation module 605, for the transference tensor of each user of transference tensor sum according to the public
Calculate tensor similarity；
Abnormal individuals determining module 606, for determining abnormal user according to the tensor similarity and similarity threshold
Body.
Optionally, the second preset quantity kind emotion includes neutral, happy, surprised, sad and angry.
Optionally, the transference tensor is represented using following form：
{ initial emotion stimulates emotion, generates emotion, probability value }.
Optionally, the similarity calculation module calculates the feelings of each user of transference tensor sum of the public using following formula
Feel the tensor similarity of transport tensor：
In formula, tensor A and B represent the transference tensor of each user of transference tensor sum of the public respectively；Am::
And Bn::It is tensor A and the piece matrix of B, Sim (Am::, Bn::) it is the similarity of tensor A and B on lower dimensional space.
It should be noted that the abnormal individual character detection device provided in an embodiment of the present invention based on dialogue is with the above method
One-to-one relationship, the implementation detail of the above method are equally applicable to above device, and the embodiment of the present invention is no longer to above-mentioned system
System is described in detail.
In the specification of the present invention, numerous specific details are set forth.It is to be appreciated, however, that the embodiment of the present invention can be with
It puts into practice without these specific details.In some instances, well known method, structure and skill is not been shown in detail
Art, so as not to obscure the understanding of this description.
The above embodiments are only used to illustrate the technical solution of the present invention., rather than its limitations；Although with reference to aforementioned each reality
Example is applied the present invention is described in detail, it will be understood by those of ordinary skill in the art that：It still can be to aforementioned each
Technical solution recorded in embodiment modifies and either carries out equivalent replacement to which part or all technical features；And
These modifications are replaced, the range for various embodiments of the present invention technical solution that it does not separate the essence of the corresponding technical solution,
It should all cover in the claim of the present invention and the range of specification.
Claims (8)
1. a kind of abnormal individual character detection method based on dialogue, which is characterized in that the method includes：
Obtain the first preset quantity dialogue data in social media；
Emotion recognition is carried out to the first preset quantity dialogue data using support vector machines, obtains the second present count
Public's emotion and each user feeling under amount kind emotion；
Public's emotion under the second preset quantity kind emotion and each user feeling are marked respectively；
Probability statistics are carried out to the transference of the second preset quantity item dialogue, the emotion for obtaining the public and each user turns
Move tensor；
According to the transference tensor computation tensor similarity of each user of transference tensor sum of the public；
Abnormal user individual is determined according to the tensor similarity and similarity threshold.
2. public's transference according to claim 1 is distributed modeling method, which is characterized in that second preset quantity
Kind emotion includes neutral, happy, surprised, sad and angry.
3. public's transference according to claim 1 is distributed modeling method, which is characterized in that the transference tensor
It is represented using following form：
{ initial emotion stimulates emotion, generates emotion, transition probability }.
4. public's transference according to claim 1 is distributed modeling method, which is characterized in that calculates the public using following formula
The each user of transference tensor sum transference tensor tensor similarity：
In formula, tensor A and B represent the transference tensor of each user of transference tensor sum of the public respectively；Am::With
Bn::It is tensor A and the piece matrix of B, Sim (Am::, Bn::) it is the similarity of tensor A and B on lower dimensional space.
5. a kind of abnormal individual character detection device based on dialogue, which is characterized in that described device includes：
Dialogue data acquisition module, for obtaining the first preset quantity dialogue data in social media；
Emotion recognition module, for carrying out emotion knowledge to the first preset quantity dialogue data using support vector machines
Not, the second preset quantity kind emotion is obtained；
Dialogue data mark module, for marking public's emotion and each user under the second preset quantity kind emotion respectively
Emotion；
Transport tensor determining module, the transference for talking with to the second preset quantity item carry out probability statistics, obtain
The public or the transference tensor of each user；
Similarity calculation module, for the transference tensor computation tensor of each user of transference tensor sum according to the public
Similarity；
Abnormal individuals determining module, for determining abnormal user individual according to the tensor similarity and similarity threshold.
6. public's transference according to claim 5 is distributed model building device, which is characterized in that second preset quantity
Kind emotion includes neutral, happy, surprised, sad and angry.
7. public's transference according to claim 5 is distributed model building device, which is characterized in that the transference tensor
It is represented using following form：
{ initial emotion stimulates emotion, generates emotion, transition probability }.
8. public's transference according to claim 5 is distributed model building device, which is characterized in that the similarity calculation mould
Block calculates the tensor similarity of the transference tensor of each user of transference tensor sum of the public using following formula：
In formula, tensor A and B represent the transference tensor of each user of transference tensor sum of the public respectively；Am::With
Bn::It is tensor A and the piece matrix of B, Sim (Am::, Bn::) it is the similarity of tensor A and B on lower dimensional space.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810015868.7A CN108197274B (en) | 2018-01-08 | 2018-01-08 | Abnormal personality detection method and device based on conversation |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810015868.7A CN108197274B (en) | 2018-01-08 | 2018-01-08 | Abnormal personality detection method and device based on conversation |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108197274A true CN108197274A (en) | 2018-06-22 |
| CN108197274B CN108197274B (en) | 2020-10-09 |
Family
ID=62588622
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810015868.7A Active CN108197274B (en) | 2018-01-08 | 2018-01-08 | Abnormal personality detection method and device based on conversation |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108197274B (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109615077A (en) * | 2018-10-17 | 2019-04-12 | 合肥工业大学 | Affective state modeling and feeling shifting method and device based on dialogue |
| CN109871002A (en) * | 2019-03-06 | 2019-06-11 | 东方证券股份有限公司 | The identification of concurrent abnormality and positioning system based on the study of tensor label |
| CN111026884A (en) * | 2019-12-12 | 2020-04-17 | 南昌众荟智盈信息技术有限公司 | Dialog corpus generation method for improving quality and diversity of human-computer interaction dialog corpus |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20130124438A1 (en) * | 2011-11-11 | 2013-05-16 | Sung-young Lee | Method of recognizing patterns based on markov chain hidden conditional random field model |
| CN103617158A (en) * | 2013-12-17 | 2014-03-05 | 苏州大学张家港工业技术研究院 | Method for generating emotion abstract of dialogue text |
| CN104516873A (en) * | 2014-12-12 | 2015-04-15 | 北京智谷睿拓技术服务有限公司 | Method and device for building emotion model |
| CN105809186A (en) * | 2016-02-25 | 2016-07-27 | 中国科学院声学研究所 | Emotion classification method and system |
| CN106599196A (en) * | 2016-12-14 | 2017-04-26 | 竹间智能科技（上海）有限公司 | Artificial intelligence conversation method and system |
| CN106970703A (en) * | 2017-02-10 | 2017-07-21 | 南京威卡尔软件有限公司 | Multilayer affection computation method based on mood index |
| CN107169426A (en) * | 2017-04-27 | 2017-09-15 | 广东工业大学 | A kind of detection of crowd's abnormal feeling and localization method based on deep neural network |
-
2018
- 2018-01-08 CN CN201810015868.7A patent/CN108197274B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20130124438A1 (en) * | 2011-11-11 | 2013-05-16 | Sung-young Lee | Method of recognizing patterns based on markov chain hidden conditional random field model |
| CN103617158A (en) * | 2013-12-17 | 2014-03-05 | 苏州大学张家港工业技术研究院 | Method for generating emotion abstract of dialogue text |
| CN104516873A (en) * | 2014-12-12 | 2015-04-15 | 北京智谷睿拓技术服务有限公司 | Method and device for building emotion model |
| CN105809186A (en) * | 2016-02-25 | 2016-07-27 | 中国科学院声学研究所 | Emotion classification method and system |
| CN106599196A (en) * | 2016-12-14 | 2017-04-26 | 竹间智能科技（上海）有限公司 | Artificial intelligence conversation method and system |
| CN106970703A (en) * | 2017-02-10 | 2017-07-21 | 南京威卡尔软件有限公司 | Multilayer affection computation method based on mood index |
| CN107169426A (en) * | 2017-04-27 | 2017-09-15 | 广东工业大学 | A kind of detection of crowd's abnormal feeling and localization method based on deep neural network |
Non-Patent Citations (1)
| Title |
|---|
| XIAO SUN等: ""Detecting users" anomalous emotion using social media for business intelligence"", 《JOURNAL OF COMPUTATIONAL SCIENCE》 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109615077A (en) * | 2018-10-17 | 2019-04-12 | 合肥工业大学 | Affective state modeling and feeling shifting method and device based on dialogue |
| CN109871002A (en) * | 2019-03-06 | 2019-06-11 | 东方证券股份有限公司 | The identification of concurrent abnormality and positioning system based on the study of tensor label |
| CN111026884A (en) * | 2019-12-12 | 2020-04-17 | 南昌众荟智盈信息技术有限公司 | Dialog corpus generation method for improving quality and diversity of human-computer interaction dialog corpus |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108197274B (en) | 2020-10-09 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Mathews et al. | Semstyle: Learning to generate stylised image captions using unaligned text | |
| CN108984530A (en) | A kind of detection method and detection system of network sensitive content | |
| WO2022141861A1 (en) | Emotion classification method and apparatus, electronic device, and storage medium | |
| Kreutzer et al. | Reliability and learnability of human bandit feedback for sequence-to-sequence reinforcement learning | |
| CN109508375A (en) | A kind of social affective classification method based on multi-modal fusion | |
| CN106095834A (en) | Intelligent dialogue method and system based on topic | |
| CN107247702A (en) | A kind of text emotion analysis and processing method and system | |
| Hossain et al. | Text to image synthesis for improved image captioning | |
| CN107278310A (en) | Batch normalizes layer | |
| CN111046670B (en) | Entity and relationship combined extraction method based on drug case legal documents | |
| CN106445919A (en) | Sentiment classifying method and device | |
| CN105895087A (en) | Voice recognition method and apparatus | |
| CN109271493A (en) | A kind of language text processing method, device and storage medium | |
| CN107122349A (en) | A kind of feature word of text extracting method based on word2vec LDA models | |
| CN111666761A (en) | Fine-grained emotion analysis model training method and device | |
| Paik et al. | The world of an octopus: How reporting bias influences a language model's perception of color | |
| CN110134765A (en) | A kind of dining room user comment analysis system and method based on sentiment analysis | |
| CN108197274A (en) | Abnormal individual character detection method and device based on dialogue | |
| CN110096587A (en) | The fine granularity sentiment classification model of LSTM-CNN word insertion based on attention mechanism | |
| CN110490444A (en) | Mark method for allocating tasks, device, system and storage medium | |
| CN112988963A (en) | User intention prediction method, device, equipment and medium based on multi-process node | |
| CN115392237B (en) | Emotion analysis model training method, device, equipment and storage medium | |
| CN108090099A (en) | A kind of text handling method and device | |
| CN114387061A (en) | Product pushing method and device, electronic equipment and readable storage medium | |
| CN108985223A (en) | A kind of human motion recognition method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |