CN109657241A - User's speech semantic analysis of network-oriented live scene - Google Patents
User's speech semantic analysis of network-oriented live scene Download PDFInfo
- Publication number
- CN109657241A CN109657241A CN201811523923.XA CN201811523923A CN109657241A CN 109657241 A CN109657241 A CN 109657241A CN 201811523923 A CN201811523923 A CN 201811523923A CN 109657241 A CN109657241 A CN 109657241A
- Authority
- CN
- China
- Prior art keywords
- user
- semantic analysis
- network
- speech
- lstm
- 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
- 238000004458 analytical method Methods 0.000 title claims abstract description 54
- 239000013598 vector Substances 0.000 claims abstract description 29
- 238000012549 training Methods 0.000 claims abstract description 24
- 230000011218 segmentation Effects 0.000 claims abstract description 5
- 238000012545 processing Methods 0.000 claims abstract description 4
- 238000000034 method Methods 0.000 claims description 15
- 230000008569 process Effects 0.000 claims description 9
- 238000012360 testing method Methods 0.000 claims description 5
- 238000013461 design Methods 0.000 claims description 3
- 238000003062 neural network model Methods 0.000 claims description 3
- 238000012546 transfer Methods 0.000 claims description 3
- 238000013528 artificial neural network Methods 0.000 description 5
- 230000000694 effects Effects 0.000 description 4
- 210000002569 neuron Anatomy 0.000 description 4
- 230000006870 function Effects 0.000 description 3
- 230000004913 activation Effects 0.000 description 2
- 230000008901 benefit Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 239000011159 matrix material Substances 0.000 description 2
- 241000251468 Actinopterygii Species 0.000 description 1
- 208000025174 PANDAS Diseases 0.000 description 1
- 208000021155 Paediatric autoimmune neuropsychiatric disorders associated with streptococcal infection Diseases 0.000 description 1
- 240000000220 Panda oleosa Species 0.000 description 1
- 235000016496 Panda oleosa Nutrition 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 230000004069 differentiation Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000013467 fragmentation Methods 0.000 description 1
- 238000006062 fragmentation reaction Methods 0.000 description 1
- 238000011478 gradient descent method Methods 0.000 description 1
- 238000012417 linear regression Methods 0.000 description 1
- 238000007477 logistic regression Methods 0.000 description 1
- 230000006386 memory function Effects 0.000 description 1
- 230000000644 propagated effect Effects 0.000 description 1
- 239000000126 substance Substances 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/30—Semantic analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/20—Natural language analysis
- G06F40/279—Recognition of textual entities
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Abstract
Present invention discloses a kind of user's speech semantic analysis of network-oriented live scene, include the following steps: S1, obtain the content of user input area in network direct broadcasting platform direct broadcasting room in real time, and is pre-processed；S2, pretreated content is subjected to word segmentation processing, keeps word sequence sequence constant；S3, the term vector of each word is obtained according to corpus dictionary, and then the vector for obtaining user's input content indicates；S4, building and the training two-way RNN semantic analysis sorter model of LSTM type；S5, judge whether user's input content includes flame according to the model established in S4, it is normal if not including to send, it otherwise reminds user and forbids sending.The present invention can be automatically completed the semantic analysis to user's speech in the case where prosthetic participates in, exercise supervision from source to user's speech in network direct broadcasting platform, can not only guarantee the real-time and validity of supervision, and also save cost of labor.
Description
Technical field
It is straight in particular to a kind of network-oriented the present invention relates to a kind of semantic analysis for user's speech
The user's speech semantic analysis for broadcasting scene, belongs to deep learning and content of text technical field.
Background technique
In recent years, network direct broadcasting industry development is extremely rapid, and various live streaming platforms emerge one after another, and the user for watching live streaming is quasi-
It is low to enter threshold, the situation that becomes younger integrally is presented in user group.And a distinguishing feature of network direct broadcasting platform is exactly that user can
Arbitrarily to deliver oneself speech in comment area, while these speeches can be shown in direct broadcasting room in the form of barrage, make to be in
All users (including main broadcaster) of the same direct broadcasting room can see.
But during actual platform operation, it has been found that, the not high user of some personal quality is always had, is
Pursuit low taste delivers indecency, unsound speech in barrage upper outlet at dirty.Due to the instantaneity of barrage and instantaneous
Property feature, speech one go out become accomplished fact, relevant measure of control does not often have the effect of any substance；While by
In the crypticity of barrage, some vulgar barrages cannot be found and be handled in time, this all causes pole to teenager, to society
Big harm.
Currently, network direct broadcasting platform also rests on the supervision of user's speech the simple shielding to barrage words mostly
On, semantic analysis is not realized really, and supervision effect is not satisfactory.Since barrage has changeableization, diversification, fragmentation
Etc. characteristics so that some do not include common sensitive vocabulary but there is the speech of vulgar entertaining meaning cannot effectively handle；
Secondly, barrage one, which goes out owner in direct broadcasting room, to be seen, carrying out simply shielding to it can not solve the problems, such as from source, from
And it cannot be guaranteed the real-time of supervision.
In conclusion a kind of semantic analysis for user's speech how is proposed on the basis of existing technology, it is real
Now to effective supervision of user's speech in network direct broadcasting platform, also just become that those skilled in that art are urgently to be resolved to be asked
Topic.
Summary of the invention
In view of the prior art, there are drawbacks described above, and the purpose of the present invention is to propose to a kind of users of network-oriented live scene
Speech semantic analysis, includes the following steps:
S1, the content for obtaining user input area in network direct broadcasting platform direct broadcasting room in real time, and pre-processed；
S2, pretreated content is subjected to word segmentation processing, keeps word sequence sequence constant；
S3, the term vector of each word is obtained according to corpus dictionary, and then the vector for obtaining user's input content indicates；
S4, building and the training two-way RNN semantic analysis sorter model of LSTM type；
S5, judge whether user's input content includes flame according to the model established in S4, it is normal if not including
It sends, otherwise remind user and forbids sending.
Preferably, pretreatment described in S1 specifically includes: the removal redundant information unrelated with word content, the extra letter
Breath includes expression picture, emoticon, numerical chracter and phonetic symbol.
Preferably, it is specifically included described in S3 according to the term vector that corpus dictionary obtains each word: each vocabulary is shown as
One only hot vector, the dimension of vector are the length of corpus dictionary；The corpus dictionary is by the corpus shape on network direct broadcasting platform
At the words in the corpus dictionary does not repeat.
Preferably, S4 specifically comprises the following steps:
User's input content on S41, collection network live streaming platform in various types live streaming, and be marked one by one, it will
Content-label comprising flame is 0, is otherwise labeled as 1；
S42, user's input content of collection is pre-processed, divides training set and test set, construct corpus dictionary；
S43, the form that user inputs content of text term vector is showed according to corpus dictionary, it is suitable according to word sequence
Term vector is attached by sequence；
S44, the training data with label is input in the two-way RNN semantic analysis classifier of LSTM type and is trained,
Optimal neural network model parameter is obtained, the two-way RNN semantic analysis sorter model of LSTM type is finally obtained.
Preferably, S44 specifically comprises the following steps:
S441, the two-way RNN structure of design LSTM type, the building two-way RNN semantic analysis classifier of LSTM type, obtain LSTM type
Two-way RNN semantic analysis sorter model；
S442, training pattern parameter complete the training of RNN semantic analysis sorter model two-way to LSTM type.
Preferably, the two-way RNN semantic analysis classifier of LSTM type described in S44 include sequentially sequentially connected input layer,
Hidden layer and output layer；
The input of the input layer is the word sequence for representing content of text；
The hidden layer is connected by multiple LSTM units, mono- including the LSTM transmitted according to word sequence forward direction
The unit of member and the LSTM according to word sequence reverse transfer.
Preferably, the output layer is classifier, and the classifier is two classifiers.
Preferably, the LSTM unit is the ad hoc network gathered around there are three door, and three doors are by Sigmoid function control
System, can selectively control the transmitting of information flow, and three doors are respectively input gate, forget door and out gate.
Preferably, the flame includes vulgar information, pornography and violence information.
Compared with prior art, advantages of the present invention is mainly reflected in the following aspects:
The present invention can be automatically completed the semantic analysis to user's speech, from source in the case where prosthetic participates in
It exercises supervision to user's speech in network direct broadcasting platform, can not only guarantee the real-time and validity of supervision, and drop
Low monitor procedure relies on for manually-operated, has saved cost of labor.
The present invention carries out semantic analysis using the two-way RNN neural network structure of LSTM type, both refers in the analysis process
The accuracy of supervision is effectively promoted referring also to Future Information in historical information.Meanwhile the present invention can be to all live streamings
All users on platform carry out the supervision with real-time and validity, and supervision range is wide, strong applicability.
In addition, the present invention also provides reference for other relevant issues in same domain, can be opened up on this basis
Extension is stretched, and is applied in same domain in the technical solution of other users speech semantic analysis, has very wide application prospect.
Just attached drawing in conjunction with the embodiments below, the embodiment of the present invention is described in further detail, so that of the invention
Technical solution is more readily understood, grasps.
Detailed description of the invention
Fig. 1 is flow diagram of the invention；
Fig. 2 is the two-way RNN semantic analysis sorter model training flow chart of LSTM type in the present invention；
Fig. 3 is the two-way RNN semantic analysis sorter model structural block diagram in the present invention.
Specific embodiment
As shown in FIG. 1 to 3, present invention discloses a kind of user's speech semantic analysis sides of network-oriented live scene
Method includes the following steps:
S1, the content for obtaining user input area in network direct broadcasting platform direct broadcasting room in real time, and pre-processed.The pre- place
Reason is specially to remove the redundant information unrelated with word content, and the redundant information includes expression picture, emoticon, digit
Number and phonetic symbol etc..
S2, pretreated content is subjected to word segmentation processing, keeps word sequence sequence constant.
S3, the term vector of each word is obtained according to corpus dictionary, each vocabulary is shown as an only hot vector, the dimension of vector
Degree is the length of corpus dictionary.The corpus dictionary is formed by the corpus on network direct broadcasting platform, the word in the corpus dictionary
Word does not repeat.And then the vector for obtaining user's input content indicates.
S4, building and the training two-way RNN semantic analysis sorter model of LSTM type.This process is as shown in Fig. 2, specific packet
Include following steps:
User's input content on S41, collection network live streaming platform in various types live streaming, and be marked one by one.
Firstly, being collected using web crawlers various types of on major network direct broadcasting platform (such as bucket fish, war flag, panda, YY etc.)
User's speech in type live streaming (such as game is broadcast live, show field live streaming, live news etc.), while platform administrator according to previous warp
The some user's speeches tested under intercepting also are collected together, form corpus.User's sentence of collection is more, then corpus is completeer
It is standby.Later, it by these statements cleans and marks, shoots off identical sentence, will include vulgar, yellow, the flames such as violence
Speech is labeled as 0, conversely, the speech for not including these information is labeled as 1.
S42, user's input content of collection is pre-processed, divides training set and test set, construct corpus dictionary.
The pretreatment includes deleting some meaningless symbols in these speeches, such as expression picture, emoticon, number
Character number and phonetic, space etc. will retain punctuation mark to keep speech semantic complete.These speeches are divided into two later
Part, wherein 75% is used as training set, remaining is test set, either in test set or in training set all includes certain ratio
The positive negative sample of example, to prevent the lack of uniformity of sample from having an impact to classification results.Then, Stanford- is utilized
For segmenter segmenter to its Chinese word segmentation, this is a open source segmenter, and using simple, participle effect is good, sentence to be processed
Unduplicated words all in corpus is combined and is formed language by the words sequence that input model is formd by participle
Expect dictionary.
S43, the form that user inputs content of text term vector is showed according to corpus dictionary, it is suitable according to word sequence
Term vector is attached by sequence.
Each word in words sequence can be indicated that the length of vector is word with a very long vector according to dictionary
The length of allusion quotation, each word are a feature in this feature vector.It include 10 words in dictionary, then if there is a dictionary
One word just needs to be indicated with 10 dimensional vectors, such as x (' beautiful ')=[0,1,0,0,0,0,0,0,0,0],
X (' generous ')=[0,0,1,0,0,0,0,0,0,0] can digitize every user's speech with this method,
It is transformed into term vector sequence, input model is facilitated to be analyzed.
S44, the training data with label is input in the two-way RNN semantic analysis classifier of LSTM type and is trained,
Optimal neural network model parameter is obtained, the two-way RNN semantic analysis sorter model of LSTM type is finally obtained.
The two-way RNN semantic analysis classifier of LSTM type includes sequentially sequentially connected input layer, hidden layer and output
Layer.
The input of the input layer is the word sequence for representing content of text.The hidden layer is connected by multiple LSTM units
It forms, including the unit of the LSTM unit and the LSTM according to word sequence reverse transfer that are transmitted according to word sequence forward direction, uses
In the semantic feature for extracting input text, semantic analysis is based on context carried out.The output layer is classifier, for according to institute
State hidden layer semantic analysis result carry out discriminant classification, both relied on during its differentiation text historical information or
The Future Information of text is relied on, so that differentiating that result is more accurate reasonable.The classifier is two classifiers, commonly
Strong classifier, as SVM, Logistic regression effect are all good.
The LSTM unit is the ad hoc network gathered around there are three door, and three doors are controlled by Sigmoid function, can be had
The transmitting of the control information flow of selectivity, three doors are respectively input gate, forget door and out gate.
The input gate is for controlling that how many information can flow into memory unit；The forgetting door is for controlling a period of time
Current memory unit can be flowed by carving how many information in memory unit；The out gate has in current memory unit for controlling
How much information can flow into current hidden state.
S44 specifically comprises the following steps:
S441, the two-way RNN structure of design LSTM type, the building two-way RNN semantic analysis classifier of LSTM type, obtain LSTM type
Two-way RNN semantic analysis sorter model.
Two-way RNN semantic analysis classifier structure in the present invention is as shown in figure 3, by the word sequence of vectorization respectively by head
It is input in two-way RNN neural network to tail and by tail to head, each LSTM unit of hidden layer has memory function, can be same
When receive the data from input layer and previous LSTM unit, they are sequentially connected, the last one LSTM unit then include with
Preceding all data informations differentiate so only the output of the last one LSTM unit of hidden layer need to be transmitted to classifier,
Classification results had both relied on former information and have been also relied on Future Information.Wherein three doors of hidden layer LSTM unit assign mind
Through first judgment, control force and memory, specific formula for calculation is as follows.
Input gate: It=σ (WI·[ht-1,xt]+BI)
Forget door: Ft=σ (WF·[ht-1,xt]+BF)；
Current time state:；Ct=Ft*Ct-1+It*tanh(Wc·[ht-1,xt]+Bc)
Out gate: Ot=σ (WO·[ht-1,xt]+BO)；
LSTM unit output: ht=Ot*tanh(Ct)；
Wherein, W is weight matrix, and B is bias matrix, xtFor the input of t moment, htFor the output of t moment hidden layer, Ct
For the state of t moment LSTM unit, σ is the activation primitive of three control doors, formula are as follows:
Activation primitive tanh formula are as follows:
Text information with this configuration, can automatically extract user's speech semantic feature, be not necessarily to manual setting feature templates,
Manpower is not only saved, versatility is good, and accuracy rate is significantly improved compared with traditional treatment method.
In order not to increase the complexity of model, which uses classifier for Logistic recurrence, i.e., in linear regression
On the basis of applied a logical function, specific formula is as follows:
As last output hθWhen >=0.5, determine user's speech without flame；Work as hθWhen≤0.5, user's speech packet is determined
Containing flame.
S442, training pattern parameter complete the training of RNN semantic analysis sorter model two-way to LSTM type.
By the words vector by user's speech in the training sample of handmarking, by the moment, first forward direction inputs two-way RNN
In, then in reversed input RNN, in forward and reverse input process, the input signal of the hidden layer of two-way RNN described in each moment
It further include the output data of the previous moment hidden layer at current time other than the input vector data comprising current time.
In the training process when prediction result and the annotation results of training sample have deviation, by being passed through in neural network
The error backpropagation algorithm of allusion quotation adjusts each weight in neural network, and error backpropagation algorithm is reversed step by step by error
All neurons for sharing each layer are propagated, obtain the error signal of each layer neuron, and then correct the weight of each neuron.It is logical
It crosses forward algorithm and successively transmits operational data, and the process for gradually modifying the weight of each neuron by backward algorithm is exactly mind
Training process through network；It repeats the above process, until the accuracy of prediction result reaches the threshold value of setting, deconditioning, this
When be believed that the trained completion of the two-way RNN model of the LSTM type.
Gradient descent method is used in the parameter of training classifier, the more new formula of parameter is as follows:
Wherein, α is that learning rate generally takes lesser value.
S5, judge whether user's input content includes flame according to the model established in S4, it is normal if not including
It sends, otherwise remind user and forbids sending.The flame includes vulgar information, pornography and violence information etc..
The present invention can be automatically completed the semantic analysis to user's speech, from source in the case where prosthetic participates in
It exercises supervision to user's speech in network direct broadcasting platform, can not only guarantee the real-time and validity of supervision, and drop
Low monitor procedure relies on for manually-operated, has saved cost of labor.
The present invention carries out semantic analysis using the two-way RNN neural network structure of LSTM type, both refers in the analysis process
The accuracy of supervision is effectively promoted referring also to Future Information in historical information.Meanwhile the present invention can be to all live streamings
All users on platform carry out the supervision with real-time and validity, and supervision range is wide, strong applicability.
In addition, the present invention also provides reference for other relevant issues in same domain, can be opened up on this basis
Extension is stretched, and is applied in same domain in the technical solution of other users speech semantic analysis, has very wide application prospect.
It is obvious to a person skilled in the art that invention is not limited to the details of the above exemplary embodiments, Er Qie
In the case where without departing substantially from spirit and essential characteristics of the invention, the present invention can be realized in other specific forms.Therefore, no matter
From the point of view of which point, the present embodiments are to be considered as illustrative and not restrictive, and the scope of the present invention is by appended power
Benefit requires rather than above description limits, it is intended that all by what is fallen within the meaning and scope of the equivalent elements of the claims
Variation is included within the present invention, and any reference signs in the claims should not be construed as limiting the involved claims.
In addition, it should be understood that although this specification is described in terms of embodiments, but not each embodiment is only wrapped
Containing an independent technical solution, this description of the specification is merely for the sake of clarity, and those skilled in the art should
It considers the specification as a whole, the technical solutions in the various embodiments may also be suitably combined, forms those skilled in the art
The other embodiments being understood that.
Claims (9)
1. a kind of user's speech semantic analysis of network-oriented live scene, which comprises the steps of:
S1, the content for obtaining user input area in network direct broadcasting platform direct broadcasting room in real time, and pre-processed；
S2, pretreated content is subjected to word segmentation processing, keeps word sequence sequence constant；
S3, the term vector of each word is obtained according to corpus dictionary, and then the vector for obtaining user's input content indicates；
S4, building and the training two-way RNN semantic analysis sorter model of LSTM type；
S5, judge whether user's input content includes flame according to the model established in S4, it is normal if not including to send,
Otherwise it reminds user and forbids sending.
2. user's speech semantic analysis of network-oriented live scene according to claim 1, which is characterized in that S1
Described in pre-process and specifically include: the removal redundant information unrelated with word content, the redundant information includes expression picture, table
Feelings symbol, numerical chracter and phonetic symbol.
3. user's speech semantic analysis of network-oriented live scene according to claim 1, which is characterized in that S3
Described in specifically included according to the term vector that corpus dictionary obtains each word: each vocabulary is shown as an only hot vector, vector
Dimension be corpus dictionary length；The corpus dictionary is formed by the corpus on network direct broadcasting platform, in the corpus dictionary
Words do not repeat.
4. user's speech semantic analysis of network-oriented live scene according to claim 1, which is characterized in that S4
Specifically comprise the following steps:
User's input content on S41, collection network live streaming platform in various types live streaming, and be marked one by one, will include
The content-label of flame is 0, is otherwise labeled as 1；
S42, user's input content of collection is pre-processed, divides training set and test set, construct corpus dictionary；
S43, the form that user inputs content of text term vector is showed according to corpus dictionary, it will according to word sequence sequence
Term vector is attached；
S44, the training data with label is input in the two-way RNN semantic analysis classifier of LSTM type and is trained, obtained
Optimal neural network model parameter finally obtains the two-way RNN semantic analysis sorter model of LSTM type.
5. user's speech semantic analysis of network-oriented live scene according to claim 4, which is characterized in that
S44 specifically comprises the following steps:
S441, the two-way RNN structure of design LSTM type, the building two-way RNN semantic analysis classifier of LSTM type, it is two-way to obtain LSTM type
RNN semantic analysis sorter model；
S442, training pattern parameter complete the training of RNN semantic analysis sorter model two-way to LSTM type.
6. user's speech semantic analysis of network-oriented live scene according to claim 4, it is characterised in that:
The two-way RNN semantic analysis classifier of LSTM type described in S44 includes sequentially sequentially connected input layer, hidden layer and output layer；
The input of the input layer is the word sequence for representing content of text；
The hidden layer is connected by multiple LSTM units, including the LSTM unit that is transmitted according to word sequence forward direction and
According to the unit of the LSTM of word sequence reverse transfer.
7. user's speech semantic analysis of network-oriented live scene according to claim 5, it is characterised in that: institute
Stating output layer is classifier, and the classifier is two classifiers.
8. user's speech semantic analysis of network-oriented live scene according to claim 5, it is characterised in that: institute
Stating LSTM unit is the ad hoc network gathered around there are three door, and three doors are controlled by Sigmoid function, can selectively be controlled
The transmitting of information flow processed, three doors are respectively input gate, forget door and out gate.
9. user's speech semantic analysis of network-oriented live scene according to claim 1, it is characterised in that: institute
Stating flame includes vulgar information, pornography and violence information.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811523923.XA CN109657241A (en) | 2018-12-13 | 2018-12-13 | User's speech semantic analysis of network-oriented live scene |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811523923.XA CN109657241A (en) | 2018-12-13 | 2018-12-13 | User's speech semantic analysis of network-oriented live scene |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109657241A true CN109657241A (en) | 2019-04-19 |
Family
ID=66114543
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811523923.XA Pending CN109657241A (en) | 2018-12-13 | 2018-12-13 | User's speech semantic analysis of network-oriented live scene |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109657241A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110659383A (en) * | 2019-08-19 | 2020-01-07 | 广州荔支网络技术有限公司 | Audio anchor scoring method based on audio content portrait |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106960042A (en) * | 2017-03-29 | 2017-07-18 | 中国科学技术大学苏州研究院 | Network direct broadcasting measure of supervision based on barrage semantic analysis |
| CN107396144A (en) * | 2017-06-30 | 2017-11-24 | 武汉斗鱼网络科技有限公司 | A kind of barrage distribution method and device |
| CN108427670A (en) * | 2018-04-08 | 2018-08-21 | 重庆邮电大学 | A kind of sentiment analysis method based on context word vector sum deep learning |
| CN108520298A (en) * | 2018-04-09 | 2018-09-11 | 中国民航大学 | A kind of land sky call semantic consistency method of calibration based on improvement LSTM-RNN |
-
2018
- 2018-12-13 CN CN201811523923.XA patent/CN109657241A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106960042A (en) * | 2017-03-29 | 2017-07-18 | 中国科学技术大学苏州研究院 | Network direct broadcasting measure of supervision based on barrage semantic analysis |
| CN107396144A (en) * | 2017-06-30 | 2017-11-24 | 武汉斗鱼网络科技有限公司 | A kind of barrage distribution method and device |
| CN108427670A (en) * | 2018-04-08 | 2018-08-21 | 重庆邮电大学 | A kind of sentiment analysis method based on context word vector sum deep learning |
| CN108520298A (en) * | 2018-04-09 | 2018-09-11 | 中国民航大学 | A kind of land sky call semantic consistency method of calibration based on improvement LSTM-RNN |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110659383A (en) * | 2019-08-19 | 2020-01-07 | 广州荔支网络技术有限公司 | Audio anchor scoring method based on audio content portrait |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107092596B (en) | Text emotion analysis method based on attention CNNs and CCR | |
| CN107563498B (en) | Image description method and system based on visual and semantic attention combined strategy | |
| CN109844743B (en) | Generating responses in automated chat | |
| CN109241255A (en) | A kind of intension recognizing method based on deep learning | |
| CN106503055A (en) | A kind of generation method from structured text to iamge description | |
| CN107480122B (en) | Artificial intelligence interaction method and artificial intelligence interaction device | |
| CN110532912B (en) | Sign language translation implementation method and device | |
| CN109308318A (en) | Training method, device, equipment and the medium of cross-domain texts sentiment classification model | |
| CN103207860B (en) | The entity relation extraction method and apparatus of public sentiment event | |
| CN107038480A (en) | A kind of text sentiment classification method based on convolutional neural networks | |
| CN108984530A (en) | A kind of detection method and detection system of network sensitive content | |
| CN108711421A (en) | A kind of voice recognition acoustic model method for building up and device and electronic equipment | |
| CN106569998A (en) | Text named entity recognition method based on Bi-LSTM, CNN and CRF | |
| CN107025284A (en) | The recognition methods of network comment text emotion tendency and convolutional neural networks model | |
| CN109388700A (en) | A kind of intension recognizing method and system | |
| CN107145514B (en) | Chinese sentence pattern classification method based on decision tree and SVM mixed model | |
| CN106547866B (en) | A kind of fine granularity sensibility classification method based on the random co-occurrence network of emotion word | |
| Shang et al. | Learning to Converse with Noisy Data: Generation with Calibration. | |
| CN112183670A (en) | Knowledge distillation-based few-sample false news detection method | |
| CN107274903A (en) | Text handling method and device, the device for text-processing | |
| CN107679225A (en) | A kind of reply generation method based on keyword | |
| CN108733675A (en) | Affective Evaluation method and device based on great amount of samples data | |
| CN111428481A (en) | Entity relation extraction method based on deep learning | |
| CN111339772B (en) | Russian text emotion analysis method, electronic device and storage medium | |
| Varshney et al. | Modelling context emotions using multi-task learning for emotion controlled dialog generation |
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: 20190419 |