CN106503101A - Electric business customer service automatically request-answering system sentence keyword extracting method - Google Patents
Electric business customer service automatically request-answering system sentence keyword extracting method Download PDFInfo
- Publication number
- CN106503101A CN106503101A CN201610900368.2A CN201610900368A CN106503101A CN 106503101 A CN106503101 A CN 106503101A CN 201610900368 A CN201610900368 A CN 201610900368A CN 106503101 A CN106503101 A CN 106503101A
- Authority
- CN
- China
- Prior art keywords
- word
- sentence
- customer service
- electric business
- extracting method
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/33—Querying
- G06F16/3331—Query processing
- G06F16/3332—Query translation
- G06F16/3335—Syntactic pre-processing, e.g. stopword elimination, stemming
-
- 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
- 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
- G06F16/36—Creation of semantic tools, e.g. ontology or thesauri
- G06F16/367—Ontology
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Databases & Information Systems (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Computational Linguistics (AREA)
- Mathematical Physics (AREA)
- Artificial Intelligence (AREA)
- Human Computer Interaction (AREA)
- Life Sciences & Earth Sciences (AREA)
- Animal Behavior & Ethology (AREA)
- Machine Translation (AREA)
Abstract
The invention discloses electric business customer service automatically request-answering system sentence keyword extracting method, comprises the following steps：Words partition system is built, and is set up deactivation vocabulary and key word is chosen by neural network characteristics.The present invention enables network customer service robot to have clearly understanding to the intention of consumer by the method for keyword extraction, achieve the actual conversation of consumer and network customer service robot, so that shopping at network is convenient, the purchase experiences of consumer are substantially increased.
Description
Technical field
The present invention relates to a kind of keyword extracting method, particularly a kind of electric business customer service automatically request-answering system sentence key word
Extracting method.
Background technology
With the development of network, shopping at network is quietly risen, and people are chosen various articles and got over by network
Become a kind of fashion to get over, but present network customer service robot can only provide the consumer with list consultancy service, i.e., to disappearing
The person of expense provides a list, allows consumer oneself to select the service for needing, consumer carry out Instant-Counseling, so can be directly
Reduce the purchase experiences of consumer.If network robot to be realized carries out reply in real time to consumer disappeared it is necessary to accurate understanding
The intention of the person of expense, if it is desired that obtaining the intention of accurate understanding consumer of robot, just has to close the sentence of consumer
Keyword is extracted.
Content of the invention
For solving the above problems, it is an object of the invention to provide a kind of electric business customer service automatically request-answering system sentence key word
Extracting method.
The technical scheme adopted by its problem of solution of the invention is electric business customer service automatically request-answering system sentence keyword extraction
Method, comprises the following steps：Words partition system is built, and is set up deactivation vocabulary and key word is chosen by neural network characteristics.
Further, the Words partition system is NLPIR Chinese word segmentation systems, the Words partition system have Chinese word segmentation function,
Part-of-speech tagging function, name Entity recognition function, definition user-oriented dictionary function and new word discovery function.
Further, described deactivation vocabulary include the high Chinese word character of English character, mathematical character, punctuate, frequency, onomatopoeia word,
Can only be in the first word for occurring, the noun of locality and interjection.
Further, key word is chosen by neural network characteristics refer to the key word obtained after user's sentence carries out participle
Key word is extracted in set, is extracted according to tripartite's surface information of key word when extracting key word, believed in terms of described three
Breath includes semantic information, self information and positional information.
Further, semantic information includes word part of speech, word association degree, sentence name Entity recognition and removes stop words；
Word part of speech refers to that various parts of speech in user's sentence become the probability of key word and have differences, to different words in keyword extraction
Property key word give different score values, for extracting score value calculating；Word association degree refer in the sentence of user word and other
There is complex relationship between word, each word is the equal of the semantic node one by one in space in a word, and they are mutual
Between exist certain associate, thus this factor is considered wherein, to calculate word association angle value using Word2vec by this method；Sentence
Son name Entity recognition refers to that name entity acquires a special sense in sentence, the content for being keyword extraction to their identification
One of；Stop words is gone to refer to that some key words frequency of occurrences in sentence is higher, but its effect very little, so carrying out
These effects little word will be removed when keyword extraction.
Further, self information includes word frequency and word length；When word frequency refers to that user repeatedly mentions a certain word, it into
Probability for key word is just very big；Word length refers to that longer word represents abundanter information, and which becomes key word
Probability is also bigger.
Further, positional information includes position first and word span, if the position that then use occurs first that single occurs
Expression is put, with occurring position first and word span is represented if 2 times or more occur；PositionWherein L is that sentence is long,
liPosition for word；Word span Hi,Wherein li2Put for finally there is lexeme, li1Put for occurring lexeme first.
The invention has the beneficial effects as follows：The present invention is electric business customer service automatically request-answering system sentence keyword extracting method, this
Invent and enable network customer service robot to have clearly understanding to the intention of consumer by the method for keyword extraction, realize
The actual conversation of consumer and network customer service robot so that shopping at network is convenient, substantially increase the purchase of consumer
Object is tested.
Description of the drawings
The invention will be further described with example below in conjunction with the accompanying drawings.
Fig. 1 is the Custom Dictionaries of the present invention；
Fig. 2 is the deactivation vocabulary of the present invention；
Fig. 3 is the part of speech score table of the present invention；
Fig. 4 is the sample result of the present invention.
Specific embodiment
The identification are intended to by user's commodity is realized, first have to do is exactly the understanding to user input sentence.This reason
The element task of solution is exactly to extract the key word in sentence, and the key wordses in sentence are the external presentations of user view expression.
Extraction in question answering system to user's key word, carries out participle to the sentence of user first, then removes some for wherein including
Network address connection, punctuation mark etc., carry out keyword extraction to remaining word, and the main contents in step are as follows：Build participle
System, foundation disable vocabulary and choose key word by neural network characteristics.
Need to carry out participle with Chinese sentence unlike English, the quality of participle has the extraction of key word to be affected,
The present invention chooses NLPIR Chinese word segmentation systems, that is, before ICTCLAS2013, the system has multiple function such as：Chinese
Participle, part-of-speech tagging, name Entity recognition, and define user-oriented dictionary, new word discovery etc.；That participated in 2003 is " international
SIGHAN participle contests " achieve comprehensive first achievement, are one of classic Chinese automatic word-cut, at present, Global Subscriber
Break through 200,000.Electric business customer service robot uses this Words partition system, simultaneously because its service market is cosmetic industry can have
Some specialized vocabularies need to be added, and construct exclusive User Defined dictionary here, individual comprising vocabulary more than 660, user
Front 9 row of Custom Dictionaries is as shown in Figure 1.
In the set obtained after participle, it is found that some invalid words, they can be excluded from the candidate of key word
Collection.As system uses scene different, so disabling the characteristics of will considering net purchase when vocabulary is set up.Numerous scholars
The conclusion of stop words is also carried out, by " the high Chinese word character of English character, number, measure word, mathematical character, punctuate, frequency, onomatopoeia
Word, can only be in the first word for occurring, pronoun, the noun of locality, interjection " etc. listed the range of choice of stop words in, for different places
Can there are some differences in the selection of reason text these words.
In the system, some numbers and pronoun etc. are all without stop words is put into, because customer can be related to business in shopping
The quantity of product, price etc., these are the key messages in sentence, it is clear that can not be ignored；Additionally, pronoun is also important,
In question answering process, customer often can be referred to by the way of referring to the commodity that said, therefore pronoun is also important sentence information
One of；But " greeting word " often occurred in shopping at network is for example, " parent, hello, " etc. key letter all to sentence
Breath has no impact, so it is put into deactivation vocabulary, but in question answering system, these greeting words have corresponding greeting instead
When feedback, i.e. user are greeted, robot also can enthusiasm greeting response.In addition, some individual character auxiliary words and onomatopoeia word, word do not include yet
Therefore they also listed in by important information, and the user's chat language material for interacting language material and collection by statistical machine people carries out word frequency
Statistics, then carry out screen conclusion obtain disable vocabulary as shown in Figure 2.
Question sentence keyword feature is extracted, and the actual question sentence to user obtains C=[c after carrying out participle1, c2…cn]；To ciEnter
Row feature extraction, the feature selection of word launch to be semantic, itself and position from tripartite's surface information.
Semantic information：(1) word part of speech：In user's sentence, various parts of speech become the probability of key word and can have differences,
Different score values are given to key word according to different parts of speech in keyword extraction, for extracting score value calculating, part of speech score table
As shown in Figure 3.(2) word association degree：Miscellaneous cyberrelationship is incorporated in language, and make use of the parameter structure in complex network
The complex network of 15 kinds of language is made, in the sentence of user, between word and other words, there is complex relationship, each in a word
Individual word is the equal of the semantic node one by one in space, and they have certain association each other, thus the present invention by this
Factor considers wherein, to calculate word association angle value using Word2vec.(3) sentence name Entity recognition：Name entity is in sentence
Often acquire a special sense in son, to one of content that their identification is a lot of keyword extractions.(4) stop words：In sentence
The middle frequency of occurrences is higher, but its effect very little, many times stop words will be removed when keyword extraction is carried out.
Self information：(1) word frequency：In the language of user when word frequency gradually increases, such as user repeatedly mentions a certain word
When, the probability that it becomes key word is very big.(2) word length：There is statistics to find that longer word represents abundanter information,
The probability which becomes key word is very big, and in the system, some proper nouns can have longer situation, for example：Robot customer service
" biological fiber ", " Fructus Rubi " in key word tree etc..
Positional information：Position and word span first；The positional information that word occurs in sentence, if what single occurred
Then with the positional representation that occurs first, with occurring position first and word span is represented if 2 times or more occur；PositionWherein L is that sentence is long, liPosition for word；Word span Hi,Wherein li2Put for finally there is lexeme, li1For
Occur lexeme first to put.
Electric business customer service robot has compiled language material according to certain cosmetic on-line shop shopping language material in conjunction with itself product and field
Storehouse, is tested from wherein having extracted unduplicated 1000 sentences here, and it is a lot of that reason is that the corpus data for arranging is present
Similarity is limited plus current language material quantity, and experimental procedure is as follows：
(1) carry out participle using Chinese Academy of Sciences NLPIR Chinese word segmentation systems, due to electric business customer service robot application background and
The reason for product object, be inevitably present wrong participle after participle being carried out using Words partition system；Therefore, the present invention is to language material
After the participle in storehouse is audited, the participle of mistake is carried out extraction and forms raising point in self-defined dictionary for word segmentation addition Words partition system
The accuracy of word.
(2) 1000 sentences to extracting carry out participle after simple process, obtain a word segmentation result about more than 6600, are utilizing
Chinese Academy of Sciences's Words partition system is achieved that the mark for naming the identification and part of speech of entity to word during carrying out participle；The present invention
Give different parts of speech different score values, according to the participle standard that the Chinese Academy of Sciences uses, work out part of speech score table such as table 3, in addition
Name entity is for 1, is not for 0.
(3) and then to this more than 6600 words count other above-mentioned eigenvalues, word frequency, word length and lexeme put and
Word span statistical computation；Stop words statistics is for 1 using stop words list notation, is not for 0；The calculating of word association degree is related to
And the use to Word2vec, have collected the language material from 4 bulks such as shop shopping, news, comment, cosmetics crawls first,
It is trained using Word2vec on the server, training obtains the bin file comprising vocabulary vector after terminating, using bin texts
Part is calculating the respective average degree of association of more than 6600 word.Here all of eigenvalue is obtained, and data are ready.
Data are tested using matlabR2014a, the present invention is processed to 6600 datas for obtaining, deleted
It is invalid that some judge, if Word2vec values are FAULT, because some words that limit of language material resource fail to obtain its word
The expression of vector；Find that the part of speech score value that the part-of-speech tagging such as word " neutrality ", " mixed type " is the distinction word of b etc. is 0 simultaneously, real
In border, these words illustrate the attribute of skin, and its part of speech value is set to 0.7 therefore；Finally, 6592 valid data have been obtained, whole
Manage data publication address is as follows：http：// 120.237.31.12/E_Bot_backstage/date.html, entitled " customer service
Robot BP neural network experimental data ".The present invention is using the gui tool nprtool in matl-ab, data set used in which
Default allocation is as follows：Training sample 70%, checking sample 15%, test sample 15%, neutral net is hiding in the training process
The setting of node layer is first according to experience and is set to 5, finds to train and result when hidden layer nodes are 10 through hands-on
Relatively good.In view of the present invention adds the word association degree calculated using Word2vec in neutral net, here in training
Respectively to plus this feature and be not added with the data of this feature and carry out Comparison of experiment results, i.e., neural network node combination be respectively (8,
10,1) with (7,10,1) process experiment obtaining both confusion matrixs.
In addition, the present invention carries out keyword extraction this method using neutral net, carry out in document keyword extraction
Application, the result of the present invention will be contrasted with which., through model experiment, test is accurate for the accuracy rate of key word identification of the present invention
Really rate is relatively stable more than 88%, and optimum has reached 90.7%, and test accuracy rate is higher than to realize document using BP neural network
83.8% for extracting, illustrates the feasibility that BP neural network is applied in sentence keyword extraction；But its data volume that tests
It is news and journal of writings, its record is all 200, and vocabulary quantity is huge；The present invention chooses 1000 sentences, quantity tool in unit
There is comparability, additionally, the result for doing more big data quantity can have more persuasion property.
Test through before, the present invention saves the optimum training simulation model of experiment and actual sentence is imitated
Very, the result of the example of presentation is as shown in Figure 4：
Above four examples experienced participle and keyword extraction, and its result is presented below as：
Ex1：With this facial film, the whiter people of Bulbus Lilii facial film can not apply to how this nurses？
Participle：Bulbus Lilii/n facial films/n compares/d is white/a /udel people/n can/r facial film/n this with/v use/p /y not /d
Be suitable for/v should/v how/ryv nursing/n
Keyword extraction result：Bulbus Lilii facial film compares how white man is nursed with this facial film is inapplicable
Ex2：How about the Bulbus Lilii of this plate goes the effect of fat granule
Participle：This/rz plates/n /ude1 Bulbus Liliies/n goes/v fat granules/n /ude1 effects/n how/ryv
Keyword extraction result：How this plate Bulbus Lilii goes fat granule effect
Ex3：Heavy wool is suitable for anti-acne silk mask on the face？
Participle：On the face/s heavy wools/a be suitable for/v anti-acnes/v silkworm silks/n facial films/n /y
Keyword extraction result：Heavy wool is suitable for anti-acne silk mask on the face
Ex4：The not handy not freight charges of can guaranteeing replacement of Bulbus Lilii facial film？
Participle：Bulbus Lilii/n facial films/n not /d is handy/a can guarantee replacement with/v/v not /d freight charges/n /y
Keyword extraction result：The not handy not freight charges of guaranteeing replacement of Bulbus Lilii facial film
The above, simply presently preferred embodiments of the present invention, the invention is not limited in above-mentioned embodiment, as long as
Which reaches the technique effect of the present invention with identical means, should all belong to protection scope of the present invention.
Claims (7)
1. electric business customer service automatically request-answering system sentence keyword extracting method, it is characterised in that comprise the following steps：Build participle
System, foundation disable vocabulary and choose key word by neural network characteristics.
2. electric business customer service automatically request-answering system sentence keyword extracting method according to claim 1, it is characterised in that：Institute
It is NLPIR Chinese word segmentation systems to state Words partition system, and the Words partition system has Chinese word segmentation function, part-of-speech tagging function, name
Entity recognition function, definition user-oriented dictionary function and new word discovery function.
3. electric business customer service automatically request-answering system sentence keyword extracting method according to claim 1, it is characterised in that：Institute
State and disable vocabulary and include the high Chinese word character of English character, mathematical character, punctuate, frequency, onomatopoeia word, can only occur the first
Word, the noun of locality and interjection.
4. electric business customer service automatically request-answering system sentence keyword extracting method according to claim 1, it is characterised in that：Logical
Cross neural network characteristics selection key word and refer to extraction key word in the keyword set obtained after user's sentence carries out participle,
Extract key word when extracted according to tripartite's surface information of key word, tripartite's surface information include semantic information, from
Body information and positional information.
5. electric business customer service automatically request-answering system sentence keyword extracting method according to claim 4, it is characterised in that：Language
Adopted information includes word part of speech, word association degree, sentence name Entity recognition and removes stop words；Word part of speech refers to user's language
In sentence, various parts of speech become the probability of key word and have differences, and the key word of different parts of speech are given in keyword extraction different
Score value, for extracting score value calculating；Word association degree is referred to
System, in a word, each word is the equal of the semantic node one by one in space, and they have certain association each other, because
And this factor is considered wherein, to calculate word association angle value using Word2vec by this method；Sentence name Entity recognition is referred to
Name entity acquires a special sense in sentence, to one of content that their identification is keyword extraction；Stop words is gone to refer to
Some key words in sentence the frequency of occurrences this is higher, but its effect very little, so when keyword extraction is carried out
The little word of these effects will be removed.
6. electric business customer service automatically request-answering system sentence keyword extracting method according to claim 4, it is characterised in that：From
Body information includes word frequency and word length；When word frequency refers to that user repeatedly mentions a certain word, it becomes the probability of key word
Just very big；Word length refers to that longer word represents abundanter information, and the probability which becomes key word is also bigger.
7. electric business customer service automatically request-answering system sentence keyword extracting method according to claim 4, it is characterised in that：Position
Confidence breath includes position and word span first, if single occur then with the positional representation for occurring first, if 2 times or with
Upper occur then with occurring position first and word span is represented；PositionWherein L is that sentence is long, liPosition for word；Word span
Hi,Wherein li2Put for finally there is lexeme, li1Put for occurring lexeme first.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610900368.2A CN106503101A (en) | 2016-10-14 | 2016-10-14 | Electric business customer service automatically request-answering system sentence keyword extracting method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610900368.2A CN106503101A (en) | 2016-10-14 | 2016-10-14 | Electric business customer service automatically request-answering system sentence keyword extracting method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN106503101A true CN106503101A (en) | 2017-03-15 |
Family
ID=58293998
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610900368.2A Pending CN106503101A (en) | 2016-10-14 | 2016-10-14 | Electric business customer service automatically request-answering system sentence keyword extracting method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106503101A (en) |
Cited By (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107861942A (en) * | 2017-10-11 | 2018-03-30 | 国网浙江省电力公司电力科学研究院 | A kind of electric power based on deep learning is doubtful to complain work order recognition methods |
| CN108334490A (en) * | 2017-04-07 | 2018-07-27 | 腾讯科技（深圳）有限公司 | Keyword extracting method and keyword extracting device |
| CN108536678A (en) * | 2018-04-12 | 2018-09-14 | 腾讯科技（深圳）有限公司 | Text key message extracting method, device, computer equipment and storage medium |
| CN108628882A (en) * | 2017-03-20 | 2018-10-09 | 北京京东尚科信息技术有限公司 | Method and system for prejudging problem |
| WO2019000170A1 (en) * | 2017-06-26 | 2019-01-03 | Microsoft Technology Licensing, Llc | Generating responses in automated chatting |
| CN109145293A (en) * | 2018-08-06 | 2019-01-04 | 中国地质大学（武汉） | A kind of keyword extracting method and system towards merit |
| CN109525737A (en) * | 2018-12-03 | 2019-03-26 | 商客通尚景信息技术江苏有限公司 | A kind of call Cut-in control method and system |
| CN109635278A (en) * | 2018-11-20 | 2019-04-16 | 北京小谛机器人科技有限公司 | Semantic representation method and device |
| CN110020015A (en) * | 2017-12-29 | 2019-07-16 | 中国科学院声学研究所 | A kind of conversational system answers generation method and system |
| CN110457447A (en) * | 2019-05-15 | 2019-11-15 | 国网浙江省电力有限公司电力科学研究院 | A kind of power grid Task conversational system |
| CN110532390A (en) * | 2019-08-26 | 2019-12-03 | 南京邮电大学 | A kind of news keyword extracting method based on NER and Complex Networks Feature |
| WO2021012382A1 (en) * | 2019-07-25 | 2021-01-28 | 深圳壹账通智能科技有限公司 | Method and apparatus for configuring chat robot, computer device and storage medium |
| CN112559769A (en) * | 2020-12-15 | 2021-03-26 | 上海钛米机器人股份有限公司 | Method and device for generating question-answer database |
| CN112988984A (en) * | 2021-05-24 | 2021-06-18 | 腾讯科技（深圳）有限公司 | Feature acquisition method and device, computer equipment and storage medium |
| CN114118060A (en) * | 2021-11-10 | 2022-03-01 | 北京深维智信科技有限公司 | Method and system for automatically identifying key events from sales session |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101196904A (en) * | 2007-11-09 | 2008-06-11 | 清华大学 | News keyword abstraction method based on word frequency and multi-component grammar |
| CN102033919A (en) * | 2010-12-07 | 2011-04-27 | 北京新媒传信科技有限公司 | Method and system for extracting text key words |
| US20110202332A1 (en) * | 2001-03-16 | 2011-08-18 | Meaningful Machines, LLC | Knowledge System Method and Apparatus |
| CN105843897A (en) * | 2016-03-23 | 2016-08-10 | 青岛海尔软件有限公司 | Vertical domain-oriented intelligent question and answer system |
-
2016
- 2016-10-14 CN CN201610900368.2A patent/CN106503101A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20110202332A1 (en) * | 2001-03-16 | 2011-08-18 | Meaningful Machines, LLC | Knowledge System Method and Apparatus |
| CN101196904A (en) * | 2007-11-09 | 2008-06-11 | 清华大学 | News keyword abstraction method based on word frequency and multi-component grammar |
| CN102033919A (en) * | 2010-12-07 | 2011-04-27 | 北京新媒传信科技有限公司 | Method and system for extracting text key words |
| CN105843897A (en) * | 2016-03-23 | 2016-08-10 | 青岛海尔软件有限公司 | Vertical domain-oriented intelligent question and answer system |
Non-Patent Citations (3)
| Title |
|---|
| 岳芸: ""电商客服自动问答系统的商品意图识别"", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
| 杨凯艳: ""基于改进的TFIDF关键词自动提取算法研究"", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
| 黄轩 等: ""基于多特征的中文关键词抽取方法"", 《计算机与现代化》 * |
Cited By (23)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108628882A (en) * | 2017-03-20 | 2018-10-09 | 北京京东尚科信息技术有限公司 | Method and system for prejudging problem |
| CN108334490A (en) * | 2017-04-07 | 2018-07-27 | 腾讯科技（深圳）有限公司 | Keyword extracting method and keyword extracting device |
| CN108334490B (en) * | 2017-04-07 | 2021-05-07 | 腾讯科技（深圳）有限公司 | Keyword extraction method and keyword extraction device |
| US11586810B2 (en) | 2017-06-26 | 2023-02-21 | Microsoft Technology Licensing, Llc | Generating responses in automated chatting |
| WO2019000170A1 (en) * | 2017-06-26 | 2019-01-03 | Microsoft Technology Licensing, Llc | Generating responses in automated chatting |
| CN107861942B (en) * | 2017-10-11 | 2021-10-26 | 国网浙江省电力有限公司营销服务中心 | Suspected power complaint work order identification method based on deep learning |
| CN107861942A (en) * | 2017-10-11 | 2018-03-30 | 国网浙江省电力公司电力科学研究院 | A kind of electric power based on deep learning is doubtful to complain work order recognition methods |
| CN110020015A (en) * | 2017-12-29 | 2019-07-16 | 中国科学院声学研究所 | A kind of conversational system answers generation method and system |
| CN108536678B (en) * | 2018-04-12 | 2023-04-07 | 腾讯科技（深圳）有限公司 | Text key information extraction method and device, computer equipment and storage medium |
| CN108536678A (en) * | 2018-04-12 | 2018-09-14 | 腾讯科技（深圳）有限公司 | Text key message extracting method, device, computer equipment and storage medium |
| CN109145293A (en) * | 2018-08-06 | 2019-01-04 | 中国地质大学（武汉） | A kind of keyword extracting method and system towards merit |
| CN109145293B (en) * | 2018-08-06 | 2021-05-28 | 中国地质大学（武汉） | Case-oriented keyword extraction method and system |
| CN109635278A (en) * | 2018-11-20 | 2019-04-16 | 北京小谛机器人科技有限公司 | Semantic representation method and device |
| CN109525737A (en) * | 2018-12-03 | 2019-03-26 | 商客通尚景信息技术江苏有限公司 | A kind of call Cut-in control method and system |
| CN110457447A (en) * | 2019-05-15 | 2019-11-15 | 国网浙江省电力有限公司电力科学研究院 | A kind of power grid Task conversational system |
| WO2021012382A1 (en) * | 2019-07-25 | 2021-01-28 | 深圳壹账通智能科技有限公司 | Method and apparatus for configuring chat robot, computer device and storage medium |
| CN110532390A (en) * | 2019-08-26 | 2019-12-03 | 南京邮电大学 | A kind of news keyword extracting method based on NER and Complex Networks Feature |
| CN112559769A (en) * | 2020-12-15 | 2021-03-26 | 上海钛米机器人股份有限公司 | Method and device for generating question-answer database |
| CN112559769B (en) * | 2020-12-15 | 2024-04-16 | 上海钛米机器人股份有限公司 | Method and device for generating question-answer database |
| CN112988984B (en) * | 2021-05-24 | 2021-08-06 | 腾讯科技（深圳）有限公司 | Feature acquisition method and device, computer equipment and storage medium |
| CN112988984A (en) * | 2021-05-24 | 2021-06-18 | 腾讯科技（深圳）有限公司 | Feature acquisition method and device, computer equipment and storage medium |
| CN114118060A (en) * | 2021-11-10 | 2022-03-01 | 北京深维智信科技有限公司 | Method and system for automatically identifying key events from sales session |
| CN114118060B (en) * | 2021-11-10 | 2022-09-27 | 北京深维智信科技有限公司 | Method and system for automatically identifying key events from sales session |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106503101A (en) | Electric business customer service automatically request-answering system sentence keyword extracting method | |
| Sidorov et al. | Empirical study of machine learning based approach for opinion mining in tweets | |
| Hussain et al. | An approach to detect abusive bangla text | |
| US7865354B2 (en) | Extracting and grouping opinions from text documents | |
| KR101005337B1 (en) | System for extraction and analysis of opinion in web documents and method thereof | |
| Anthony | Visualisation in corpus-based discourse studies | |
| CN110569354B (en) | Barrage emotion analysis method and device | |
| CN108885617B (en) | Sentence analysis system and program | |
| CN107291680A (en) | A kind of system and implementation method that automatically generate composition based on template | |
| CN109241525B (en) | Keyword extraction method, device and system | |
| CN113591489B (en) | Voice interaction method and device and related equipment | |
| CN112631436B (en) | Method and device for filtering sensitive words of input method | |
| CN107491435A (en) | Method and device based on Computer Automatic Recognition user feeling | |
| CN110287319A (en) | Students' evaluation text analyzing method based on sentiment analysis technology | |
| CN108108468A (en) | A kind of short text sentiment analysis method and apparatus based on concept and text emotion | |
| Bansal et al. | Code-switching patterns can be an effective route to improve performance of downstream NLP applications: A case study of humour, sarcasm and hate speech detection | |
| CN112860896A (en) | Corpus generalization method and man-machine conversation emotion analysis method for industrial field | |
| CN102929860A (en) | Chinese clause emotion polarity distinguishing method based on context | |
| CN110119443A (en) | A kind of sentiment analysis method towards recommendation service | |
| Saranya et al. | A Machine Learning-Based Technique with IntelligentWordNet Lemmatize for Twitter Sentiment Analysis. | |
| TW201820172A (en) | System, method and non-transitory computer readable storage medium for conversation analysis | |
| Hasan et al. | Sentiment classification in bangla textual content: A comparative study | |
| CN110610003A (en) | Method and system for assisting text annotation | |
| CN105302859B (en) | A kind of intelligent interactive system Internet-based | |
| Patil et al. | Novel technique for script translation using NLP: performance evaluation |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | 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: 20170315 |
|
| RJ01 | Rejection of invention patent application after publication |