CN110299140A - A kind of key content extraction algorithm based on Intelligent dialogue - Google Patents
A kind of key content extraction algorithm based on Intelligent dialogue Download PDFInfo
- Publication number
- CN110299140A CN110299140A CN201910524215.6A CN201910524215A CN110299140A CN 110299140 A CN110299140 A CN 110299140A CN 201910524215 A CN201910524215 A CN 201910524215A CN 110299140 A CN110299140 A CN 110299140A
- Authority
- CN
- China
- Prior art keywords
- robot
- word
- key content
- vector
- text
- 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
- 238000000605 extraction Methods 0.000 title claims abstract description 20
- 239000000284 extract Substances 0.000 claims abstract description 9
- 238000010606 normalization Methods 0.000 claims abstract 2
- 238000012549 training Methods 0.000 claims description 5
- 230000008859 change Effects 0.000 claims description 2
- 238000005096 rolling process Methods 0.000 claims 1
- 238000004891 communication Methods 0.000 abstract description 4
- 238000013473 artificial intelligence Methods 0.000 abstract description 3
- 230000008901 benefit Effects 0.000 description 4
- 238000005457 optimization Methods 0.000 description 4
- 238000003058 natural language processing Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000005611 electricity Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000009434 installation Methods 0.000 description 1
- 238000000034 method Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000011017 operating method Methods 0.000 description 1
- 238000002360 preparation method Methods 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/20—Natural language analysis
- G06F40/279—Recognition of textual entities
- G06F40/289—Phrasal analysis, e.g. finite state techniques or chunking
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L15/00—Speech recognition
- G10L15/26—Speech to text systems
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L15/00—Speech recognition
- G10L15/28—Constructional details of speech recognition systems
- G10L15/30—Distributed recognition, e.g. in client-server systems, for mobile phones or network applications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04M—TELEPHONIC COMMUNICATION
- H04M3/00—Automatic or semi-automatic exchanges
- H04M3/22—Arrangements for supervision, monitoring or testing
- H04M3/2281—Call monitoring, e.g. for law enforcement purposes; Call tracing; Detection or prevention of malicious calls
Landscapes
- Engineering & Computer Science (AREA)
- Computational Linguistics (AREA)
- Health & Medical Sciences (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Human Computer Interaction (AREA)
- Acoustics & Sound (AREA)
- Multimedia (AREA)
- Computer Security & Cryptography (AREA)
- Signal Processing (AREA)
- Artificial Intelligence (AREA)
- General Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Technology Law (AREA)
- Machine Translation (AREA)
Abstract
The invention discloses a kind of key content extraction algorithm based on Intelligent dialogue, more particularly to network communication field, including robot, the robot interior is integrated with ASR speech recognition system, the robot connects nlp server, the specific steps are as follows: S1, voice are converted into text and are sent to nlp server;S2, text word cutting are simultaneously indicated with vector;S3, two-way gru is run to each word, obtains the expression of another vector;S4, each word distribute a lable label and are converted to probability by softmax layers by score normalization;S5, it decodes to obtain optimal sequence using Viterbi, improve viterbi algorithm and is optimized using array is rolled;S6, the sequence that will have been marked identify harassing call.The present invention extracts crucial information by artificial intelligence from dialogue, allows robot to answer and answer, key message is recorded, while user can be facilitated to carry out decision and judge whether to call back to avoid important phone is lost.
Description
Technical field
The present invention relates to network communication technology fields, it is more particularly related to a kind of pass based on Intelligent dialogue
Key contents extraction algorithm.
Background technique
Network communication nowadays has become the major way of the daily communication of people, also expedites the emergence of while bringing convenient for people
Various Dark Industry Links.Wherein harassing call is the most serious, or even influences the routine work of people.How effectively to avoid
Harassing call also becomes the hot spot paid close attention in recent years.
At present in the industry for harassing call by following several: carrying out phone mark by user, electricity can be harassed to avoid part
Words, but a large amount of harassing call is broadcasted automatically by machine, is labeled in mass data, is needed to expend a large amount of manpower object
Power;Robot is allowed to answer, presets simple question-response form, while recording the content of dialogue, but need user again complete
Whole listens a conversation content, and whether be harassing call, under efficiency is very low if screening.
To sum up, existing Barassment preventing telephone is still to be improved in efficiency.
Summary of the invention
In order to overcome the drawbacks described above of the prior art, the embodiment of the present invention is provided in a kind of key based on Intelligent dialogue
Hold extraction algorithm, by artificial intelligence, crucial information is extracted from dialogue, allows robot to answer and answer, by key message
Record, while user can be facilitated to carry out decision and judge whether to call back to avoid important phone is lost, robot by sound
Sound is sent to ASR identification in real time, obtains the recognition result of textual form, is used for natural language processing, excavates key message, obtains
Preset question and answer, plays back with audio form, accomplishes robot auto-pickup, while mentioning to key content
Take record.
To achieve the above object, the invention provides the following technical scheme: a kind of key content based on Intelligent dialogue extracts
Algorithm, including robot, the robot interior are integrated with ASR speech recognition system, and the robot connects nlp server,
Specific step is as follows for key content extraction algorithm:
S1, when visitor or robot initiate speech exchange, visitor's sound is sent into ASR speech recognition system in real time by robot
Speech recognition is carried out in system, converts text for voice, is sent on nlp server after saving in the form of text, is returned crucial
Information;
The text of input is carried out word cutting by S2, nlp server, is indicated to each word with term vector, and term vector
It is made of 3 parts: the term vector of word2vec training, character level other vector sum word position vector, to position, character
Embedding runs a two-way lstm, then splices the hidden layer of these three, obtains the vector of a regular length;
S3, the term vector according to obtained in step S2 run two-way gru to each word in sentence, then obtain another
Vector indicates;
S4, finally with a full articulamentum lable label is distributed to each word, and by softmax layers by score
It is normalized, is converted to probability;
S5, decode to obtain optimal sequence using Viterbi, and viterbi algorithm improved, using roll array into
Row optimization;
S6, the sequence that will finally mark extract desired information by state automata, identify harassing call.
In a preferred embodiment, in one layer of crf of softmax layers of connection in the step S4.
In a preferred embodiment, it in the step S5, can also be needed to carry out mark to text according to business,
Identify harassing call.
In a preferred embodiment, the robot interior is provided with audio player, can be used in extract
Key content played back with audio form, accomplish robot auto-pickup, while record is extracted to key content.
Technical effect and advantage of the invention:
1, the present invention extracts crucial information by artificial intelligence from dialogue, allows robot to answer and answer, will be crucial
Information is recorded, while user can be facilitated to carry out decision and judge whether to call back, and closed to avoid important phone is lost
During key contents extraction, ASR speech recognition system identifies the speech exchange between visitor or robot, is converted into text, so
After be sent to nlp server and carry out word cutting, then each word is indicated with term vector, to position, character
Embedding runs a two-way lstm, then splices the hidden layer of these three, obtains the vector of a regular length, so
Two-way gru is run to each word in sentence afterwards, the expression of another vector is then obtained, finally with a full articulamentum to each word
A lable label is distributed, and score is normalized by softmax layers, probability is converted to, is decoded using Viterbi
Optimal sequence is obtained, and viterbi algorithm is improved, is optimized using array is rolled, greatly saves space, finally
The sequence that will have been marked is extracted desired information by state automata, or needs to play text according to business
Mark identifies harassing call;
2, it is connected to one layer of crf for softmax layers of the present invention, advantage is can during it is labeled for a position
To use the information marked before this, after avoiding the occurrence of softmax layers of output, the case where verb is followed by verb, benefit are generated
Optimal sequence is obtained with Viterbi decoding, to save space, being improved to viterbi algorithm, is carried out using array is rolled
Optimization, keyword extraction accuracy are high;
3, robot by sound be sent in real time ASR identification, obtain the recognition result of textual form, be used for natural language
Processing is excavated key message, obtains preset question and answer, played back with audio form, accomplish that robot connects automatically
It listens, while record is extracted to key content.
Detailed description of the invention
Fig. 1 is the overview flow chart of the key content extraction algorithm of the invention based on Intelligent dialogue.
Fig. 2 is the key content extraction algorithm schematic diagram of the invention based on Intelligent dialogue.
Specific embodiment
Following will be combined with the drawings in the embodiments of the present invention, and technical solution in the embodiment of the present invention carries out clear, complete
Site preparation description, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.It is based on
Embodiment in the present invention, it is obtained by those of ordinary skill in the art without making creative efforts every other
Embodiment shall fall within the protection scope of the present invention.
Embodiment 1:
A kind of key content extraction algorithm based on Intelligent dialogue according to shown in Fig. 1-2, including robot, the machine
People has been internally integrated ASR speech recognition system, and the robot connects nlp server, key content extraction algorithm specific steps
It is as follows:
S1, when visitor or robot initiate speech exchange, visitor's sound is sent into ASR speech recognition system in real time by robot
Speech recognition is carried out in system, converts text for voice, is sent on nlp server after saving in the form of text, is returned crucial
Information;
The text of input is carried out word cutting by S2, nlp server, is indicated to each word with term vector, and term vector
It is made of 3 parts: the term vector of word2vec training, character level other vector sum word position vector, to position, character
Embedding runs a two-way lstm, then splices the hidden layer of these three, obtains the vector of a regular length;
S3, the term vector according to obtained in step S2 run two-way gru to each word in sentence, then obtain another
Vector indicates;
S4, finally with a full articulamentum lable label is distributed to each word, and by softmax layers by score
It is normalized, is converted to probability;
S5, it decodes to obtain optimal sequence using Viterbi, to save space, and improving viterbi algorithm, use
Array is rolled to optimize;
S6, the sequence that will finally mark extract desired information by state automata, identify harassing call.
Further, do not have after softmax layers of output in the step S4 in one layer of crf of softmax layers of connection
The relationship for considering front and back label, can generate the case where verb is followed by verb, for the generation for avoiding such case, Wo Men
Softmax layers have met one layer of crf, and advantage is to can use before this during it is labeled for a position
The information of mark.
Further, the robot interior is provided with audio player, and the key content that can be used in extract is with sound
Frequency form plays back, and accomplishes robot auto-pickup, while extracting record to key content.
Embodiment 2:
A kind of key content extraction algorithm based on Intelligent dialogue, including robot, the robot interior are integrated with ASR
Speech recognition system, the robot connect nlp server, and specific step is as follows for key content extraction algorithm:
S1, when visitor or robot initiate speech exchange, visitor's sound is sent into ASR speech recognition system in real time by robot
Speech recognition is carried out in system, converts text for voice, is sent on nlp server after saving in the form of text, is returned crucial
Information;
The text of input is carried out word cutting by S2, nlp server, is indicated to each word with term vector, and term vector
It is made of 3 parts: the term vector of word2vec training, character level other vector sum word position vector, to position, character
Embedding runs a two-way lstm, then splices the hidden layer of these three, obtains the vector of a regular length;
S3, the term vector according to obtained in step S2 run two-way gru to each word in sentence, then obtain another
Vector indicates;
S4, finally with a full articulamentum lable label is distributed to each word, and by softmax layers by score
It is normalized, is converted to probability;
S5, decode to obtain optimal sequence using Viterbi, and viterbi algorithm improved, using roll array into
Row optimization;
S6, finally the sequence marked is needed to carry out mark to text according to business, identifies harassing call.
Embodiment 3:
As shown in Fig. 2, specific operating procedure is as follows:
S1, when visitor or robot initiate speech exchange, visitor's sound is sent into ASR speech recognition system in real time by robot
Speech recognition is carried out in system, such as the text of identification is " you hundred answer ", at this point, ASR speech recognition system converts voice to
Text is sent on nlp server after saving in the form of text, returns to key message;
The text of input is carried out word cutting by S2, nlp server, is indicated to each word with term vector, as X1, X2,
X3, X4, and term vector is made of 3 parts: the term vector of word2vec training, the other vector sum word position vector of character level, it is right
Position, character embedding run a two-way lstm, i.e. corresponding two h3 of X1 corresponding two h1, X2 corresponding two h2, X3,
X4 corresponds to two h4, then splices the hidden layer of these three, obtains the vector of a regular length;
S3, the term vector according to obtained in step S2 run two-way gru to each word in sentence, then obtain another
Vector indicates, obtains P1, P2, P3, P4;
S4, finally with a full articulamentum to each word distribute a lable label, i.e. lable1, lable2,
Lable3, lable4, and score is normalized by softmax layers, it is converted to probability;
S5, decode to obtain optimal sequence using Viterbi, and viterbi algorithm improved, using roll array into
Row optimization;
S6, the sequence that will finally mark extract desired information by state automata, identify harassing call.
The several points that should finally illustrate are: firstly, in the description of the present application, it should be noted that unless otherwise prescribed and
It limits, term " installation ", " connected ", " connection " shall be understood in a broad sense, can be mechanical connection or electrical connection, be also possible to two
Connection inside element, can be directly connected, and "upper", "lower", "left", "right" etc. are only used for indicating relative positional relationship, when
The absolute position for being described object changes, then relative positional relationship may change;
Secondly: the present invention discloses in embodiment attached drawing, relates only to the structure being related to the embodiment of the present disclosure, other knots
Structure, which can refer to, to be commonly designed, and under not conflict situations, the same embodiment of the present invention and different embodiments be can be combined with each other;
Last: the foregoing is only a preferred embodiment of the present invention, is not intended to restrict the invention, all in the present invention
Spirit and principle within, any modification, equivalent replacement, improvement and so on, should be included in protection scope of the present invention it
It is interior.
Claims (4)
1. a kind of key content extraction algorithm based on Intelligent dialogue, it is characterised in that: including robot, the robot interior
It is integrated with ASR speech recognition system, the robot connects nlp server, and specific step is as follows for key content extraction algorithm:
S1, when visitor or robot initiate speech exchange, visitor's sound is sent into ASR speech recognition system by robot in real time
Speech recognition is carried out, converts text for voice, is sent on nlp server after saving in the form of text, key message is returned;
The text of input is carried out word cutting by S2, nlp server, is indicated to each word with term vector, and term vector is by 3
It is grouped as: the term vector of word2vec training, the other vector sum word position vector of character level, to the embedding of position, character
A two-way lstm is run, then the hidden layer of these three is spliced, obtains the vector of a regular length;
S3, the term vector according to obtained in step S2 run two-way gru to each word in sentence, then obtain another vector
It indicates;
S4, finally with a full articulamentum lable label is distributed to each word, and is carried out score by softmax layers
Normalization, is converted to probability;
S5, it decodes to obtain optimal sequence using Viterbi, and viterbi algorithm is improved, array progress is excellent using rolling
Change;
S6, the sequence that will finally mark extract desired information by state automata, identify harassing call.
2. a kind of key content extraction algorithm based on Intelligent dialogue according to claim 1, it is characterised in that: the step
In one layer of crf of softmax layers of connection in rapid S4.
3. a kind of key content extraction algorithm based on Intelligent dialogue according to claim 1, it is characterised in that: the step
In rapid S5, it can also be needed to carry out mark to text according to business, identify harassing call.
4. a kind of key content extraction algorithm based on Intelligent dialogue according to claim 1, it is characterised in that: the machine
Device people is internally provided with audio player, can be used in playing back the key content of extraction with audio form, accomplishes machine
People's auto-pickup, while record is extracted to key content.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910524215.6A CN110299140A (en) | 2019-06-18 | 2019-06-18 | A kind of key content extraction algorithm based on Intelligent dialogue |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910524215.6A CN110299140A (en) | 2019-06-18 | 2019-06-18 | A kind of key content extraction algorithm based on Intelligent dialogue |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110299140A true CN110299140A (en) | 2019-10-01 |
Family
ID=68028209
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910524215.6A Pending CN110299140A (en) | 2019-06-18 | 2019-06-18 | A kind of key content extraction algorithm based on Intelligent dialogue |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110299140A (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110930985A (en) * | 2019-12-05 | 2020-03-27 | 携程计算机技术(上海)有限公司 | Telephone speech recognition model, method, system, device and medium |
CN111368526A (en) * | 2020-03-03 | 2020-07-03 | 支付宝(杭州)信息技术有限公司 | Sequence labeling method and system |
CN115017286A (en) * | 2022-06-09 | 2022-09-06 | 北京邮电大学 | Search-based multi-turn dialog system and method |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103281213A (en) * | 2013-04-18 | 2013-09-04 | 西安交通大学 | Method for extracting, analyzing and searching network flow and content |
CN107346340A (en) * | 2017-07-04 | 2017-11-14 | 北京奇艺世纪科技有限公司 | A kind of user view recognition methods and system |
CN108388560A (en) * | 2018-03-17 | 2018-08-10 | 北京工业大学 | GRU-CRF meeting title recognition methods based on language model |
CN108829818A (en) * | 2018-06-12 | 2018-11-16 | 中国科学院计算技术研究所 | A kind of file classification method |
KR20190019661A (en) * | 2017-08-18 | 2019-02-27 | 동아대학교 산학협력단 | Method for Natural Langage Understanding Based on Distribution of Task-specific Labels |
-
2019
- 2019-06-18 CN CN201910524215.6A patent/CN110299140A/en active Pending
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103281213A (en) * | 2013-04-18 | 2013-09-04 | 西安交通大学 | Method for extracting, analyzing and searching network flow and content |
CN107346340A (en) * | 2017-07-04 | 2017-11-14 | 北京奇艺世纪科技有限公司 | A kind of user view recognition methods and system |
KR20190019661A (en) * | 2017-08-18 | 2019-02-27 | 동아대학교 산학협력단 | Method for Natural Langage Understanding Based on Distribution of Task-specific Labels |
CN108388560A (en) * | 2018-03-17 | 2018-08-10 | 北京工业大学 | GRU-CRF meeting title recognition methods based on language model |
CN108829818A (en) * | 2018-06-12 | 2018-11-16 | 中国科学院计算技术研究所 | A kind of file classification method |
Non-Patent Citations (1)
Title |
---|
程晓锦等: "《有限状态自动机及在字符串搜索中的应用》", 《北京印刷学院学报》 * |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110930985A (en) * | 2019-12-05 | 2020-03-27 | 携程计算机技术(上海)有限公司 | Telephone speech recognition model, method, system, device and medium |
CN110930985B (en) * | 2019-12-05 | 2024-02-06 | 携程计算机技术(上海)有限公司 | Telephone voice recognition model, method, system, equipment and medium |
CN111368526A (en) * | 2020-03-03 | 2020-07-03 | 支付宝(杭州)信息技术有限公司 | Sequence labeling method and system |
CN111368526B (en) * | 2020-03-03 | 2023-04-25 | 支付宝(杭州)信息技术有限公司 | Sequence labeling method and system |
CN115017286A (en) * | 2022-06-09 | 2022-09-06 | 北京邮电大学 | Search-based multi-turn dialog system and method |
CN115017286B (en) * | 2022-06-09 | 2023-04-07 | 北京邮电大学 | Search-based multi-turn dialog system and method |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN105096941B (en) | Audio recognition method and device | |
CN106409289B (en) | Environment self-adaption method, speech recognition equipment and the household electrical appliance of speech recognition | |
CN104050160B (en) | Interpreter's method and apparatus that a kind of machine is blended with human translation | |
CN108305643B (en) | Method and device for determining emotion information | |
US8219397B2 (en) | Data processing system for autonomously building speech identification and tagging data | |
CN110299140A (en) | A kind of key content extraction algorithm based on Intelligent dialogue | |
Bach et al. | Noisy BiLSTM-Based Models for Disfluency Detection. | |
CN107977183A (en) | voice interactive method, device and equipment | |
CN108984529A (en) | Real-time court's trial speech recognition automatic error correction method, storage medium and computing device | |
CN109036391A (en) | Audio recognition method, apparatus and system | |
US8165887B2 (en) | Data-driven voice user interface | |
CN110197279B (en) | Transformation model training method, device, equipment and storage medium | |
CN111477216A (en) | Training method and system for pronunciation understanding model of conversation robot | |
CN107578769A (en) | Speech data mask method and device | |
CN103003876A (en) | Modification of speech quality in conversations over voice channels | |
CN101923854A (en) | Interactive speech recognition system and method | |
CN108376144A (en) | Man-machine more wheel dialogue methods that scene based on deep neural network automatically switches | |
CN109410949B (en) | Text content punctuation adding method based on weighted finite state converter | |
CN107451131A (en) | A kind of audio recognition method and device | |
CN107910004A (en) | Voiced translation processing method and processing device | |
CN110459202A (en) | A kind of prosodic labeling method, apparatus, equipment, medium | |
CN111737991A (en) | Text sentence break position identification method and system, electronic device and storage medium | |
CN1945692B (en) | Intelligent method for improving prompting voice matching effect in voice synthetic system | |
CN110196897B (en) | Case identification method based on question and answer template | |
CN116092472A (en) | Speech synthesis method and synthesis system |
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 | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20191001 |