CN109241241A - A kind of mobile subscriber's relationship type recognition methods based on communication data - Google Patents

A kind of mobile subscriber's relationship type recognition methods based on communication data Download PDF

Info

Publication number
CN109241241A
CN109241241A CN201810949662.1A CN201810949662A CN109241241A CN 109241241 A CN109241241 A CN 109241241A CN 201810949662 A CN201810949662 A CN 201810949662A CN 109241241 A CN109241241 A CN 109241241A
Authority
CN
China
Prior art keywords
data
communication data
relationship type
parameter
mobile subscriber
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
Application number
CN201810949662.1A
Other languages
Chinese (zh)
Inventor
郝阳
俞鹏飞
高恒
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Telecommunication Science And Technology Fifth Research Institute Co Ltd
Original Assignee
Telecommunication Science And Technology Fifth Research Institute Co Ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Telecommunication Science And Technology Fifth Research Institute Co Ltd filed Critical Telecommunication Science And Technology Fifth Research Institute Co Ltd
Priority to CN201810949662.1A priority Critical patent/CN109241241A/en
Publication of CN109241241A publication Critical patent/CN109241241A/en
Pending legal-status Critical Current

Links

Classifications

    • 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
    • G06N3/048—Activation functions
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/285—Selection of pattern recognition techniques, e.g. of classifiers in a multi-classifier system
    • G—PHYSICS
    • G10—MUSICAL INSTRUMENTS; ACOUSTICS
    • G10L—SPEECH ANALYSIS OR SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING; SPEECH OR AUDIO CODING OR DECODING
    • G10L15/00—Speech recognition
    • G10L15/06—Creation of reference templates; Training of speech recognition systems, e.g. adaptation to the characteristics of the speaker's voice
    • G10L15/063—Training
    • G—PHYSICS
    • G10—MUSICAL INSTRUMENTS; ACOUSTICS
    • G10L—SPEECH ANALYSIS OR SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING; SPEECH OR AUDIO CODING OR DECODING
    • G10L15/00—Speech recognition
    • G10L15/26—Speech to text systems

Abstract

Mobile subscriber's relationship type recognition methods based on communication data that the invention discloses a kind of, the communication data of mobile subscriber is converted textual form by this method, the communication data of different modes is normalized, unified communication format is converted by different-format, different types of communication data, then it is identified using mode identification technology.The present invention automatically analyzes relationship type contained in communication data by mode identification technology, is not necessarily to manual intervention.The present invention automatically analyzes the relevance between the communication data of different time points by the information processing technology, is merged into an entirety and is analyzed, recognition methods strong applicability of the invention.

Description

A kind of mobile subscriber's relationship type recognition methods based on communication data
Technical field
The present invention relates to fields of communication technology, and in particular to a kind of mobile subscriber's relationship type identification based on communication data Method.
Background technique
It is which kind of relationship, such as Peer Relationships, friends, family's relationship etc. between two people that relationship type, which refers to,.It closes There is no fixed standard in for the division of set type, but is divided according to actual needs.Judge two shiftings communicated Employing the relationship type between family is the basic task in the application such as data mining, social network analysis, is analysis mobile subscriber Between relevance basis.
Analysis for relationship type, common processing method is will be by the way of artificial directly between mobile subscriber Communication data analyzed, the relationship type between mobile subscriber is judged by manually.There are following for this processing method The shortcoming of various aspects:
1, there are redundancies for communication data, and ununified format, data are different because of different communication mode, can not be intuitive Ground reflects whether it is valuable for relationship analysis.
2, the quantity of communication data is very huge, and by the method for manual analysis, there are inefficiency, dry as dust, easy The problem of error.
3, the record of communication data often carries out sequentially in time, and single communication data can not often reflect movement Relationship type between user, and need to analyze a plurality of communication data.There are certain passes between different communication data Connection property needs to carry out these communication datas unified analysis when judging relationship type, while excluding unrelated with relationship type Data.Communication data with relevance may be spaced for a long time in time, when by manual analysis, be needed to pervious logical Letter data is remembered, and could find the relevance of present communications data and former communication data, and is further discovered that relation object Type.When the traffic is very huge, by carrying out artificial memory and to analyze relationship type be unpractical.
Summary of the invention
In order to solve it is existing by manual identified customer relationship type inefficiency, recognition accuracy is not high enough, high limitation Property etc. technical problems, the present invention provides solve the above problems it is a kind of based on communication data mobile subscriber's relationship type identification The communication data of mobile subscriber is converted textual form by method, this method, analyzes the relevance between each communication data, will Related communication data merges into an entirety, is then identified using mode identification technology to it.The present invention is not necessarily to people Work intervention can be realized customer relationship type and classify automatically.
The present invention is achieved through the following technical solutions:
A kind of mobile subscriber's relationship type recognition methods based on communication data, method includes the following steps:
Step 1: the text and voice data of user's communication are acquired by data acquisition and language data process unit, and The voice communication data of acquisition is converted into textcommunication data；
Step 2: the textcommunication data of different-format are converted to the polymerization of user's dimension by normalized unit Unified format textcommunication data；
Step 3: carrying out the training of model identification parameter and model ginseng to normalization textcommunication data by model database Number saves；
Step 4: being used by pattern recognition unit mobile based on the model identification parameter that identification model database provides Family communicates text data and carries out pattern-recognition, output mobile customer relationship type.
Preferably, in the step 1, voice communication data is converted into textcommunication data specifically: using being based on The machine learning algorithm of neural network, is trained neural network model by way of supervised learning, it is made to have identification Then the function of voice is encapsulated as fixed component, for voice communication data to be converted to textcommunication data.
Preferably, the normalized unit chooses suitable characteristic parameter for different communication type, completes from non- Structured text communication data forms a unified text communication format to the conversion of structured text communication data.
Preferably, the model database is made of multiple training servers and a parameter server, wherein the instruction Practice server to be responsible for being trained the textcommunication data of input, the training server is provided dependent on parameter server Model parameter, after a training server completes training, the model parameter of update can be returned to parameter server by it, thus The update of implementation model；What is stored in the parameter server is finally obtained model parameter, and responds and come from pattern-recognition The request of unit, and required model parameter is passed into pattern recognition unit, so that it is completed identification mission.
Preferably, the pattern recognition unit uses natural language processing technique, opens up to a recurrent neural network Open calculating, the word in one sentence of each node processing after expansion, the last one word of each sentence it is defeated The output of corresponding relationship type out.
Preferably, to a recurrent neural network progress unfolding calculation, detailed process is as follows:
xtIndicate the input of t (t=1,2,3...) step；
stFor the state that the t of hidden layer is walked, it is the memory unit of network；stIt is hidden according to current input layer and previous step The state of hiding layer is calculated: st=f (Uxt+Wst-1), wherein f is nonlinear activation primitive, is calculating s0When, by s-1It sets For 0 vector；
otIt is the output of t step, it carries out softmax transformation: o to the output of a upper layer networkt=softmax (Vst)。
The present invention has the advantage that and the utility model has the advantages that
The communication data of different modes is normalized in mobile subscriber's relationship type analysis method of the invention, will Different-format, different types of communication data are converted into unified communication format.Automatic relationship type analysis, passes through pattern-recognition Technology automatically analyzes relationship type contained in communication data, is not necessarily to manual intervention.Automatic time association analysis, passes through letter Breath processing technique automatically analyzes the relevance between the communication data of different time points, is merged into an entirety and is divided Analysis.
Mobile subscriber's relationship type analysis method strong applicability of the invention, main fusion is in the number to information of mobile user According to excavate and social network analysis system in, but the present disclosure applies equally to similar other information processing systems.
Detailed description of the invention
Attached drawing described herein is used to provide to further understand the embodiment of the present invention, constitutes one of the application Point, do not constitute the restriction to the embodiment of the present invention.In the accompanying drawings:
Fig. 1 is the functional block diagram of the invention.
Fig. 2 is data acquisition and language data process cell schematics of the invention.
Fig. 3 is model database schematic diagram of the invention.
Fig. 4 is the unfolding calculation schematic diagram of neural network of the invention.
Specific embodiment
To make the objectives, technical solutions, and advantages of the present invention clearer, below with reference to embodiment and attached drawing, to this Invention is described in further detail, and exemplary embodiment of the invention and its explanation for explaining only the invention, are not made For limitation of the invention.
Embodiment
As shown in Figure 1, the present embodiment is acquired by data and language data process unit, normalization unit, pattern-recognition list Mobile subscriber's relationship type identifying system of member and model database composition, the voice communication data of mobile subscriber is converted to Textcommunication data analyze the relevance between each communication data, related communication data are merged into an entirety, so It is identified using mode identification technology afterwards.
Wherein, data acquisition and language data process unit are the acquisitions for completing text and voice class data, and complete language Sound communication data is converted into text.Voice communication data conversion uses machine learning techniques neural network based, first passes through The mode of supervised learning is trained neural network model, and the ability for making it have identification voice, it is solid for being then encapsulated Fixed component, for converting text for voice communication data.The unit has the ability of on-line study, can by switch with When switch between training and recognition mode.When being switched to training mode, which is then converted to text for voice communication data This communication data is trained for model database；When being switched to recognition mode, which then turns voice communication data It is changed to textcommunication data and carries out relationship type identification for pattern recognition unit.The schematic diagram of the unit is as shown in Figure 2.
Normalization unit is to convert the textcommunication data of different-format to the unified format polymerizeing with user's dimension Textcommunication data.The unit chooses suitable characteristic parameter for different communication type, completes to communicate from non-structured text Data form a unified text communication format to the conversion of structured text communication data.The partial data of the format is such as Shown in lower:
Initiate communication object Receive communication object Communication start time The sign off time Communication position
180****2681 135****5623 2017-10-01 9:00 2017-10-01 9:05 Chengdu
The effect of model database is with communication data training and preservation model.The unit is implementation relation type analysis Core component, it is made of a distributed server cluster, can complete the training and store function to text data.
It is made of multiple training servers and a parameter server, and wherein training server is responsible for the text to input Data are trained, the model parameter that it depends on parameter server to provide.After a training server completes training, its meeting The parameter of update is returned into parameter server, thus the update of implementation model.
What is stored in parameter server is to finally obtain model parameter, it responds the request from pattern recognition unit, and Required parameter is passed to the former, so that it completes identification mission.The block diagram of the unit is as shown in Figure 3.
For model parameter training as to traditional neural metwork training, it is same using being based on error back propagation Gradient descent algorithm, but with some difference.If neural network is unfolded, the parameter of different nodes is shared , and no for traditional neural network.And when using gradient descent algorithm in the neural network of expansion, Mei Yibu Output not only rely on the network currently walked, and also rely on the state of several step networks in front.For example, also being needed in t=4 Three steps are transmitted backward, and subsequent three step is required plus various gradients.Therefore, when conventional exercises method can not solve long Dependence Problem, i.e., current output is related with one section of very long sequence of front, and conventional exercises method can bring gradient to disappear Or gradient explosion issues.The present embodiment has used a kind of improved method based on LSTMs, specially to cope with this problem.
Pattern recognition unit function is to carry out pattern-recognition to textcommunication data, obtains relationship type.The unit uses Natural language processing technique identifies text information using the model parameter saved in parameter server.The unit executes Core procedure be unfolding calculation to a recurrent neural network, as shown in Figure 4.
It is unfolded after one sentence of each node processing in a word, the last one word of each sentence it is defeated The output of corresponding relationship type out.Neural network is carried out as shown in Figure 4 to be launched into a full neural network.For example, to one Sentence comprising 5 words, then the network being unfolded is one five layers of neural network, each layer represents a word.For this The calculating process of network is as follows:
xtIndicate the input of t (t=1,2,3...) step；For example, x1For second word one-hot encoding (one-hot) to Measure (according to Fig.4, x0For first word)；
stFor the state that the t of hidden layer is walked, it is the memory unit of network；stIt is hidden according to current input layer and previous step The state of hiding layer is calculated: st=f (Uxt+Wst-1), wherein f is nonlinear activation primitive, such as tanh or ReLU function, Calculating s0When, i.e. the hiding layer state of first word needs to use s-1, but itself and be not present, in the present embodiment will It is set as the realization of 0 vector；
otIt is the output of t step, such as vector of next word indicates, it carries out softmax to the output of a upper layer network Transformation: ot=softmax (Vst)。
In the present embodiment, the specific identification process of mobile subscriber's relationship type is as follows:
Step 1: data acquisition and language data process unit carry out the text of user's communication and adopting for voice class data Collection, and complete voice communication data and be converted into input of the text data as normalized unit.
Step 2: normalized unit converts the textcommunication data of different-format to the system polymerizeing with user's dimension The textcommunication data of one format.
Step 3: the training of model identification parameter is carried out based on normalization communication text data, and identification model will be saved as Database.
Step 4: text data is communicated to mobile subscriber based on model identification parameter and carries out pattern-recognition, output mobile is used Family relationship type.
Above-described specific embodiment has carried out further the purpose of the present invention, technical scheme and beneficial effects It is described in detail, it should be understood that being not intended to limit the present invention the foregoing is merely a specific embodiment of the invention Protection scope, all within the spirits and principles of the present invention, any modification, equivalent substitution, improvement and etc. done should all include Within protection scope of the present invention.

Claims (6)

1. a kind of mobile subscriber's relationship type recognition methods based on communication data, which is characterized in that this method includes following step It is rapid:
Step 1: acquiring the text and voice data of user's communication by data acquisition and language data process unit, and will adopt The voice communication data of collection is converted to textcommunication data；
Step 2: the textcommunication data of different-format are converted to the system polymerizeing with user's dimension by normalized unit The textcommunication data of one format；
Step 3: carrying out the training of model identification parameter and model parameter guarantor to normalization textcommunication data by model database It deposits；
Step 4: based on the model identification parameter that identification model database provides, it is logical to mobile subscriber by pattern recognition unit Message notebook data carries out pattern-recognition, output mobile customer relationship type.
2. a kind of mobile subscriber's relationship type recognition methods based on communication data according to claim 1, feature exist In in the step 1, voice communication data is converted to textcommunication data specifically: use machine neural network based Device learning algorithm, is trained neural network model by way of supervised learning, so that it is had the function of identification voice, so It is encapsulated afterwards as fixed component, for voice communication data to be converted to textcommunication data.
3. a kind of mobile subscriber's relationship type recognition methods based on communication data according to claim 1, feature exist In the normalized unit chooses suitable characteristic parameter for different communication type, completes logical from non-structured text Letter data forms a unified text communication format to the conversion of structured text communication data.
4. a kind of mobile subscriber's relationship type recognition methods based on communication data according to claim 1, feature exist In the model database is made of multiple training servers and a parameter server, wherein the training server is responsible for The textcommunication data of input are trained, the model parameter that the training server depends on parameter server to provide, when After one training server completes training, the model parameter of update can be returned to parameter server by it, thus implementation model It updates；What is stored in the parameter server is finally obtained model parameter, and responds the request from pattern recognition unit, And required model parameter is passed into pattern recognition unit, so that it is completed identification mission.
5. a kind of mobile subscriber's relationship type recognition methods based on communication data according to claim 1, feature exist In the pattern recognition unit uses natural language processing technique, carries out unfolding calculation, expansion to a recurrent neural network A word in one sentence of each node processing afterwards, the output corresponding relationship type of the last one word of each sentence Output.
6. a kind of mobile subscriber's relationship type recognition methods based on communication data according to claim 5, feature exist In carrying out unfolding calculation to a recurrent neural network, detailed process is as follows:
xtIndicate the input of t (t=1,2,3...) step；
stFor the state that the t of hidden layer is walked, it is the memory unit of network；stAccording to current input layer and previous step hidden layer State calculated: st=f (Uxt+Wst-1), wherein f is nonlinear activation primitive, is calculating s0When, by s-1Be set to 0 to Amount；
otIt is the output of t step, it carries out softmax transformation: o to the output of a upper layer networkt=soft max (Vst)。
CN201810949662.1A 2018-08-20 2018-08-20 A kind of mobile subscriber's relationship type recognition methods based on communication data Pending CN109241241A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201810949662.1A CN109241241A (en) 2018-08-20 2018-08-20 A kind of mobile subscriber's relationship type recognition methods based on communication data

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201810949662.1A CN109241241A (en) 2018-08-20 2018-08-20 A kind of mobile subscriber's relationship type recognition methods based on communication data

Publications (1)

Publication Number Publication Date
CN109241241A true CN109241241A (en) 2019-01-18

Family

ID=65071015

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201810949662.1A Pending CN109241241A (en) 2018-08-20 2018-08-20 A kind of mobile subscriber's relationship type recognition methods based on communication data

Country Status (1)

Country Link
CN (1) CN109241241A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111652451A (en) * 2020-08-06 2020-09-11 腾讯科技（深圳）有限公司 Social relationship obtaining method and device and storage medium
CN113780107A (en) * 2021-08-24 2021-12-10 电信科学技术第五研究所有限公司 Radio signal detection method based on deep learning dual-input network model

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102682769A (en) * 2011-03-07 2012-09-19 埃森哲环球服务有限公司 Natural language-based control of digital network
CN106062786A (en) * 2014-09-12 2016-10-26 微软技术许可有限责任公司 Computing system for training neural networks
CN108280115A (en) * 2017-10-24 2018-07-13 腾讯科技（深圳）有限公司 Identify the method and device of customer relationship

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102682769A (en) * 2011-03-07 2012-09-19 埃森哲环球服务有限公司 Natural language-based control of digital network
CN106062786A (en) * 2014-09-12 2016-10-26 微软技术许可有限责任公司 Computing system for training neural networks
CN108280115A (en) * 2017-10-24 2018-07-13 腾讯科技（深圳）有限公司 Identify the method and device of customer relationship

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111652451A (en) * 2020-08-06 2020-09-11 腾讯科技（深圳）有限公司 Social relationship obtaining method and device and storage medium
CN113780107A (en) * 2021-08-24 2021-12-10 电信科学技术第五研究所有限公司 Radio signal detection method based on deep learning dual-input network model
CN113780107B (en) * 2021-08-24 2024-03-01 电信科学技术第五研究所有限公司 Radio signal detection method based on deep learning dual-input network model

Similar Documents

Publication Publication Date Title
CN110188331A (en) Model training method, conversational system evaluation method, device, equipment and storage medium
CN105094315B (en) The method and apparatus of human-machine intelligence's chat based on artificial intelligence
CN104346366B (en) Extend the method and apparatus of test data
CN104408149A (en) Criminal suspect mining association method and system based on social network analysis
CN110020433A (en) A kind of industrial and commercial senior executive's name disambiguation method based on enterprise's incidence relation
CN108268441A (en) Sentence similarity computational methods and apparatus and system
WO2023045417A1 (en) Fault knowledge graph construction method and apparatus
CN106021366A (en) API (Application Programing Interface) tag recommendation method based on heterogeneous information
CN109886554A (en) Unlawful practice method of discrimination, device, computer equipment and storage medium
CN110825968A (en) Information pushing method and device, storage medium and computer equipment
CN104765729A (en) Cross-platform micro-blogging community account matching method
CN105282123A (en) Network protocol identification method and device
CN109344887A (en) Short video classification methods, system and medium based on multi-modal dictionary learning
CN111324679A (en) Method, device and system for processing address information
CN109902859B (en) Queuing peak period estimation method based on big data and machine learning algorithm
CN110019519A (en) Data processing method, device, storage medium and electronic device
CN111046213B (en) Knowledge base construction method based on image recognition
CN112528639A (en) Object recognition method and device, storage medium and electronic equipment
CN115391553B (en) Method for automatically searching time sequence knowledge graph completion model
CN109241241A (en) A kind of mobile subscriber's relationship type recognition methods based on communication data
CN111581390A (en) Knowledge graph construction method and device and electronic equipment
CN104657466A (en) Method and device for identifying user interest based on forum post features
CN110110218A (en) A kind of Identity Association method and terminal
CN111240656A (en) Efficient deep learning model deployment framework
CN117237559B (en) Digital twin city-oriented three-dimensional model data intelligent analysis method and 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: 20190118