CN110059182A - Art recommended method and device towards customer service - Google Patents
Art recommended method and device towards customer service Download PDFInfo
- Publication number
- CN110059182A CN110059182A CN201910217963.XA CN201910217963A CN110059182A CN 110059182 A CN110059182 A CN 110059182A CN 201910217963 A CN201910217963 A CN 201910217963A CN 110059182 A CN110059182 A CN 110059182A
- Authority
- CN
- China
- Prior art keywords
- recommended
- words art
- art
- answer
- words
- 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
- G06F16/332—Query formulation
- G06F16/3322—Query formulation using system suggestions
-
- 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
- G06F16/35—Clustering; Classification
- G06F16/355—Class or cluster creation or modification
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
- G06Q30/0281—Customer communication at a business location, e.g. providing product or service information, consulting
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Business, Economics & Management (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Finance (AREA)
- Accounting & Taxation (AREA)
- Development Economics (AREA)
- Computational Linguistics (AREA)
- Strategic Management (AREA)
- Human Computer Interaction (AREA)
- Artificial Intelligence (AREA)
- Entrepreneurship & Innovation (AREA)
- Game Theory and Decision Science (AREA)
- Economics (AREA)
- Marketing (AREA)
- General Business, Economics & Management (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
This specification embodiment provides a kind of art recommended method and device towards customer service, method includes: first against active user's question sentence, using the information above of active user's question sentence and active user's question sentence as solicited message, by solicited message input long text disaggregated model trained in advance, the solicited message is obtained corresponding to preset probability of all categories by the output of the long text disaggregated model, and the corresponding high frequency of each classification talks about art；Then preset probability of all categories is corresponded to according to the solicited message, sorted of all categories from high to low according to the probability, using the high frequency for the preceding predetermined number classification that sorts words art as the first words art to be recommended；Finally according at least to the described first words art to be recommended, the consequently recommended words art recommended to customer service is determined, so as to make recommendation, art is stated to ask with user matches, and meets the needs of customer service.
Description
Technical field
This specification one or more embodiment is related to computer field, more particularly to the art recommended method towards customer service
And device.
Background technique
With the universal and application of mobile Internet, the diversification of terminal device, the diversification of communication modes, the network information
Also personalization, mobile, real time implementation and big data are gradually showed.For a user, more and more allegro life, shifting
The arrival of dynamic Internet age, so that people are for customer service, more stringent requirements are proposed: in time, fast and accurately understand user
Demand simultaneously provides service.In order to help customer service to be familiar with operation flow as early as possible, during customer service and user's text chat, chat window
Real-time prompting correlation talks about art beside mouthful, and customer problem is answered in auxiliary customer service, so that the speech of customer service seems more proper.
If existing in art recommended method, then the mainly conversation log of collected offline a large number of users and customer service carries out
Cluster, excavate building question and answer pair, be stored in it is spare in knowledge base, knowledge base include multiple groups question and answer pair, every group of question and answer to include pair
The typical problem and answer answered.When user and customer service chat, the typical problem in active user's question sentence and knowledge base is done into language
Adopted similarity calculation chooses the several typical problems most like with active user's question sentence and releases corresponding answer.Due to user
Question sentence is usually relatively simple, and colloquial style, is usually stated and is not asked not with user by the answer that art recommended method if existing is released
Match, is unable to satisfy the demand of customer service.
Accordingly, it would be desirable to there is improved plan, art if recommending can be made to state to ask the need for matching, and meeting customer service with user
It asks.
Summary of the invention
This specification one or more embodiment describes a kind of art recommended method and device towards customer service, can make
Art is stated to ask with user and be matched if recommendation, and meets the needs of customer service.
In a first aspect, providing one kind, towards customer service, art recommended method, method include:
For active user's question sentence, using the information above of active user's question sentence and active user's question sentence as asking
Information is sought, by solicited message input long text disaggregated model trained in advance, passes through the defeated of the long text disaggregated model
The solicited message is obtained out corresponding to preset probability of all categories, and the corresponding high frequency of each classification talks about art；
According to the solicited message correspond to preset probability of all categories, by it is of all categories according to the probability by height
To low sequence, using the high frequency for the preceding predetermined number classification that sorts words art as the first words art to be recommended；
According at least to the described first words art to be recommended, the consequently recommended words art recommended to customer service is determined.
It is described to be directed to active user's question sentence in a kind of possible embodiment, by active user's question sentence and described
The information above of active user's question sentence is as solicited message, by solicited message input long text disaggregated model trained in advance
Before, the method also includes:
From multiple answers that frequency of occurrence is more than default value in the conversation log of user and customer service, are obtained, answered multiple
Case cluster determines that the corresponding high frequency of each classification talks about art according to cluster result；
It is inputted the information above of each answer as the sample of the long text disaggregated model, by the corresponding classification of each answer
As the sample label of the long text disaggregated model, the long text disaggregated model is trained.
In a kind of possible embodiment, the high frequency words art using the preceding predetermined number classification that sorts is as the
After one words art to be recommended, the method also includes:
According to the described first words art to be recommended, retrieval obtains the described first words art to be recommended from the knowledge base pre-established
Similar answer and the corresponding typical problem of the similar answer；Wherein, the knowledge base includes multiple groups question and answer pair, and every group is asked
It answers questions including corresponding typical problem and answer；
Described first words art to be recommended typical problem corresponding with the similar answer is established into corresponding relationship, for determining
The corresponding typical problem of the consequently recommended words art.
It is described according at least to the described first words art to be recommended in a kind of possible embodiment, it determines and recommends to customer service
Consequently recommended words art before, the method also includes:
According to the solicited message, retrieval obtains preset number answer from the knowledge base pre-established, and each answers
The corresponding typical problem of case；Wherein, the knowledge base include multiple groups question and answer pair, every group of question and answer to include corresponding typical problem and
Answer；
By deep semantic model trained in advance, the matching degree of each preset number answer and the solicited message is obtained
Score；
According to the sequence of the score from high to low, selected section answer is as second from the preset number answer
Words art to be recommended；
It is described according at least to the described first words art to be recommended, determine that the consequently recommended words art recommended to customer service includes:
According to the described first words art to be recommended and the second words art to be recommended, the consequently recommended words art is determined.
Further, described according to the described first words art to be recommended and the second words art to be recommended, it determines described final
Recommend words art, comprising:
According to the pre-set priority of the described first words art to be recommended and the second words art to be recommended, from described the
The consequently recommended words art recommended to customer service is selected in one words art to be recommended and the second words art to be recommended.
Further, described according to the described first words art to be recommended and the second words art to be recommended, it determines described final
Recommend words art, comprising:
Using the described first words art to be recommended and the second words art to be recommended as the input of the decision model pre-established,
The consequently recommended words art recommended to customer service is obtained by the output of the decision model.
It is described according at least to the described first words art to be recommended in a kind of possible embodiment, it determines and recommends to customer service
Consequently recommended words art before, the method also includes:
Question and answer in active user's question sentence and the knowledge base pre-established are matched with problem to doing problem, determine institute
State typical problem similar with active user's question sentence and similarity in knowledge base；Wherein, the knowledge base is asked comprising multiple groups
It answers questions, every group of question and answer are to including corresponding typical problem and answer；
According to the sequence of similarity from high to low, the selected section from typical problem similar with active user's question sentence
Typical problem, using the corresponding answer of the part of standards problem as third words art to be recommended；
It is described according at least to the described first words art to be recommended, determine that the consequently recommended words art recommended to customer service includes:
According to the described first words art to be recommended and third words art to be recommended, the consequently recommended words art is determined.
Further, described according to the described first words art to be recommended and third words art to be recommended, it determines described final
Recommend words art, comprising:
According to the pre-set priority of the described first words art to be recommended and third words art to be recommended, from described the
The consequently recommended words art recommended to customer service is selected in one words art to be recommended and third words art to be recommended.
Further, described according to the described first words art to be recommended and third words art to be recommended, it determines described final
Recommend words art, comprising:
Using the described first words art to be recommended and third words art to be recommended as the input of the decision model pre-established,
The consequently recommended words art recommended to customer service is obtained by the output of the decision model.
Second aspect, providing one kind, art recommendation apparatus, device include: towards customer service
Taxon, for being directed to active user's question sentence, by active user's question sentence and active user's question sentence
Solicited message input long text disaggregated model trained in advance is passed through the long article as solicited message by information above
The output of this disaggregated model obtains the solicited message corresponding to preset probability of all categories, and each classification is one corresponding
High frequency talks about art；
First recommendation unit, the solicited message for being obtained according to the taxon correspond to preset each
The probability of classification sorts of all categories according to the probability from high to low, by the high frequency of preceding predetermined number classification that sorts
Art is talked about as the first words art to be recommended；
Determination unit, the first words art to be recommended for obtaining according at least to first recommendation unit are determined to customer service
The consequently recommended words art recommended.
The third aspect provides a kind of computer readable storage medium, is stored thereon with computer program, when the calculating
When machine program executes in a computer, enable computer execute first aspect method.
Fourth aspect provides a kind of calculating equipment, including memory and processor, and being stored in the memory can hold
Line code, when the processor executes the executable code, the method for realizing first aspect.
The method and apparatus provided by this specification embodiment, first against active user's question sentence, by the current use
The information above of family question sentence and active user's question sentence is as solicited message, by solicited message input length trained in advance
Textual classification model obtains the solicited message corresponding to preset all kinds of by the output of the long text disaggregated model
Other probability, the corresponding high frequency of each classification talk about art；Then corresponded to according to the solicited message preset of all categories
Probability, sort of all categories from high to low according to the probability, the high frequency of the preceding predetermined number classification that sorts talked about into art
As the first words art to be recommended；Finally according at least to the described first words art to be recommended, the consequently recommended words recommended to customer service are determined
Art.In this specification embodiment, has and combine context understanding ability, while the robustness of disaggregated model determines the model
Accuracy rate is very high, has the ability for more information above accurately being combined to recommend high frequency words art to customer service, and since words art is visitor
Common high frequency words art is taken, therefore is more easier to be received by customer service.Generally speaking, art if recommending can be made to state and ask with user
Matching, and meet the needs of customer service.
Detailed description of the invention
In order to illustrate the technical solution of the embodiments of the present invention more clearly, required use in being described below to embodiment
Attached drawing be briefly described, it should be apparent that, drawings in the following description are only some embodiments of the invention, for this
For the those of ordinary skill of field, without creative efforts, it can also be obtained according to these attached drawings others
Attached drawing.
Fig. 1 is the implement scene schematic diagram of one embodiment that this specification discloses；
Fig. 2 shows the art recommended method flow charts towards customer service according to one embodiment；
Fig. 3 shows the schematic block diagram of art recommendation apparatus towards customer service according to one embodiment；
Fig. 4 shows the schematic block diagram of art recommendation apparatus towards customer service according to another embodiment.
Specific embodiment
With reference to the accompanying drawing, the scheme provided this specification is described.
Fig. 1 is the implement scene schematic diagram of one embodiment that this specification discloses.The implement scene is related to towards customer service
If art recommend.Referring to Fig.1, chat window 11 is shown on the display screen of terminal and recommends words art list 12, when terminal receives
When the active user's question sentence inputted to user, the N wheel dialogue of the user and customer service is had existed before user's question sentence in this prior,
For example, user's question sentence 1 shown in chat window 11 and answer 1 constitute a wheel dialogue, user's question sentence 2 and answer 2 constitute another
Wheel dialogue, user's question sentence N and answer N constitute another wheel dialogue, this N wheel dialogue is referred to as the information above of active user's question sentence.
Suitable answer is made for active user's question sentence in order to facilitate customer service, terminal is according to active user's question sentence and the active user
The information above of question sentence generates and recommends words art list 12, and it includes from the knowledge base pre-established which, which talks about in art list 12,
The one or more groups of question and answer pair screened, the answer of question and answer centering is as recommending words art to recommend customer service, so that customer service can be with
Recommendation words art is directly inputted, words art will be recommended as the answer for being directed to active user's question sentence, slightly repaired alternatively, customer service can input
The recommendation words art changed, using the recommendation words art being modified slightly as the answer for being directed to active user's question sentence.
In the scene, when receiving active user's question sentence, according to active user's question sentence and active user's question sentence
Information above generates the recommendation shown to customer service and talks about art list, customer service can be helped to be familiar with operation flow as early as possible, in customer service and uses
During the text chat of family, real-time prompting correlation talks about art beside chat window, and customer problem is answered in auxiliary customer service, so that customer service
Speech seems more proper.
This specification embodiment provide the art recommended method towards customer service, can make recommend words art list in include
Recommendation words art, which is stated to ask with user, to be matched, and meets the needs of customer service.
Fig. 2 shows the art recommended method flow chart towards customer service according to one embodiment, this method can be based on Fig. 1
Shown in application scenarios.As shown in Fig. 2, in the embodiment towards customer service art recommended method the following steps are included: step 21,
For active user's question sentence, using the information above of active user's question sentence and active user's question sentence as solicited message,
By solicited message input long text disaggregated model trained in advance, institute is obtained by the output of the long text disaggregated model
Solicited message is stated corresponding to preset probability of all categories, the corresponding high frequency of each classification talks about art；Step 22, according to institute
Solicited message is stated corresponding to preset probability of all categories, sorts, will arrange from high to low according to the probability by of all categories
The high frequency words art of the preceding predetermined number classification of sequence is as the first words art to be recommended；Step 23, according at least to described first to
Recommend words art, determines the consequently recommended words art recommended to customer service.The specific executive mode of above each step is described below.
First in step 21, for active user's question sentence, by active user's question sentence and active user's question sentence
Solicited message input long text disaggregated model trained in advance is passed through the long article as solicited message by information above
The output of this disaggregated model obtains the solicited message corresponding to preset probability of all categories, and each classification is one corresponding
High frequency talks about art.It is understood that can also include presetting multiple classifications, and determine each in this specification embodiment
The process of the corresponding high frequency words art of classification, and process trained in advance is carried out to the long text disaggregated model.
It in one example, is more than pre- from frequency of occurrence in the conversation log of user and customer service, is obtained before step 21
If multiple answers of numerical value, multiple answers are clustered, determine that the corresponding high frequency of each classification talks about art according to cluster result；It will respectively answer
The information above of case is inputted as the sample of the long text disaggregated model, using the corresponding classification of each answer as the long text
The sample label of disaggregated model is trained the long text disaggregated model.
Wherein, after by multiple answers cluster, one can be selected according to cluster result from same category of each answer
Answer talks about art as the corresponding high frequency of the category.For example, according to the number in same category of each answer including business word, selection
The maximum answer of the number talks about art as the corresponding high frequency of the category；Alternatively, according to the number of words of same category of each answer, selection
The maximum answer of the number of words talks about art as the corresponding high frequency of the category；Alternatively, whether being customer service according to same category of each answer
The answer of collection, the answer for selecting customer service to collect is as the corresponding high frequency words art of the category.
In addition, multiple groups question and answer pair may be had existed comprising more wheel question and answer before active user's question sentence, it can be preparatory
The number for the question and answer pair that the information above of setting active user's question sentence includes, or preset the information above of active user's question sentence
The number for the question sentence for including, or preset the number for the answer that the information above of active user's question sentence includes.For example, referring to figure
1, question and answer are taken turns comprising N before active user's question sentence, that is, has existed N group question and answer pair, active user's question sentence can be preset
The numbers of the information above question and answer pair that include be 2, then the information above of active user's question sentence includes user's question sentence 2, returns at this time
It answers 2, user's question sentence 1, answer 1.
Similarly, the number for the question and answer pair that the information above of answer includes can be preset, or presets answer
The number for the question sentence that information above includes, or preset the number for the answer that the information above of answer includes.
Then in step 22, preset probability of all categories is corresponded to according to the solicited message, is pressed of all categories
It sorts from high to low according to the probability, using the high frequency for the preceding predetermined number classification that sorts words art as the first words to be recommended
Art.It is understood that words art to be recommended is only the form of answer, in order to be more convenient use or choosing of the customer service to words art to be recommended
It selects, typical problem can be mixed for words art to be recommended, it is subsequent words art to be recommended and typical problem to be supplied to customer service together.
In one example, after step 22, according to the described first words art to be recommended, from the knowledge base pre-established
Retrieval obtains the similar answer and the corresponding typical problem of the similar answer of the described first words art to be recommended；Wherein, described to know
Knowing library includes multiple groups question and answer pair, and every group of question and answer are to including corresponding typical problem and answer；By the described first words art to be recommended with
The corresponding typical problem of the similar answer establishes corresponding relationship, when the described first words art to be recommended is consequently recommended words art,
For determining the corresponding typical problem of the consequently recommended words art.
Finally the consequently recommended words art recommended to customer service is determined according at least to the described first words art to be recommended in step 23.
It is understood that various ways can be taken to determine words art to be recommended, respectively using various ways in this specification embodiment
Advantage, promoted words art recommend accuracy rate.
In one example, it before step 23, according to the solicited message, is retrieved from the knowledge base pre-established
To preset number answer and the corresponding typical problem of each answer；Wherein, the knowledge base include multiple groups question and answer pair, every group
Question and answer are to including corresponding typical problem and answer；By deep semantic model trained in advance, obtains each preset number and answer
The matching degree score of case and the solicited message；According to the sequence of the score from high to low, from the preset number answer
Middle selected section answer is as the second words art to be recommended；Step 23 is specifically, according to the described first words art to be recommended and described
Two words arts to be recommended determine the consequently recommended words art.
Further, according to the pre-set preferential of the described first words art to be recommended and the second words art to be recommended
Grade selects the consequently recommended words art recommended to customer service from the described first words art to be recommended and the second words art to be recommended.Or
Person passes through using the described first words art to be recommended and the second words art to be recommended as the input of the decision model pre-established
The output of the decision model obtains the consequently recommended words art recommended to customer service.
In one example, before step 23, by asking in active user's question sentence and the knowledge base pre-established
It answers questions the problem of doing to match with problem, determines in the knowledge base typical problem similar to active user's question sentence and similar
Degree；Wherein, the knowledge base includes multiple groups question and answer pair, and every group of question and answer are to including corresponding typical problem and answer；According to similar
The sequence of degree from high to low, the selected section typical problem from typical problem similar with active user's question sentence will be described
The corresponding answer of part of standards problem is as third words art to be recommended；Step 23 is specifically, according to the described first words art to be recommended
With third words art to be recommended, the consequently recommended words art is determined.
Further, the pre-set preferential of art is talked about according to the described first words art to be recommended and the third are to be recommended
Grade selects the consequently recommended words art recommended to customer service from the described first words art to be recommended and third words art to be recommended.Or
Person passes through using the described first words art to be recommended and third words art to be recommended as the input of the decision model pre-established
The output of the decision model obtains the consequently recommended words art recommended to customer service.
It is understood that step 23 can also be, according to the described first words art to be recommended, described second recommend words art and
The third words art to be recommended, determines the consequently recommended words art.
Further, according to the described first words art, the second recommendation words art and third words art to be recommended to be recommended
Pre-set priority, recommend words art and third words art to be recommended from the described first words art to be recommended, described second
The middle consequently recommended words art for selecting to recommend to customer service.Alternatively, by the described first words art, the second recommendation words art and institute to be recommended
Input of the third words art to be recommended as the decision model pre-established is stated, is obtained by the output of the decision model to customer service
The consequently recommended words art recommended.
The method provided by this specification embodiment, first against active user's question sentence, by active user's question sentence
Information above with active user's question sentence is as solicited message, by solicited message input long article one's duty trained in advance
Class model, by the output of the long text disaggregated model obtain the solicited message correspond to it is preset of all categories general
Rate, the corresponding high frequency of each classification talk about art；Then preset probability of all categories is corresponded to according to the solicited message,
It sorts of all categories from high to low according to the probability, using the high frequency for the preceding predetermined number classification that sorts words art as first
Words art to be recommended；Finally according at least to the described first words art to be recommended, the consequently recommended words art recommended to customer service is determined.This explanation
In book embodiment, has and combine context understanding ability, while the robustness of disaggregated model determines the accuracy rate of the model very
Height has the ability for more information above accurately being combined to recommend high frequency words art to customer service, and since words art is that customer service is common
High frequency talks about art, therefore is more easier to be received by customer service.Generally speaking, art if recommending can be made to state to ask with user and match, and
Meets the needs of customer service.
According to the embodiment of another aspect, also providing one kind, art recommendation apparatus, the device are used to execute towards customer service
The art recommended method towards customer service that this specification embodiment provides.What Fig. 3 was shown according to one embodiment towards customer service
Talk about the schematic block diagram of art recommendation apparatus.As shown in figure 3, the device 300 includes:
Taxon 31, for being directed to active user's question sentence, by active user's question sentence and active user's question sentence
Information above as solicited message, by solicited message input long text disaggregated model trained in advance, pass through the length
The output of textual classification model obtains the solicited message corresponding to preset probability of all categories, each classification corresponding one
A high frequency talks about art；
First recommendation unit 32, the solicited message for being obtained according to the taxon 31, which corresponds to, to be preset
Probability of all categories, sort of all categories from high to low according to the probability, by the preceding predetermined number classification that sorts
High frequency talks about art as the first words art to be recommended；
Determination unit 33, the first words art to be recommended for being obtained according at least to first recommendation unit 32, determine to
The consequently recommended words art that customer service is recommended.
Optionally, as one embodiment, described device further include:
Cluster cell, for being directed to active user's question sentence in the taxon 31, by active user's question sentence and institute
The information above of active user's question sentence is stated as solicited message, by solicited message input long text classification mould trained in advance
Before type, from multiple answers that frequency of occurrence is more than default value in the conversation log of user and customer service, are obtained, by multiple answers
Cluster determines that the corresponding high frequency of each classification talks about art according to cluster result；
Training unit, the information above of each answer for obtaining the cluster cell is as long text classification mould
The sample of type inputs, using the corresponding classification of each answer as the sample label of the long text disaggregated model, to the long text
Disaggregated model is trained.
Optionally, as one embodiment, described device further include:
First retrieval unit, in first recommendation unit 32 by the high frequency for the preceding predetermined number classification that sorts
After art is talked about as the first words art to be recommended, according to the described first words art to be recommended, retrieved from the knowledge base pre-established
To the similar answer and the corresponding typical problem of the similar answer of the described first words art to be recommended；Wherein, the knowledge base packet
Question and answer containing multiple groups pair, every group of question and answer are to including corresponding typical problem and answer；
Associative cell, the similar answer pair for retrieving the described first words art to be recommended to first retrieval unit
The typical problem answered establishes corresponding relationship, for determining the corresponding typical problem of the consequently recommended words art.
Optionally, as one embodiment, described device further include:
Second retrieval unit, for, according at least to the described first words art to be recommended, being determined to visitor in the determination unit 33
Before taking the consequently recommended words art recommended, according to the solicited message, retrieval obtains present count from the knowledge base pre-established
Mesh answer and the corresponding typical problem of each answer；Wherein, the knowledge base includes multiple groups question and answer pair, and every group of question and answer are to packet
Include corresponding typical problem and answer；
First matching unit, for obtaining the second retrieval unit retrieval by deep semantic model trained in advance
The matching degree score of each preset number answer and the solicited message arrived；
Second recommendation unit, the sequence of score from high to low for being obtained according to first matching unit, from described
Selected section answer is as the second words art to be recommended in preset number answer；
The determination unit 33, specifically for the first words art to be recommended obtained according to first recommendation unit 32 and institute
The the second recommendation unit obtains second words art to be recommended is stated, determines the consequently recommended words art.
Further, the determination unit 33, specifically for according to the described first words art to be recommended and described second wait push away
The pre-set priority for recommending words art, selects from the described first words art to be recommended and the second words art to be recommended to customer service
The consequently recommended words art recommended.
Further, the determination unit 33 is specifically used for the described first words art to be recommended and described second to be recommended
Input of the art as the decision model pre-established is talked about, is finally pushed away by what the output of the decision model obtained recommending to customer service
Recommend words art.
Optionally, as one embodiment, described device further include:
Second matching unit, for, according at least to the described first words art to be recommended, being determined to visitor in the determination unit 33
Before taking the consequently recommended words art recommended, by the question and answer in active user's question sentence and the knowledge base that pre-establishes to doing problem
It is matched with problem, determines typical problem similar with active user's question sentence and similarity in the knowledge base；Wherein, described
Knowledge base includes multiple groups question and answer pair, and every group of question and answer are to including corresponding typical problem and answer；
Third recommendation unit, the similarity sequence from high to low for being determined according to second matching unit, from
Selected section typical problem in the similar typical problem of active user's question sentence, by the corresponding answer of the part of standards problem
As third words art to be recommended；
The determination unit 33, specifically for the first words art to be recommended obtained according to first recommendation unit 32 and institute
The third words art to be recommended that third recommendation unit obtains is stated, determines the consequently recommended words art.
Further, the determination unit 33, specifically for waiting pushing away according to the described first words art to be recommended and the third
The pre-set priority for recommending words art, selects from the described first words art to be recommended and third words art to be recommended to customer service
The consequently recommended words art recommended.
Further, the determination unit 33 is specifically used for the described first words art to be recommended and the third is to be recommended
Input of the art as the decision model pre-established is talked about, is finally pushed away by what the output of the decision model obtained recommending to customer service
Recommend words art.
The device provided by this specification embodiment, first taxon 31 are directed to active user's question sentence, work as by described in
The information above of preceding user's question sentence and active user's question sentence inputs training in advance as solicited message, by the solicited message
Long text disaggregated model, by the output of the long text disaggregated model obtain the solicited message correspond to it is preset
Probability of all categories, the corresponding high frequency of each classification talk about art；Then the first recommendation unit 32 is corresponding according to the solicited message
It in preset probability of all categories, sorts of all categories from high to low according to the probability, will sort preceding predetermined number
The high frequency words art of mesh classification is as the first words art to be recommended；Last determination unit 33 is according at least to the described first words to be recommended
Art determines the consequently recommended words art recommended to customer service.In this specification embodiment, has and combine context understanding ability, simultaneously
The robustness of disaggregated model determines that the accuracy rate of the model is very high, has and more accurately information above is combined to recommend to customer service
High frequency talks about the ability of art, and since words art is the common high frequency words art of customer service, is more easier to be received by customer service.Total comes
It says, art if recommending can be made to state to ask with user and match, and meet the needs of customer service.
Fig. 4 shows the schematic block diagram of art recommendation apparatus towards customer service according to another embodiment.Such as Fig. 4 institute
Show, the device 400 include: natural language understanding (natural language understanding, NLU) module 41, up and down
Module 46, knot are retrieved/recalled to literary management module 42, words art categorization module 44, answer recommending module 45, at semantic matches module 43
Close the answer sorting module 47 of text.
For customer problem, initially enters NLU module 41 and carry out Chinese word segmentation, syntactic analysis and information extraction, then
The context of dialogue is obtained by context management module 42.
Semantic matches module 43 mainly uses depth learning technology, with the question and answer in user's current problem and knowledge base to doing
Problem-problem matching, finds out with problem most like in knowledge base, goes out front three (TOP3) according to sequencing of similarity and ask accordingly
Topic-answer pair.
It talks about art categorization module 44 and excavates high frequency words art and corresponding contextual information, the corresponding long article one's duty of training offline
Class model.Online service part carries out long text classification using user and customer service context, chooses the classification words art conduct of TOP3
Output words art, and similar to search is done in knowledge base, choose the answer in knowledge base with currently talking about art or jaccard
Similarity factor as similar answer and releases corresponding problem greater than 0.8.Wherein, jaccard similarity factor (jaccard
Similarity coefficient) for comparing similitude and otherness between finite sample collection.
Answer recommending module 45 then use user's current problem and preceding 5 word as request, calling retrieve/recall module
46 recall preceding 10 question and answer to as candidate question and answer from knowledge base, and then this 10 answers are input in conjunction with answer above
In sorting module 47, the answer of information above and candidate that answer sorting module 47 is talked with according to epicycle is done deep semantic and is sentenced
It is disconnected, thus the problem of obtaining the score of this 10 answers and sorted according to score, exporting TOP3-answer pair.
Finally, NLU module 41 exports semantic matches module 43, words art categorization module 44 and answer recommending module 45
As a result it merges, exports answer of the final TOP3 answer as user's current problem.
In this specification embodiment, semantic matches model workable for semantic matches module 43 has semantic similarity model
(word average model, WAM), shot and long term memory network (long short-term memory network, LSTM),
Convolutional neural networks (convolutional neural network, CNN), time convolutional network (temporal
Convolutional network, TCN), ESIM, MVLSTM, MATCHPYAMID even depth semantic matches model；
The model that words art categorization module 44 can be used has textual classification model FastText, CNN, character level convolutional Neural
Network (character convolutional neural network, Char-CNN), LSTM, the shot and long term based on attention
Memory network (attention-based long short-term memory network, attention-based LSTM)
Even depth learning classification model；
RNN, LSTM, a kind of answer recommended models of combination context can be used in conjunction with answer sorting module 47 above
A kind of (deep learning-to-rank, DL2R), multi-angle question and answer recommended models (multi-view response
Selection for human-computer conversation, Multi-view), a kind of answer of combination context pushes away
Recommend model (sequential matching network, SMN) even depth learning model；
Retrieving/recall module 46 can be used keyword search with weight metric algorithm BM25, TF-IDF calculating weight etc..
In this specification embodiment, the current described content of user is not only considered, it is also contemplated that user and customer service are hereinbefore
Described content, comprehensively considers contextual information in this way, can more accurately understand that user is intended to, and by information input above
In model, customer service is recommended by the suitable answer of Automatic Model Selection and is used；Words art disaggregated model can excavate visitor offline
Art if clothes are collected or is common, and the art disaggregated model in conjunction with context is trained, have and combine context understanding ability,
The robustness of disaggregated model determines that the accuracy rate of the model is very high simultaneously, has and more accurately combines information above to customer service
Recommend the ability of high frequency words art, and since words art is the common high frequency words art of customer service, is more easier to be received by customer service.
Comprehensively consider user's current utterance and information above, can more accurate understanding user be currently intended to, and from knowledge
Recommend suitable words art in library；Multi-model integration program can integrate simple sentence semantic matches, in conjunction with art classification, knot if context
The respective advantage of answer sequence hereafter is closed, promotes the accuracy rate that words art is recommended according to Model Fusion；Not only consider to utilize knowledge
Library is done words art and is recommended, while considering that customer service high frequency talks about art, so that art is easier to be received by customer service if releasing, promotes customer service and talks about art
Clicking rate.
According to the embodiment of another aspect, a kind of computer readable storage medium is also provided, is stored thereon with computer journey
Sequence enables computer execute method described in conjunction with Figure 2 when the computer program executes in a computer.
According to the embodiment of another further aspect, a kind of calculating equipment, including memory and processor, the memory are also provided
In be stored with executable code, when the processor executes the executable code, realize method described in conjunction with Figure 2.
Those skilled in the art are it will be appreciated that in said one or multiple examples, function described in the invention
It can be realized with hardware, software, firmware or their any combination.It when implemented in software, can be by these functions
Storage in computer-readable medium or as on computer-readable medium one or more instructions or code transmitted.
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 any modification, equivalent substitution, improvement and etc. on the basis of technical solution of the present invention, done should all
Including within protection scope of the present invention.
Claims (20)
1. one kind art recommended method towards customer service, which comprises
For active user's question sentence, believe the information above of active user's question sentence and active user's question sentence as request
Solicited message input long text disaggregated model trained in advance is passed through exporting for the long text disaggregated model by breath
Correspond to preset probability of all categories to the solicited message, the corresponding high frequency of each classification talks about art；
According to the solicited message correspond to preset probability of all categories, by it is of all categories according to the probability from high to low
Sequence, using the high frequency for the preceding predetermined number classification that sorts words art as the first words art to be recommended；
According at least to the described first words art to be recommended, the consequently recommended words art recommended to customer service is determined.
2. the method for claim 1, wherein described be directed to active user's question sentence, by active user's question sentence and institute
The information above of active user's question sentence is stated as solicited message, by solicited message input long text classification mould trained in advance
Before type, the method also includes:
From multiple answers that frequency of occurrence is more than default value in the conversation log of user and customer service, are obtained, multiple answers are gathered
Class determines that the corresponding high frequency of each classification talks about art according to cluster result；
Using the information above of each answer as the long text disaggregated model sample input, using the corresponding classification of each answer as
The sample label of the long text disaggregated model is trained the long text disaggregated model.
3. the method for claim 1, wherein the high frequency words art using the preceding predetermined number classification that sorts as
After first words art to be recommended, the method also includes:
According to the described first words art to be recommended, retrieval obtains the phase of the described first words art to be recommended from the knowledge base pre-established
Like answer and the corresponding typical problem of the similar answer；Wherein, the knowledge base includes multiple groups question and answer pair, every group of question and answer pair
Including corresponding typical problem and answer；
Described first words art to be recommended typical problem corresponding with the similar answer is established into corresponding relationship, described in determining
The corresponding typical problem of consequently recommended words art.
4. the method for claim 1, wherein described according at least to the described first words art to be recommended, determination is pushed away to customer service
Before the consequently recommended words art recommended, the method also includes:
According to the solicited message, retrieval obtains preset number answer and each answer pair from the knowledge base pre-established
The typical problem answered；Wherein, the knowledge base includes multiple groups question and answer pair, and every group of question and answer are to including corresponding typical problem and answering
Case；
By deep semantic model trained in advance, obtains each preset number answer and the matching degree of the solicited message obtains
Point；
According to the sequence of the score from high to low, selected section answer is as second wait push away from the preset number answer
Recommend words art；
It is described according at least to the described first words art to be recommended, determine that the consequently recommended words art recommended to customer service includes:
According to the described first words art to be recommended and the second words art to be recommended, the consequently recommended words art is determined.
5. method as claimed in claim 4, wherein described according to the described first words art to be recommended and second words to be recommended
Art determines the consequently recommended words art, comprising:
According to described first it is to be recommended words art and it is described second it is to be recommended words art pre-set priority, from described first to
Recommend to select the consequently recommended words art recommended to customer service in words art and the second words art to be recommended.
6. method as claimed in claim 4, wherein described according to the described first words art to be recommended and second words to be recommended
Art determines the consequently recommended words art, comprising:
Using the described first words art to be recommended and the second words art to be recommended as the input of the decision model pre-established, pass through
The output of the decision model obtains the consequently recommended words art recommended to customer service.
7. the method for claim 1, wherein described according at least to the described first words art to be recommended, determination is pushed away to customer service
Before the consequently recommended words art recommended, the method also includes:
Question and answer in active user's question sentence and the knowledge base pre-established are matched with problem to doing problem, determine described in know
Know typical problem similar with active user's question sentence and similarity in library；Wherein, the knowledge base includes multiple groups question and answer pair,
Every group of question and answer are to including corresponding typical problem and answer；
According to the sequence of similarity from high to low, the selected section standard from typical problem similar with active user's question sentence
Problem, using the corresponding answer of the part of standards problem as third words art to be recommended；
It is described according at least to the described first words art to be recommended, determine that the consequently recommended words art recommended to customer service includes:
According to the described first words art to be recommended and third words art to be recommended, the consequently recommended words art is determined.
8. the method for claim 7, wherein described according to the described first words art to be recommended and third words to be recommended
Art determines the consequently recommended words art, comprising:
According to described first it is to be recommended words art and the third it is to be recommended words art pre-set priority, from described first to
Recommend to select the consequently recommended words art recommended to customer service in words art and third words art to be recommended.
9. the method for claim 7, wherein described according to the described first words art to be recommended and third words to be recommended
Art determines the consequently recommended words art, comprising:
Using the described first words art to be recommended and third words art to be recommended as the input of the decision model pre-established, pass through
The output of the decision model obtains the consequently recommended words art recommended to customer service.
10. a kind of, towards customer service, art recommendation apparatus, described device include:
Taxon, for being directed to active user's question sentence, by the above of active user's question sentence and active user's question sentence
Solicited message input long text disaggregated model trained in advance is passed through the long article one's duty as solicited message by information
The output of class model obtains the solicited message corresponding to preset probability of all categories, the corresponding high frequency of each classification
Talk about art；
First recommendation unit, the solicited message for being obtained according to the taxon correspond to preset of all categories
Probability, sort of all categories from high to low according to the probability, the high frequency of the preceding predetermined number classification that sorts talked about into art
As the first words art to be recommended；
Determination unit, the first words art to be recommended for obtaining according at least to first recommendation unit determine and recommend to customer service
Consequently recommended words art.
11. device as claimed in claim 10, wherein described device further include:
Cluster cell, for being directed to active user's question sentence in the taxon, by active user's question sentence and described current
The information above of user's question sentence as solicited message, by solicited message input long text disaggregated model trained in advance it
Before, from multiple answers that frequency of occurrence is more than default value in the conversation log of user and customer service, are obtained, multiple answers are gathered
Class determines that the corresponding high frequency of each classification talks about art according to cluster result；
Training unit, the information above of each answer for obtaining the cluster cell is as the long text disaggregated model
Sample input classifies to the long text using the corresponding classification of each answer as the sample label of the long text disaggregated model
Model is trained.
12. device as claimed in claim 10, wherein described device further include:
First retrieval unit, for making the high frequency words art for the preceding predetermined number classification that sorts in first recommendation unit
After the first words art to be recommended, according to the described first words art to be recommended, retrieval obtains described from the knowledge base pre-established
The similar answer and the corresponding typical problem of the similar answer of first words art to be recommended；Wherein, the knowledge base includes multiple groups
Question and answer pair, every group of question and answer are to including corresponding typical problem and answer；
Associative cell, for the described first words art to be recommended is corresponding with the similar answer that first retrieval unit retrieves
Typical problem establishes corresponding relationship, for determining the corresponding typical problem of the consequently recommended words art.
13. device as claimed in claim 10, wherein described device further include:
Second retrieval unit, for, according at least to the described first words art to be recommended, determining and recommending to customer service in the determination unit
Consequently recommended words art before, according to the solicited message, retrieval obtains preset number and answers from the knowledge base pre-established
Case and the corresponding typical problem of each answer；Wherein, the knowledge base includes multiple groups question and answer pair, and every group of question and answer are to including corresponding to
Typical problem and answer；
First matching unit, for obtaining what second retrieval unit retrieved by deep semantic model trained in advance
The matching degree score of each preset number answer and the solicited message；
Second recommendation unit, the sequence of score from high to low for being obtained according to first matching unit are preset from described
Selected section answer is as the second words art to be recommended in number answer；
The determination unit is pushed away specifically for the first words art to be recommended obtained according to first recommendation unit and described second
The unit obtains second words art to be recommended is recommended, determines the consequently recommended words art.
14. device as claimed in claim 13, wherein the determination unit is specifically used for according to the described first words to be recommended
The pre-set priority of art and the second words art to be recommended, from the described first words art to be recommended and described second to be recommended
The consequently recommended words art recommended to customer service is selected in words art.
15. device as claimed in claim 13, wherein the determination unit is specifically used for the described first words art to be recommended
With the described second input of the words art as the decision model pre-established to be recommended, by the output of the decision model obtain to
The consequently recommended words art that customer service is recommended.
16. device as claimed in claim 10, wherein described device further include:
Second matching unit, for, according at least to the described first words art to be recommended, determining and recommending to customer service in the determination unit
Consequently recommended words art before, by the question and answer in active user's question sentence and the knowledge base pre-established to doing problem and problem
Matching, determines typical problem similar with active user's question sentence and similarity in the knowledge base；Wherein, the knowledge base
Comprising multiple groups question and answer pair, every group of question and answer are to including corresponding typical problem and answer；
Third recommendation unit, the similarity sequence from high to low for being determined according to second matching unit, from it is described
Selected section typical problem in the similar typical problem of active user's question sentence, using the corresponding answer of the part of standards problem as
Third words art to be recommended；
The determination unit is pushed away specifically for the first words art to be recommended obtained according to first recommendation unit and the third
The third words art to be recommended that unit obtains is recommended, determines the consequently recommended words art.
17. device as claimed in claim 16, wherein the determination unit is specifically used for according to the described first words to be recommended
The pre-set priority of art and third words art to be recommended, it is to be recommended from the described first words art to be recommended and the third
The consequently recommended words art recommended to customer service is selected in words art.
18. device as claimed in claim 16, wherein the determination unit is specifically used for the described first words art to be recommended
With third input of the words art as the decision model pre-established to be recommended, by the output of the decision model obtain to
The consequently recommended words art that customer service is recommended.
19. a kind of computer readable storage medium, is stored thereon with computer program, when the computer program in a computer
When execution, computer perform claim is enabled to require the method for any one of 1-9.
20. a kind of calculating equipment, including memory and processor, executable code, the processing are stored in the memory
When device executes the executable code, the method for any one of claim 1-9 is realized.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910217963.XA CN110059182A (en) | 2019-03-21 | 2019-03-21 | Art recommended method and device towards customer service |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910217963.XA CN110059182A (en) | 2019-03-21 | 2019-03-21 | Art recommended method and device towards customer service |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110059182A true CN110059182A (en) | 2019-07-26 |
Family
ID=67316258
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910217963.XA Pending CN110059182A (en) | 2019-03-21 | 2019-03-21 | Art recommended method and device towards customer service |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110059182A (en) |
Cited By (19)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110647621A (en) * | 2019-09-27 | 2020-01-03 | 支付宝(杭州)信息技术有限公司 | Method and device for selecting dialogs in robot customer service guide conversation |
| CN110910694A (en) * | 2019-11-28 | 2020-03-24 | 大唐融合通信股份有限公司 | Intelligent customer service training system |
| CN110933239A (en) * | 2019-12-30 | 2020-03-27 | 秒针信息技术有限公司 | Method and apparatus for detecting dialect |
| CN111104505A (en) * | 2019-12-30 | 2020-05-05 | 浙江阿尔法人力资源有限公司 | Information prompting method, device, equipment and storage medium |
| CN111143537A (en) * | 2019-12-30 | 2020-05-12 | 税友软件集团股份有限公司 | Service method, device, equipment and medium based on intelligent customer service system |
| CN111224863A (en) * | 2019-12-10 | 2020-06-02 | 平安国际智慧城市科技股份有限公司 | Session task generation method and device, computer equipment and storage medium |
| CN111291172A (en) * | 2020-03-05 | 2020-06-16 | 支付宝(杭州)信息技术有限公司 | Method and device for processing text |
| CN111340233A (en) * | 2020-02-20 | 2020-06-26 | 支付宝(杭州)信息技术有限公司 | Training method and device of machine learning model, and sample processing method and device |
| CN111353028A (en) * | 2020-02-20 | 2020-06-30 | 支付宝(杭州)信息技术有限公司 | Method and device for determining customer service call cluster |
| CN111414468A (en) * | 2020-03-26 | 2020-07-14 | 支付宝(杭州)信息技术有限公司 | Method and device for selecting dialect and electronic equipment |
| CN111428023A (en) * | 2020-03-26 | 2020-07-17 | 支付宝(杭州)信息技术有限公司 | Method and device for recommending dialect and electronic equipment |
| CN111522937A (en) * | 2020-05-15 | 2020-08-11 | 支付宝(杭州)信息技术有限公司 | Method and device for recommending dialect and electronic equipment |
| CN111553701A (en) * | 2020-05-14 | 2020-08-18 | 支付宝(杭州)信息技术有限公司 | Session-based risk transaction determination method and device |
| CN111782793A (en) * | 2020-08-11 | 2020-10-16 | 深圳市北科瑞声科技股份有限公司 | Intelligent customer service processing method, system and equipment |
| CN112866491A (en) * | 2020-12-31 | 2021-05-28 | 安徽迪科数金科技有限公司 | Multi-meaning intelligent question-answering method based on specific field |
| CN113064977A (en) * | 2020-01-02 | 2021-07-02 | 北大方正集团有限公司 | Data processing system |
| CN113392175A (en) * | 2020-03-12 | 2021-09-14 | 北京有限元科技有限公司 | Method, apparatus and storage medium for predicting dialect combinations |
| CN113779226A (en) * | 2021-09-27 | 2021-12-10 | 泰康保险集团股份有限公司 | Session requirement processing method and device, computer readable medium and electronic equipment |
| CN113821625A (en) * | 2021-10-11 | 2021-12-21 | 中国平安人寿保险股份有限公司 | Artificial intelligence based tactical recommendation method, device, equipment and medium |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104699708A (en) * | 2013-12-09 | 2015-06-10 | 中国移动通信集团北京有限公司 | Self-learning method and device for customer service robot |
| CN108170749A (en) * | 2017-12-21 | 2018-06-15 | 北京百度网讯科技有限公司 | Dialogue method, device and computer-readable medium based on artificial intelligence |
| CN108270661A (en) * | 2017-01-04 | 2018-07-10 | 北京搜狗科技发展有限公司 | The method, apparatus and equipment of a kind of information-reply |
| CN109033262A (en) * | 2018-07-09 | 2018-12-18 | 北京寻领科技有限公司 | Question and answer knowledge base update method and device |
| CN109446302A (en) * | 2018-09-25 | 2019-03-08 | 中国平安人寿保险股份有限公司 | Question and answer data processing method, device and computer equipment based on machine learning |
-
2019
- 2019-03-21 CN CN201910217963.XA patent/CN110059182A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104699708A (en) * | 2013-12-09 | 2015-06-10 | 中国移动通信集团北京有限公司 | Self-learning method and device for customer service robot |
| CN108270661A (en) * | 2017-01-04 | 2018-07-10 | 北京搜狗科技发展有限公司 | The method, apparatus and equipment of a kind of information-reply |
| CN108170749A (en) * | 2017-12-21 | 2018-06-15 | 北京百度网讯科技有限公司 | Dialogue method, device and computer-readable medium based on artificial intelligence |
| CN109033262A (en) * | 2018-07-09 | 2018-12-18 | 北京寻领科技有限公司 | Question and answer knowledge base update method and device |
| CN109446302A (en) * | 2018-09-25 | 2019-03-08 | 中国平安人寿保险股份有限公司 | Question and answer data processing method, device and computer equipment based on machine learning |
Cited By (27)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110647621A (en) * | 2019-09-27 | 2020-01-03 | 支付宝(杭州)信息技术有限公司 | Method and device for selecting dialogs in robot customer service guide conversation |
| CN110910694A (en) * | 2019-11-28 | 2020-03-24 | 大唐融合通信股份有限公司 | Intelligent customer service training system |
| CN111224863B (en) * | 2019-12-10 | 2021-06-22 | 平安国际智慧城市科技股份有限公司 | Session task generation method and device, computer equipment and storage medium |
| CN111224863A (en) * | 2019-12-10 | 2020-06-02 | 平安国际智慧城市科技股份有限公司 | Session task generation method and device, computer equipment and storage medium |
| CN111104505B (en) * | 2019-12-30 | 2023-08-25 | 浙江阿尔法人力资源有限公司 | Information prompting method, device, equipment and storage medium |
| CN111143537A (en) * | 2019-12-30 | 2020-05-12 | 税友软件集团股份有限公司 | Service method, device, equipment and medium based on intelligent customer service system |
| CN111104505A (en) * | 2019-12-30 | 2020-05-05 | 浙江阿尔法人力资源有限公司 | Information prompting method, device, equipment and storage medium |
| CN110933239A (en) * | 2019-12-30 | 2020-03-27 | 秒针信息技术有限公司 | Method and apparatus for detecting dialect |
| CN113064977A (en) * | 2020-01-02 | 2021-07-02 | 北大方正集团有限公司 | Data processing system |
| CN111340233A (en) * | 2020-02-20 | 2020-06-26 | 支付宝(杭州)信息技术有限公司 | Training method and device of machine learning model, and sample processing method and device |
| CN111353028A (en) * | 2020-02-20 | 2020-06-30 | 支付宝(杭州)信息技术有限公司 | Method and device for determining customer service call cluster |
| CN111340233B (en) * | 2020-02-20 | 2022-08-09 | 支付宝(杭州)信息技术有限公司 | Training method and device of machine learning model, and sample processing method and device |
| CN111353028B (en) * | 2020-02-20 | 2023-04-18 | 支付宝(杭州)信息技术有限公司 | Method and device for determining customer service call cluster |
| CN111291172B (en) * | 2020-03-05 | 2023-08-04 | 支付宝(杭州)信息技术有限公司 | Method and device for processing text |
| CN111291172A (en) * | 2020-03-05 | 2020-06-16 | 支付宝(杭州)信息技术有限公司 | Method and device for processing text |
| CN113392175A (en) * | 2020-03-12 | 2021-09-14 | 北京有限元科技有限公司 | Method, apparatus and storage medium for predicting dialect combinations |
| CN111428023A (en) * | 2020-03-26 | 2020-07-17 | 支付宝(杭州)信息技术有限公司 | Method and device for recommending dialect and electronic equipment |
| CN111414468A (en) * | 2020-03-26 | 2020-07-14 | 支付宝(杭州)信息技术有限公司 | Method and device for selecting dialect and electronic equipment |
| CN111414468B (en) * | 2020-03-26 | 2023-07-14 | 支付宝(杭州)信息技术有限公司 | Speaking operation selection method and device and electronic equipment |
| CN111428023B (en) * | 2020-03-26 | 2023-07-14 | 支付宝(杭州)信息技术有限公司 | Speaking recommendation method and device and electronic equipment |
| CN111553701A (en) * | 2020-05-14 | 2020-08-18 | 支付宝(杭州)信息技术有限公司 | Session-based risk transaction determination method and device |
| CN111522937B (en) * | 2020-05-15 | 2023-04-28 | 支付宝(杭州)信息技术有限公司 | Speaking recommendation method and device and electronic equipment |
| CN111522937A (en) * | 2020-05-15 | 2020-08-11 | 支付宝(杭州)信息技术有限公司 | Method and device for recommending dialect and electronic equipment |
| CN111782793A (en) * | 2020-08-11 | 2020-10-16 | 深圳市北科瑞声科技股份有限公司 | Intelligent customer service processing method, system and equipment |
| CN112866491A (en) * | 2020-12-31 | 2021-05-28 | 安徽迪科数金科技有限公司 | Multi-meaning intelligent question-answering method based on specific field |
| CN113779226A (en) * | 2021-09-27 | 2021-12-10 | 泰康保险集团股份有限公司 | Session requirement processing method and device, computer readable medium and electronic equipment |
| CN113821625A (en) * | 2021-10-11 | 2021-12-21 | 中国平安人寿保险股份有限公司 | Artificial intelligence based tactical recommendation method, device, equipment and medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110059182A (en) | Art recommended method and device towards customer service | |
| CN107609101B (en) | Intelligent interaction method, equipment and storage medium | |
| CN110869969B (en) | Virtual assistant for generating personalized responses within a communication session | |
| CN106649704B (en) | Intelligent conversation control method and system | |
| CN107797984B (en) | Intelligent interaction method, equipment and storage medium | |
| CN110543552B (en) | Conversation interaction method and device and electronic equipment | |
| CN108153800B (en) | Information processing method, information processing apparatus, and recording medium | |
| CN111639168B (en) | Multi-round dialogue processing method and device, electronic equipment and storage medium | |
| CN107688398B (en) | It determines the method and apparatus of candidate input and inputs reminding method and device | |
| CN109033257A (en) | Talk about art recommended method, device, computer equipment and storage medium | |
| CN107818781A (en) | Intelligent interactive method, equipment and storage medium | |
| CN109299344A (en) | The generation method of order models, the sort method of search result, device and equipment | |
| CN111414462B (en) | Dialogue statement determination method, device, computer equipment and medium | |
| CN108038165B (en) | Information searching method, searching device, terminal equipment and server equipment | |
| CN111179929B (en) | Voice processing method and device | |
| CN111078856B (en) | Group chat conversation processing method and device and electronic equipment | |
| CN110516057B (en) | Petition question answering method and device | |
| CN111460120A (en) | Conversation management method, device, equipment and storage medium | |
| CN113806503A (en) | Dialog fusion method, device and equipment | |
| CN110209768B (en) | Question processing method and device for automatic question answering | |
| CN115017289A (en) | Method and device for serving customers based on knowledge base | |
| CN107807949A (en) | Intelligent interactive method, equipment and storage medium | |
| CN109408621B (en) | Dialogue emotion analysis method and system | |
| CN109726002B (en) | Processing flow adjusting method and device | |
| CN113111157B (en) | Question-answer processing method, device, computer equipment and storage medium |
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 | ||
| TA01 | Transfer of patent application right |
Effective date of registration: 20200929 Address after: Cayman Enterprise Centre, 27 Hospital Road, George Town, Grand Cayman Islands Applicant after: Innovative advanced technology Co.,Ltd. Address before: Cayman Enterprise Centre, 27 Hospital Road, George Town, Grand Cayman Islands Applicant before: Advanced innovation technology Co.,Ltd. Effective date of registration: 20200929 Address after: Cayman Enterprise Centre, 27 Hospital Road, George Town, Grand Cayman Islands Applicant after: Advanced innovation technology Co.,Ltd. Address before: A four-storey 847 mailbox in Grand Cayman Capital Building, British Cayman Islands Applicant before: Alibaba Group Holding Ltd. |
|
| TA01 | Transfer of patent application right |