CN106056220A - Intelligent communication platform at automobile maintenance angle - Google Patents

Abstract

An intelligent communication platform at an automobile maintenance angle belongs to the field of intelligent communication. The intelligent communication platform solves a problem that an answer of a problem solution exists in an answer library but a problem of the answer cannot be matched due to a problem of expression of the problem, and then the intelligent communication platform achieves high matching of the problem and the answer and accelerates a problem matching speed. Technical main points are in that the intelligent communication platform comprises a keyword extraction unit, an approximate word matching unit and a question and answer implementation unit; according to the keyword extraction unit, an answer library of automobile maintenance problems is established, word segmentation is performed on customer problems in the answer library, and keywords are obtained through extraction; according to the approximate word matching unit, an automotive professional corpus is used in approximate word matching, a plurality of approximate words of the keywords in the answer library are obtained, and similarity degrees of the keywords and the approximate words are obtained through calculations; and, according to the question and answer implementation unit, automatic machine question and answer implementation is performed, word segmentation is performed on a customer problem, a keyword is extracted, the extracted keyword matches keywords in the answer library, an answer is returned when matching is successful, and, when the matching is not successful, approximate words of the keyword are matched, and the keywords in the answer library are re-matched.

Description

The intelligent intercourse platform in automobile maintenance direction

Technical field

The invention belongs to intelligence communication field, relate to automobile maintenance problem automaton question and answer communication method foundation and Application.

Background technology

In recent years, increasing along with automobile pollution, the demand of vehicle failure maintenance consulting is the most increasing, tradition 4 S auto shop far away not as good as the increment of automobile, causes service not put in place due to its increment, and some 4S shops exist not Sincere behavior, therefore drawback highlights day by day.

It addition, the appearance of the automobile maintenance directionality problem man-machine communication technology of intelligence so that customer issue can be collected, group Knit technical staff customer issue is answered, form the answer storehouse corresponding to customer issue；Through substantial amounts of investigation, to problem In keyword add up, customer issue often concentrates on several aspect, most problem, can seek in answer storehouse Answer, in order to can searching problem answers quickly and precisely, need answer storehouse carries out participle and extracts keyword；And often Huge due to language system, customer issue is the most corresponding with the keyword in answer storehouse, or with keyword in answer storehouse Approximation vocabulary shows, this can reduce the accuracy rate of answer, there is a case in which, the answer of the answer that has problems in answer storehouse, but It is, due to the statement problem of problem, but cannot to mate this answer.

Summary of the invention

In order to solve the answer answered that has problems in above-mentioned answer storehouse, but, due to the statement problem of problem, but cannot The problem mating this answer, the invention provides the intelligent intercourse platform in a kind of automobile maintenance direction, realizes with this, problem with answer The matched of case, accelerates problem matching speed.

To achieve these goals, technical solution of the present invention is characterized by:

A kind of intelligent intercourse platform in automobile maintenance direction,

Beneficial effect: the present invention can answer car owner's problem about automobile maintenance method timely and effectively, and extends to The answer of the problem in terms of fault and car insurance, during problem coupling, matching speed is fast, and precision is high, it is also possible to cannot mate In the case of, carry out the approximation word coupling of key word, add the scope of question answering, it is to avoid to state problem poor due to client Different, cause has answer, the problem but cannot mated.

Accompanying drawing explanation

Fig. 1 is the flow chart that in embodiment, automaton question and answer implement step.

Detailed description of the invention

Embodiment 1: the intelligent communication method in a kind of automobile maintenance direction, comprises the steps:

S1. setting up the answer storehouse of automobile maintenance problem, by the customer issue participle in answer storehouse, and extraction obtains key Word；

S2. approximation word coupling made in automobile major class corpus, obtains multiple approximation words of keyword in answer storehouse, and calculates Obtain keyword and the similarity of approximation word；

Wherein, the step that preferably automobile major class corpus work approximation word mates is: by the term vector meter of neutral net Calculate model, automobile major class corpus is done approximation word coupling, obtains multiple approximation words of keyword in answer storehouse, and by remaining String similarity based method obtains the similarity of keyword and similar word, is retained as similar word by word maximum for similarity.

S3. automaton question and answer are implemented, and to customer issue participle, and extract keyword, are answered by the keyword match of extraction Keyword in case storehouse, returns answer when the match is successful；When mating unsuccessful, the approximation word of coupling keyword, and again mate Keyword in answer storehouse.

Wherein, the step that preferably automated machine question and answer are implemented is:

The first step: obtain customer issue, by conventional language natural method based on hidden horse theorem by customer issue participle；

Second step: obtained the keyword in participle by TF-IDF method；

3rd step: by the keyword match of the keyword in customer issue Yu answer storehouse, the match is successful then directly obtains solution The certainly answer of problem；

4th step: the 3rd step does not has that the match is successful, then obtain customer issue by the term vector computation model of neutral net In the approximation word of keyword the most again mate answer storehouse keyword, until having been resolved the answer of problem.

As the present embodiment optimal technical scheme, described customer issue participle, and the method extracting keyword it is:

The first step: calculate word frequency；

Second step: calculate inverse document frequency；

3rd step: calculate TF-IDF.

Implementing of this above-mentioned steps is:

The step of described calculating word frequency is:

One customer issue occurs in that (a₁, a₂... a_m) it being total to m word, the frequency that each word occurs in customer issue is respectively For (n₁, n₂... n_m), then word frequency TF of i-th (1≤i≤m) individual word is

$TF\left(a_i\right)=\frac{n_i}{n_1+n_2+\mathrm{...}+n_m}---\left(\mathrm{3.2.1}\right)$

The described step calculating inverse document frequency is:

Using environment by automobile major class corpus simulation language, the total number of documents in automobile major class corpus is q, The total number of documents comprising described i-th word is p, then word a_iInverse document frequency IDF be:

$IDF\left(a_i\right)=log\left(\frac{q}{p+1}\right)---\left(\mathrm{3.2.2}\right)$

The step of described calculating TF-IDF is:

TF-IDF(a_i)=TF (a_i)×IDF(a_i) (3.2.3)

By TF-IDF algorithm, for participle and keyword extraction, and in conjunction with approximation word coupling, coupling can accelerated Meanwhile, the width of problem coupling can be increased；The probability that problem is answered increases, and avoids raising speed, the most sacrificial The domestic animal defect of accuracy.

Embodiment 2: the intelligent communicating device in a kind of automobile maintenance direction, including:

Module is set up in answer storehouse, sets up the answer storehouse of automobile maintenance problem, by the customer issue participle in answer storehouse, and carries Obtain keyword；

Approximation word matching module, automobile major class corpus is made approximation word coupling, is obtained the multiple of keyword in answer storehouse Approximation word, and it is calculated keyword and the similarity of approximation word；

Automaton question and answer implement module, and automaton question and answer are implemented, and to customer issue participle, and extract keyword, will Keyword in the keyword match answer storehouse extracted, returns answer when the match is successful；When mating unsuccessful, coupling keyword Approximation word, and again mate the keyword in answer storehouse.

This device is corresponding with any means technical scheme in embodiment 1, and this device can be used for performing described in embodiment 1 The technical scheme of intelligent intercourse platform in any number of automobile maintenance direction.

Embodiment 3: the present embodiment describes the method in embodiment 1 and the formation of the device technique scheme in embodiment 2 Route:

The biggest data statistics result

Before setting up said method and device or setting up system, we are for the counseling problem of thousand of Audi car owners Do a large amount of careful big data sort research, found that car owner is asked a question the several aspects being concentrated mainly on shown in table 1:

Table 1

In above-mentioned enquirement, the problem of 76% can be set up by technical director and special vehicle insurance industry specialists Answer storehouse obtains answer.

3.2. answer storehouse is set up

By the extraction to thousands of car owner's problems, we constitute special answer team and answer, define answer Storehouse, and it is extracted the key word of problem by TF-IDF algorithm, to facilitate next stage machine independently to answer the realization of technology.As Shown in table 2:

Key word	Solution	Reason accounting
			Tire pressure alerts	Check pressure of tire	60%
Tire pressure alerts	Change bearing	30%
			Tire pressure alerts	Change vehicle speed sensor	10%

TF-IDF algorithm steps is as follows:

The first step: calculate word frequency

Assume that car owner's problem occurs in that (a₁, a₂... a_m) it is total to m word, the frequency that each word occurs in car owner's problem It is respectively (n₁, n₂... n_m), then word frequency TF of i-th (1≤i≤m) individual word is

$TF\left(a_i\right)=\frac{n_i}{n_1+n_2+\mathrm{...}+n_m}---\left(\mathrm{3.2.1}\right)$

Second step: calculate inverse document frequency

In second step, the corpus that we are correlated with by an automobile, simulate the use environment of language.

It is assumed that the total number of documents in corpus is q, the total number of documents comprising this word is p, then word a_iInverse document frequency IDF is

$IDF\left(a_i\right)=log\left(\frac{q}{p+1}\right)---\left(\mathrm{3.2.2}\right)$

By 3.2.2 formula, if a word is the most common, then denominator is the biggest, inverse document frequency IDF (a_i) more Close to 0.

3rd step: calculate TF-IDF

TF-IDF(a_i)=TF (a_i)×IDF(a_i) (3.2.3)

By 3.2.3 formula it will be seen that TF-IDF and word occurrence number in a document is directly proportional, with this word whole Occurrence number in individual language is inversely proportional to.So, the algorithm automatically extracting key word is exactly the TF-of each word calculating document IDF value, arranges the most in descending order, takes and comes several words of foremost as key word.

3.3. corpus coupling approximation word

By the term vector computation model of neutral net, automobile major class corpus has been done approximation word coupling by us, Arrive multiple approximation words of keyword in answer storehouse, and it is similar with similar word to have obtained keyword by cosine similarity method Degree, retains word maximum for similarity as similar word.As tire pressure warning can obtain approximation vocabulary as shown in table 3, we Retain " abnormal tyre pressure " similar word as " tire pressure warning "

Keyword	Similar word	Similarity
			Tire pressure alerts	Abnormal tyre pressure	84%
Tire pressure alerts	Tire pressure is high	79%
			Tire pressure alerts	Tire pressure	72%

Table 3

3.4. automaton question answering system implements step

The first step: after obtaining customer issue, by conventional language natural method based on hidden horse theorem by customer issue participle；

Second step: obtained the key word in participle by TF-IDF method；

3rd step: by the keyword match of the key word in customer issue Yu answer storehouse, if the match is successful, directly To solution；

4th step: the match is successful if the 3rd step does not has, then obtain key by the term vector computation model of neutral net The approximation word of word mates answer storehouse keyword the most again, has been resolved scheme.

Idiographic flow is shown in Fig. 1.For the technical scheme of embodiment 1 and 2, statistical results based on big data, by traditional Question-answering mode is by developing under line on line face-to-face, solves the problem that the client of 76% is proposed, meets the consulting of car owner Demand, and in the way of machine answer, also save the time of car owner, decrease the economic expenditure of car owner simultaneously.

Embodiment 4: the intelligent communication system in a kind of automobile maintenance direction, including:

Answer storehouse, described answer storehouse has automobile maintenance problem, and the customer issue in answer storehouse is by participle, and extraction obtains The keyword of customer issue；

Automobile major class corpus, provides multiple approximation words for keyword in answer storehouse, and is calculated keyword with near Similarity like word；

Coupling storehouse, implements for automaton question and answer, the problem participle proposing client, and extracts keyword, will extract Keyword match answer storehouse in keyword, when the match is successful return answer；When mating unsuccessful, the approximation of coupling keyword Word, and again mate the keyword in answer storehouse.

This system is corresponding with any means technical scheme in embodiment 1, and this device can be used for performing described in embodiment 1 The technical scheme of intelligent intercourse platform in any number of automobile maintenance direction.

Embodiment 5: a kind of TF-IDF algorithm intelligent communication system in automobile maintenance direction or the intelligence in automobile maintenance direction Application in energy intercourse platform.Its application process, detailed in Example 1-4, separately, the present embodiment also disclose approximation word coupling and TF-IDF algorithm works in coordination with answering in the intelligent communication system in automobile maintenance direction or the intelligent intercourse platform in automobile maintenance direction With.

Embodiment 6: the intelligent intercourse platform in a kind of automobile maintenance direction, including

Extract key element, set up the answer storehouse of automobile maintenance problem, by the customer issue participle in answer storehouse, and carry Obtain keyword；

Approximation word matching unit, automobile major class corpus is made approximation word coupling, is obtained the multiple of keyword in answer storehouse Approximation word, and it is calculated keyword and the similarity of approximation word；

Question and answer implementation unit, the enforcement of automaton question and answer, to customer issue participle, and extract keyword, the pass that will extract Keyword in key word coupling answer storehouse, returns answer when the match is successful；When mating unsuccessful, the approximation word of coupling keyword, And again mate the keyword in answer storehouse.

Platform described in the present embodiment, realizes the method described in embodiment 1-5 for computer.

In described extraction key element, customer issue participle, and the method extracting keyword is:

The first step: calculate word frequency；

Second step: calculate inverse document frequency；

3rd step: calculate TF-IDF.

The step of described calculating word frequency is:

One customer issue occurs in that (a₁, a₂... a_m) it being total to m word, the frequency that each word occurs in customer issue is respectively For (n₁, n₂... n_m), then word frequency TF of i-th (1≤i≤m) individual word is

$TF\left(a_i\right)=\frac{n_i}{n_1+n_2+\mathrm{...}+n_m}---\left(\mathrm{3.2.1}\right)$

The described step calculating inverse document frequency is:

Using environment by automobile major class corpus simulation language, the total number of documents in automobile major class corpus is q, The total number of documents comprising described i-th word is p, then word a_iInverse document frequency IDF be:

$IDF\left(a_i\right)=log\left(\frac{q}{p+1}\right)---\left(\mathrm{3.2.2}\right)$

The step of described calculating TF-IDF is:

TF-IDF(a_i)=TF (a_i)×IDF(a_i) (3.2.3)

Described approximation word matching unit, automobile major class corpus is made the step of approximation word coupling and is: pass through neutral net Term vector computation model, automobile major class corpus is done approximation word coupling, obtain multiple approximations of keyword in answer storehouse Word, and obtained the similarity of keyword and similar word by cosine similarity method, using word maximum for similarity as similar word Retain.

Described question and answer implementation unit, the step that automaton question and answer are implemented is:

The first step: obtain customer issue, by conventional language natural method based on hidden horse theorem by customer issue participle；

Second step: obtained the keyword in participle by TF-IDF method；

3rd step: by the keyword match of the keyword in customer issue Yu answer storehouse, the match is successful then directly obtains solution The certainly answer of problem；

4th step: the 3rd step does not has that the match is successful, then obtain customer issue by the term vector computation model of neutral net In the approximation word of keyword the most again mate answer storehouse keyword, until having been resolved the answer of problem.

The above, only the invention preferably detailed description of the invention, but the protection domain of the invention is not Being confined to this, any those familiar with the art is in the technical scope that the invention discloses, according to the present invention The technical scheme created and inventive concept thereof in addition equivalent or change, all should contain the invention protection domain it In.

Claims (5)

1. the intelligent intercourse platform in an automobile maintenance direction, it is characterised in that: include

Extract key element, set up the answer storehouse of automobile maintenance problem, by the customer issue participle in answer storehouse, and extract To keyword；

Approximation word matching unit, automobile major class corpus is made approximation word coupling, is obtained multiple approximations of keyword in answer storehouse Word, and it is calculated keyword and the similarity of approximation word；

Question and answer implementation unit, the enforcement of automaton question and answer, to customer issue participle, and extract keyword, the keyword that will extract Keyword in coupling answer storehouse, returns answer when the match is successful；When mating unsuccessful, the approximation word of coupling keyword, lay equal stress on Keyword in new coupling answer storehouse.

2. the intelligent intercourse platform in automobile maintenance direction as claimed in claim 1, it is characterised in that described extraction keyword list In unit, customer issue participle, and the method extracting keyword is:

The first step: calculate word frequency；

Second step: calculate inverse document frequency；

3rd step: calculate TF-IDF.

3. the intelligent intercourse platform in automobile maintenance direction as claimed in claim 2, it is characterised in that

The step of described calculating word frequency is:

One customer issue occurs in that (a₁, a₂... a_m) it being total to m word, the frequency that each word occurs in customer issue is respectively (n₁, n₂... n_m), then word frequency TF of i-th (1≤i≤m) individual word is

The described step calculating inverse document frequency is:

Using environment by automobile major class corpus simulation language, the total number of documents in automobile major class corpus is q, comprises The total number of documents of described i-th word is p, then word a_iInverse document frequency IDF be:

The step of described calculating TF-IDF is:

TF-IDF(a_i)=TF (a_i)×IDF(a_i) (3.2.3)。

4. the intelligent intercourse platform in the automobile maintenance direction as described in claim 1 or 4, it is characterised in that described approximation word Joining unit, automobile major class corpus is made the step of approximation word coupling and is: by the term vector computation model of neutral net, to vapour Car specialty class corpus does approximation word coupling, obtains multiple approximation words of keyword in answer storehouse, and by cosine similarity side Method obtains the similarity of keyword and similar word, is retained as similar word by word maximum for similarity.

5. the intelligent intercourse platform in the automobile maintenance direction as described in claim 1 or 4, it is characterised in that described question and answer are implemented Unit, the step that automaton question and answer are implemented is:

The first step: obtain customer issue, by conventional language natural method based on hidden horse theorem by customer issue participle；

Second step: obtained the keyword in participle by TF-IDF method；

3rd step: by the keyword match of the keyword in customer issue Yu answer storehouse, the match is successful then directly obtains solution and asks The answer of topic；

4th step: the 3rd step does not has that the match is successful, then obtained in customer issue by the term vector computation model of neutral net The approximation word of keyword mates answer storehouse keyword the most again, until having been resolved the answer of problem.