A kind of household electrical appliances apply to install the method and system that the time is applied to install in scene intelligent typingTechnical field
Scene intelligent typing, which is applied to install, the present invention relates to information extraction technology field more particularly to a kind of household electrical appliances applies to install the timeMethod and system.
Background technique
Household electrical appliances apply to install the traditional approach that the time is applied to install in scene client reservation are as follows: customers dial household electrical appliances firm telephone, after artificialPlatform receives calls, and by interpersonal interaction, completion household electrical appliances, which are visited, applies to install preregistration.
By manual service enrollment time, have the disadvantage in that
One, this mode needs household electrical appliances company to prepare a large amount of manpower to receive calls, and human cost is very high, than if any 10,000A client makes a phone call to reserve simultaneously, that household electrical appliances company needs to prepare 10,000 manpowers just to receive calls, and otherwise say can not be just by clientOften reservation.
Two, the household electrical appliances company contact staff dusk, After Hours client can not then reserve, can only be until in second day household electrical appliances companyClass, could start to reserve, be unable to satisfy 7*24 hours needs of client.
, there is artificial intelligence in the shortcomings that in order to overcome manual service, in such a way that artificial intelligence carries out voice matchKeyword extraction is carried out to carry out time reservation.But existing voice match, it is by prestoring a large amount of keyword and clientSpeech content matched, if client hits any one keyword, it is corresponding to extract the keyword that this is hitTime is as subscription time.Due to the influence of the factors such as customer voice, a time needs to match multiple keywords, such as indicates" one day " the corresponding keyword on date have " one day, No.1, hello, the good, old-style Chinese private bank of silver, uh it is good " etc..So this keywordThe voice extracting mode matched, have the defects that it is very big, it cannot be guaranteed that the correct and efficiency of time reservation.
Summary of the invention
Technical problems based on background technology, when applying to install scene intelligent typing the invention proposes a kind of household electrical appliances and apply to installBetween method.
A kind of household electrical appliances proposed by the present invention apply to install the method that the time is applied to install in scene intelligent typing, preset a plurality of corpus model,Then the customer voice and corpus Model Matching that will acquire, and subscription time is extracted according to matching result.
Preferably, comprising the following steps:
S1, a plurality of corpus model and multiple protocol names are preset, each protocol name is the text of an expression time spanWord；
S2, customer voice is obtained by verbal system, and obtain and the matched corpus model of customer voice；Then according to languageExpect that customer voice is converted to target text information by model；
S3, protocol name is obtained from the target text information, and word is extracted according to protocol name acquisition time；
S4, the time nearest for each protocol name one extraction time of screening in the target text information extract wordAs the object time word of the protocol name, and all object time morphologies of extraction are combined into extracting time information.
Preferably, agreement entitled year, the moon, week, day, day, morning, the noon, evening, when or point.
Preferably, in step S1, corpus model is obtained by conditional random field models training.
Preferably, further comprising the steps of after step s4:
S5, extracting time information is fed back into client, obtains the reply voice of client, and obtain the reply voice with clientThen reply voice is converted to reply text information according to corpus model by corresponding corpus model；
S6, judge exist in target text information with the presence or absence of protocol name, then delete extracting time information and by target textWord information update is to reply text information, then return step S3；
S7, it is not present, is then registered extracting time information as subscription time.
Preferably, step S3 specifically: slot position mark, and root are carried out to text each in target text information according to protocol nameWord is extracted according to annotation results acquisition time.
Preferably, in step S1, each corresponding preset protocol name is associated with one or more time word models, walksIn rapid S3, word is extracted by protocol name each in the time word Model Matching acquisition target text information corresponding time.
It is a kind of to apply to install the system that the time is applied to install in scene intelligent typing using household electrical appliances, comprising:
Voice terminal obtains voice messaging for conversing with client；
Corpus data library, internal preset have a plurality of corpus model；
Text conversion module is separately connected voice terminal and corpus data library, for passing through voice messaging and corpus modelMatching, voice messaging is converted into target text information；
Time extraction module is connect with text conversion module, for extracting target text according to the protocol name of internal presetTime word and composition extracting time information in information；
Control module is separately connected time extraction module, voice terminal, text conversion module, when for will acquire extractionBetween information confirmed by voice terminal and client；
Registration Module is connect with control module, for obtaining client's acknowledged extraction time as subscription time and stepping onNote.
A kind of household electrical appliances proposed by the present invention apply to install the method that the time is applied to install in scene intelligent typing, by customer voice directly withExpect Model Matching, ensure that integrality of the customer voice information in conversion process, realize connected applications scene to clientVoice is analyzed, and be ensure that customer voice carries out the accuracy of contents extraction, is avoided the accent during keyword extractionEtc. adverse effect caused by factors.
In step S2 of the invention, complete customer voice is matched by corpus model, ensure that final acquisitionTarget text information and customer voice content consistency, believed by Word Input acquisition time in target text informationBreath, advantageously ensures that the accuracy of extraction.In the present invention, by customer voice and Model Matching is expected, due to customer voiceIntegrality and content-length are conducive to improve corpus Model Matching efficiency, and it is accurate to improve matching relative to keyword matchDegree.
Detailed description of the invention
Fig. 1 is that a kind of household electrical appliances that embodiment 1 proposes apply to install the method flow diagram that the time is applied to install in scene intelligent typing；
Fig. 2 is that a kind of household electrical appliances that embodiment 2 proposes apply to install the method flow diagram that the time is applied to install in scene intelligent typing；
Fig. 3 is that the slot position in embodiment 3 marks schematic diagram.
Specific embodiment
Embodiment 1
Referring to Fig.1, a kind of household electrical appliances proposed by the present invention apply to install the method that the time is applied to install in scene intelligent typing, preset a plurality of languageExpect model, the customer voice that then will acquire and corpus Model Matching, and subscription time is extracted according to matching result.
In this way, by customer voice directly with expectation Model Matching, ensure that customer voice information exists in present embodimentIntegrality in conversion process realizes connected applications scene and analyzes customer voice, ensure that in customer voice progressHold the accuracy extracted, avoids adverse effect caused by the factors such as the accent during keyword extraction.
Household electrical appliances in present embodiment apply to install the method that the time is applied to install in scene intelligent typing, include the following steps.
S1, a plurality of corpus model and multiple protocol names are preset, each protocol name is the text of an expression time spanWord, when it is implemented, protocol name can for year, the moon, week, day, day, morning, the noon, evening, when, point.Specifically, including " dividing " in protocol nameWhen, then must include " when "；In protocol name comprising " when " when not necessarily include " dividing ".
In this step, material model is obtained by conditional random field models training.
S2, customer voice is obtained by verbal system, and obtain and the matched corpus model of customer voice.Then according to languageExpect that customer voice is converted to target text information by model.
In this step, complete customer voice is matched by corpus model, ensure that the target text finally obtainedThe consistency of the content of word information and customer voice, by the Word Input acquisition time information in target text information, favorablyIn the accuracy for guaranteeing to extract.
And in present embodiment, by customer voice and Model Matching is expected, due to the integrality and content of customer voiceLength is conducive to improve corpus Model Matching efficiency, and improve matching accuracy relative to keyword match.
S3, protocol name is obtained from the target text information, and word is extracted according to protocol name acquisition time.
In present embodiment, the conversion of target text information has uniqueness, so, it can directly pass through text in this stepRetrieval obtains each protocol name and the corresponding position of protocol name in target text information, so that the time be facilitated to extract obtaining for wordIt takes.
S4, the time nearest for each protocol name one extraction time of screening in the target text information extract wordAs the object time word of the protocol name, and all object time morphologies of extraction are combined into extracting time information.
In present embodiment, by the screening of object time word, the corresponding object time word of each protocol name ensure thatUniqueness, to guarantee the uniqueness of extracting time information.
For example, target text information is ", this afternoon, today is not all right, tomorrow afternoon ".To the target text intoRow extract when, first acquisition protocol name " day " and " noon ", then correspond to protocol name " day ", obtain the time extraction word " today " and" tomorrow ", corresponding protocol name " noon " obtain time extraction word " afternoon "；Word is extracted in target text information according to each timePosition, the object time word for obtaining protocol name " day " is " tomorrow ", and the object time word of protocol name " noon " is " afternoon "；ToIt combines two object time words " tomorrow " and " afternoon ", obtaining extracting time information is " tomorrow afternoon ".
Embodiment 2
It is further comprising the steps of relative to embodiment 1 in the present embodiment.
S5, extracting time information is fed back into client, obtains the reply voice of client, and obtain the reply voice with clientThen reply voice is converted to reply text information according to corpus model by corresponding corpus model.
S6, judge exist in target text information with the presence or absence of protocol name, then delete extracting time information and by target textWord information update is to reply text information, then return step S3.
S7, it is not present, is then registered extracting time information as subscription time.
In this way, verifying in conjunction with step S5-S7 to extracting time information, the reservation finally registered is further ensuredTime it is accurate.
Embodiment 3
In the present embodiment, relative to embodiment 1, a kind of preparation method of specific time extraction word is given.
In the present embodiment, step S3 specifically: slot position mark is carried out to text each in target text information according to protocol name,And word is extracted according to annotation results acquisition time.
For example, using BMEO mode, B indicates that slot position starts, and M is indicated among slot position, and E indicates slot position knot in the present embodimentBeam, O indicate non-slot position.With target text information, " it is that tomorrow tomorrow, you were people that oh uh this afternoon this afternoon is not all rightWork or what " for, it is as shown in Figure 3 that slot position mark is carried out to it
In this way, marked in the present embodiment according to slot position, can acquisition time extract word " today ", " afternoon ", " today "," afternoon ", " tomorrow ", " tomorrow ", wherein " today " and " tomorrow " is corresponding protocol name " day ", and " afternoon " is corresponding protocol name " noon ".
Embodiment 4
The present embodiment compares embodiment 2, and in step S1, each corresponding preset protocol name is associated with one or moreTime word model in step S3, obtains the corresponding time of each protocol name in target text information by time word Model Matching and mentionsTake word.
Also with target text information " oh uh this afternoon this afternoon it is not all right be tomorrow tomorrow you be it is artificial orWhat " for.
In the step S1 of the present embodiment, it is preset with protocol name " day ", " day ", " noon ", " morning ", corresponding protocol name " day " is defaultHaving time word model " today ", " tomorrow ", " day after tomorrow "；Corresponding protocol name " day " be preset with time word model " today ", " tomorrow "," rear day "；Corresponding protocol name " noon " is preset with time word model " morning ", " noon ", " afternoon "；Corresponding protocol name " morning " is defaultHaving time word model " morning ".
In this way, there are protocol name " day " and " noons " in first choice confirmation target text information, then by " day " in the present embodiment" noon " corresponding time word model " today ", " tomorrow ", " day after tomorrow ", " morning ", " noon ", " afternoon " are in target text informationIn matched, it is word " today " and " tomorrow " and right with protocol name " noon " to can get the extraction of corresponding with protocol name " day " timeThe time answered extracts word " afternoon ".
Embodiment 5
A kind of household electrical appliances are provided in the present embodiment and apply to install the system that the time is applied to install in scene intelligent typing, are suitable for any of the above-describedKind household electrical appliances apply to install the method that the time is applied to install in scene intelligent typing.The system includes:
Voice terminal obtains voice messaging, obtains client's language in the above method in step S2 for conversing with clientSound.
Corpus data library, internal preset have a plurality of corpus model.
Text conversion module is separately connected voice terminal and corpus data library, for passing through voice messaging and corpus modelMatching, voice messaging is converted into target text information.
Time extraction module is connect with text conversion module, for extracting target text according to the protocol name of internal presetTime word and composition extracting time information in information.Specifically, in present embodiment, preset agreement in time extraction moduleName may include year, the moon, week, day, day, morning, the noon, evening, when, point in it is one or several.
Specifically, also corresponding to each protocol name in present embodiment, in time extraction module equipped with one or more rightThe time word answered extracts model, and time extraction module is used to extract model according to time word to protocol name in target text informationPosition retrieved, obtaining time of each protocol name in target text information extracts word, then extracts word by each timeThe time of sequencing selection rearmost position in target text information extracts object time word of the word as corresponding protocol name, soExtracting time information is formed by composite object time word afterwards.
Control module is separately connected time extraction module, voice terminal, text conversion module, when for will acquire extractionBetween information confirmed by voice terminal and client.Specifically, control module passes through voice terminal for extracting time information with voiceMode is sent to client, then obtains the reply voice information of client, and then control module is obtained back by text conversion moduleThe multiple corresponding reply text information of voice messaging, and judge to reply in text information with the presence or absence of protocol name, exist, then will replyText information regains extracting time information to time extraction module as target text information conveyance, and recycles verifying, directlyProtocol name is not included into the reply voice information of client, at this point, indicating that extracting time information is correct.
Registration Module is connect with control module, for obtaining client's acknowledged extraction time as subscription time and stepping onNote.
The above, preferable specific embodiment only of the present invention, but protection scope of the present invention not officeBe limited to this, anyone skilled in the art in the technical scope disclosed by the present invention, technology according to the present inventionScheme and its inventive concept are subject to equivalent substitution or change, should be covered by the protection scope of the present invention.