CN110222167A - A kind of method and system obtaining target criteria information - Google Patents

A kind of method and system obtaining target criteria information Download PDF

Info

Publication number
CN110222167A
CN110222167A CN201910595218.9A CN201910595218A CN110222167A CN 110222167 A CN110222167 A CN 110222167A CN 201910595218 A CN201910595218 A CN 201910595218A CN 110222167 A CN110222167 A CN 110222167A
Authority
CN
China
Prior art keywords
information
standard information
group
standard
user
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.)
Granted
Application number
CN201910595218.9A
Other languages
Chinese (zh)
Other versions
CN110222167B (en
Inventor
马良庄
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Advanced New Technologies Co Ltd
Advantageous New Technologies Co Ltd
Original Assignee
Alibaba Group Holding Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Alibaba Group Holding Ltd filed Critical Alibaba Group Holding Ltd
Priority to CN201910595218.9A priority Critical patent/CN110222167B/en
Publication of CN110222167A publication Critical patent/CN110222167A/en
Application granted granted Critical
Publication of CN110222167B publication Critical patent/CN110222167B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

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
    • G06F16/3329—Natural language query formulation or dialogue systems
    • 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)
  • Business, Economics & Management (AREA)
  • Theoretical Computer Science (AREA)
  • Finance (AREA)
  • Development Economics (AREA)
  • Mathematical Physics (AREA)
  • Strategic Management (AREA)
  • Accounting & Taxation (AREA)
  • General Physics & Mathematics (AREA)
  • Marketing (AREA)
  • General Business, Economics & Management (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Economics (AREA)
  • Artificial Intelligence (AREA)
  • Human Computer Interaction (AREA)
  • Game Theory and Decision Science (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Databases & Information Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

This specification embodiment is related to a kind of method and system for obtaining target criteria information, belongs to field of artificial intelligence.Method includes the following steps: the problem of obtaining target user；First group of standard information is determined based on target user's problem and the first preset algorithm；Obtain the track factor of target user；First kind information relevant to standard information is determined based on the target user track factor and the second preset algorithm；One or more target criteria information are determined based on first group of standard information and first kind information relevant to standard information.

Description

A kind of method and system obtaining target criteria information
Technical field
One or more embodiments of this specification are related to field of artificial intelligence, in particular to one kind automatically determines mesh Mark the method and system of standard information.
Background technique
In intelligent customer service, the information that robot inputs client is automatically processed, to determine that client wants consulting Item is intended to the relevant issues answer understood.Client usually uses the text of user's input during with robot interactive Information carries out match cognization.Text information generally comprises upper in the text question sentence in user's single-wheel dialogue, and more wheel dialogues Context information.In most cases, user may use simplified dialogue to interact during with robot interactive, This text information for causing user to input may be fuzzy text information, and it is accurate to carry out being difficult to precisely be matched to when match cognization Information.For example, user may input " checking account ", interactive system is difficult to match cognization user and specifically needs that account looked into, or checks account Range etc..It is therefore desirable to provide a kind of more accurate question and answer engine.
Summary of the invention
One of one or more embodiments in this specification provide a kind of method for obtaining target criteria information, feature It is, which comprises the problem of obtaining target user；First group of standard is determined based on described problem and the first preset algorithm Information；The track factor of the target user is obtained, the track factor reflects that the target user services in one or more At least one behavior on platform；Determine that the first kind is relevant to standard information based on the track factor and the second preset algorithm Information；One or more targets are determined based on first group of standard information and first kind information relevant to standard information Standard information.
In some embodiments, the method for obtaining target criteria information further includes exporting one or more target criterias Information.
In some embodiments, the method for obtaining target criteria information further include: obtain user to one or The feedback of multiple target criteria information；First preset algorithm and/or the second pre- imputation are updated according to the feedback of user Method.
In some embodiments, first preset algorithm and/or second preset algorithm include machine learning model, The machine learning model includes Bert model.
In some embodiments, first group of standard information and first kind information relevant to standard information are based on Determine that one or more target criteria information includes: based on first kind information relevant to standard information to first group of standard Information is screened, and determines one or more target criteria information；First kind information relevant to markup information includes sieve Select auxiliary information, the screening conditions of the screening auxiliary information reflection target criteria information.
In some embodiments, first kind information relevant to standard information includes second group of standard information；It is described Second group of standard information includes one or more standard information and/or its corresponding content information.
In some embodiments, described relevant to markup information based on first group of standard information and the first kind Information determines that one or more target criteria information comprise determining that the public mark of first group of standard information Yu second group of standard information Calibration information；One or more target criteria information is determined from the public standard information.
In some embodiments, first group of standard information further includes counting with its each standard information corresponding first Value；Second group of standard information further includes second value corresponding with its each standard information.
In some embodiments, first group of standard information and first kind information relevant to markup information are based on Determine that one or more target criteria information comprises determining that first group of standard information and the public standard of second group of standard information are believed Breath；The first numerical value and second value of public standard information are subjected to operation；Based on operation result, from the public standard information Middle determining one or more target criteria information.
In some embodiments, the operation includes being multiplied or being added.
One of embodiment of this specification provides a kind of system for obtaining target criteria information, the system comprises: First obtains module, the problem of for obtaining target user；First determining module, it is true based on described problem and the first preset algorithm Fixed first group of standard information；Second obtains module, for obtaining the track factor of the target user；The track factor reflection At least one behavior of the target user on one or more service platforms；Second determining module, based on the track because Son and the second preset algorithm determine first kind information relevant to standard information；Target determination module is based on first group of mark Calibration information and first kind information relevant to standard information determine one or more target criteria information.
In some embodiments, the system also includes target output modules, for exporting one or more target criterias Information.
In some embodiments, the system also includes: thirds to obtain module, for obtaining user to one or more The feedback of a target criteria information；Algorithm optimization module, for according to the feedback of user update first preset algorithm and/or Second preset algorithm.
In some embodiments, first preset algorithm and/or second preset algorithm include machine learning model, The machine learning model includes Bert model.
In some embodiments, the target determination module is also used to based on first kind letter relevant to standard information Breath screens first group of standard information, determines one or more target criteria information；The first kind and standard information phase The information of pass includes screening auxiliary information, the screening conditions of the screening auxiliary information reflection target criteria information.
In some embodiments, first kind information relevant to standard information includes second group of standard information；It is described Second group of standard information includes one or more standard information and/or its corresponding content information.
In some embodiments, the target determination module is also used to determine that first group of standard information is believed with second group of standard The public standard information of breath；One or more target criteria information is determined from the public standard information.
In some embodiments, first group of standard information further includes counting with its each standard information corresponding first Value；Second group of standard information further includes second value corresponding with its each standard information.
In some embodiments, the target determination module is also used to: determining first group of standard information and second group of standard The public standard information of information；The first numerical value and second value of public standard information are subjected to operation；Based on operation result, from One or more target criteria information is determined in the public standard information.
In some embodiments, the operation includes being multiplied or being added.
One of embodiment of this specification provides a kind of for obtaining the device of target criteria information, including processor and deposits Reservoir, the memory is for storing computer instruction；The processor is used to execute at least portion in the computer instruction Split instruction is to realize operation described in any one embodiment in this specification.
Detailed description of the invention
One or more embodiments of this specification will further illustrate that these are exemplary in a manner of exemplary embodiment Embodiment will be described in detail by attached drawing.These embodiments are simultaneously unrestricted, in these embodiments, are identically numbered Indicate identical structure, in which:
Fig. 1 is according to shown in some embodiments of this specification for obtaining the application scenarios signal of target criteria information Figure；
Fig. 2 is the exemplary system block diagram according to shown in this specification some embodiments；
Fig. 3 is according to shown in some embodiments of this specification for obtaining the exemplary process diagram of target criteria information；
Fig. 4 is the sub-process figure schematic diagram that target criteria information is determined according to shown in the other embodiment of this specification；
Fig. 5 is the training of the relevant machine learning model of the first preset algorithm according to shown in some embodiments of this specification Flow chart schematic diagram；
Fig. 6 is the training of the relevant machine learning model of the second preset algorithm according to shown in some embodiments of this specification Flow chart schematic diagram；
Fig. 7 is the sub-process figure signal for the track factor that target user is obtained according to shown in some embodiments of this specification Figure.
Specific embodiment
In order to illustrate more clearly of the technical solution of one or more embodiments of this specification, embodiment will be retouched below Attached drawing needed in stating is briefly described.It should be evident that the accompanying drawings in the following description is only this specification Some examples or embodiment without creative efforts, may be used also for those of ordinary skill in the art One or more embodiments of this specification are applied to other similar scenes according to these attached drawings.Unless from language environment Obviously it or separately explains, identical label represents identical structure or operation in figure.
It should be appreciated that " system " used herein, " device ", " unit " and/or " module " is for distinguishing different stage Different components, component, assembly unit, part or a kind of method of assembly.However, if other words can realize identical purpose, Then the word can be replaced by other expression.
As shown in one or more embodiments and claims of this specification, unless context clearly prompts to make an exception Situation, " one ", "one", the words such as "an" and/or "the" not refer in particular to odd number, may also comprise plural number.It is, in general, that term " comprising " and "comprising" only prompts to include the steps that clearly to identify and element, and these steps and element do not constitute one it is exclusive Property is enumerated, and method or apparatus may also include other step or element.
Used in one or more embodiments of this specification flow chart be used to illustrate according to one of this specification or Operation performed by the system of multiple embodiments.It should be understood that above or below operation not necessarily comes accurately in sequence Ground executes.On the contrary, each step can be handled according to inverted order or simultaneously.It is also possible to which other operations are added to these mistakes Cheng Zhong, or a certain step or number step operation are removed from these processes.
One or more embodiments of this specification can be applied to different customer service question answering system or search engine system Deng.Different customer service question answering systems includes but is not limited to the group of one or more of finance, shopping, trip, education, medical treatment etc. It closes.For example, shopping online customer service, bank's customer service, payment platform customer service, mall shopping customer service, ticket booking customer service, convenience service visitor Clothes, educational counseling customer service, hospital guide's customer service etc. apply the customer service question answering system of machine question and answer.Search engine system includes but unlimited It is flat in financial platform search engine, shopping platform search engine, trip platform search engine, teaching platform search engine, medical treatment The combination of one or more of platform search engine, Knowledge Sharing platform search engine etc..
The different embodiment application scenarios of one or more embodiments of this specification include but is not limited to webpage, browser The group of one or more of plug-in unit, customer end A pp, custom-built system, enterprises analysis system, artificial intelligence robot etc. It closes.It should be understood that the application scenarios of the system and method for one or more embodiments of this specification are only this explanation Some examples of one or more embodiments of book are not paying creative labor for those of ordinary skill in the art Under the premise of dynamic, one or more embodiments of this specification can also be applied to other similar scenes according to these attached drawings. For example, other similar help guides system.
Fig. 1 is the target criteria acquisition of information according to shown in one or more embodiments of this specification and presentation system 100 application scenarios schematic diagram.Target criteria acquisition of information and presentation system 100 may be used to provide such as bank's customer service and ask It answers, payment platform customer service question and answer, mall shopping customer service question and answer, ticket booking customer service question and answer, convenience service customer service question and answer, educational counseling visitor Take question and answer, hospital guide's customer service question and answer etc..Target criteria acquisition of information and presentation system 100 may include server 110, storage equipment 120, user terminal 130, network 140.
Server 110 can be configured as processing information related with the inquiry that user inputs and/or data.For example, clothes Query transformation can be query vector by business device 110.In another example server 110 can be based on preset algorithm from standard information library Determine one group of standard information.It is related with the behavior of user and/or personal attribute information that server 110 can be additionally configured to processing Information and/or data.For another example server 110 can behavior in response to user and/or personal attribute information, obtain rail The mark factor, and one group of standard information is determined based on the track factor and preset algorithm.In some embodiments, server 110 can be with It is single server or server group.The server group can be centralization or distributed (for example, server 110 can be Distributed system).In some embodiments, server 110 can be local, be also possible to long-range.For example, server 110 can access the information and/or data for being stored in user terminal 130 or storing equipment 120 via network 140.In another example clothes Business device 110 can connect user terminal 130 and/or store equipment 120 to access the information and/or data of storage.In some realities It applies in example, server 110 can be implemented in cloud platform.Only as an example, the cloud platform may include private clound, it is public Cloud, mixed cloud, community cloud, distribution clouds, internal cloud, multi layer cloud etc. or any combination thereof.
Storage equipment 120 can store data and/or instruction.For example, storage equipment 120 can store pre-generated machine Device learning model.In another example storage equipment 120 can store standard information library and/or at least two historical users input-standard Information pair and/or at least two historical tracks factor-standard information pair.In another example storage equipment 120 can store one or more A user is in the historical behavior of service platform and/or the personal attribute of user.For another example storage equipment 120 can store service Device 110 can execute or the data for executing illustrative methods described in this specification one or more embodiment and/or Instruction.In some embodiments, storage equipment 120 may include bulk storage, portable storage tank, in read-write volatile It deposits, read-only memory (ROM) etc. or any combination thereof.Illustrative bulk storage may include disk, CD, solid-state magnetic Disk etc..Exemplary removable memory may include flash drive, floppy disk, CD, storage card, compact disk, tape etc..Example Property volatile read-write memory may include random access memory (RAM).Exemplary RAM may include dynamic random access memory (DRAM), double data speed synchronous dynamic RAM (DDRSDRAM), static random access memory (SRAM), Thyristor random access memory (T-RAM) and zero capacitance random access memory (Z-RAM) etc..Exemplary read-only memory can To include mask ROM (MROM), programmable read only memory (PROM), Erasable Programmable Read Only Memory EPROM (PEROM), electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) and digital multi-function magnetic Disk read-only memory etc..In some embodiments, the storage equipment 120 can be realized in cloud platform.It is only as an example, described Cloud platform may include private clound, public cloud, mixed cloud, community cloud, distribution clouds, internal cloud, multi layer cloud etc. or any combination thereof.
User terminal 130 may include target criteria acquisition of information and the application end that system 100 is presented.User terminal 130 It can be also used for receiving the inquiry data of client or user, and output data corresponding with inquiry data.In some embodiments In, the inquiry data of user may include the enquirement or problem of user.The question formulation of user may include diversified forms, such as text This input, voice input, one of image information input etc. or the combination of various ways.In some embodiments, with inquiry The corresponding output data of data may include that and the inquiry corresponding typical problem of data and/or content corresponding with typical problem is believed Breath.In some embodiments, content information corresponding with typical problem is considered as mark and asks corresponding answer.User terminal 130 It may include any electronic equipment that user uses.In some embodiments, user terminal 130 can be mobile device 130-1, Tablet computer 130-2, laptop computer 130-3, desktop computer 130-4 etc. or any combination thereof.In some embodiments In, mobile device 130-1 may include wearable device, wisdom mobile device, virtual reality device, augmented reality device etc. or its Any combination.In some embodiments, wearable device may include smart bracelet, intelligent footgear, intelligent glasses, intelligent helmet, Smartwatch, Intelligent garment, intelligent knapsack, smart accessories etc. or any combination thereof.In some embodiments, Intelligent mobile equipment It may include smart phone, personal digital assistant (PDA), game station, navigation equipment, point of sale (POS) etc. or its any group It closes.In some embodiments, virtual reality device and/or enhanced virtual real world devices may include virtual implementing helmet, void Quasi- Reality glasses, virtual reality eyeshade, the augmented reality helmet, augmented reality glasses, augmented reality eyeshade etc. or any combination thereof. For example, virtual reality device and/or augmented reality equipment may include GoogleGlassTM, RiftConTM, FragmentsTM, GearVRTM etc..In some embodiments, desktop computer 130-4 can be car-mounted computer, vehicle mounted electric Depending on etc..
In some embodiments, user terminal 130 may include at least one network port.At least one network port It can be configured as via network 140 to target criteria acquisition of information and presentation system 100 (for example, server 110, storage are set It is standby 120) in one or more component send information and/or receive from it information.
Network 140 can promote the exchange of information and/or data.In some embodiments, target criteria acquisition of information and The one or more component of presentation system 100 (for example, server 110, storage equipment 120 and user terminal 130) can be via Other assemblies of the network 140 into the target criteria acquisition of information and presentation system 100 send information and/or data.For example, Server 110 can obtain user query from user terminal 130 via network 140.In another example server 110 can be to user Terminal 130 sends at least one standard information recommended, so that at least one standard information recommended is presented in user terminal 130. In some embodiments, network 140 can be any form of wired or wireless network, or any combination thereof.Only as an example, Network 140 may include cable network, cable network, fiber optic network, telecommunications network, internal network, internet, local area network Network (LAN), wide area network (WAN), Wireless LAN (WLAN), Metropolitan Area Network (MAN) (MAN), Public Switched Telephone Network (PSTN), Any combination of blueteeth network, ZigBee network, near-field communication (NFC) network etc. or the example above.In some embodiments, network 140 may include one or more network access point.For example, network 140 may include wired or wireless network access point, such as Base station and/or internet exchange point, and target criteria acquisition of information and the one or more component that system 100 is presented can pass through net Network exchange point is connected to network 140 to exchange data and/or information.
Fig. 2 is the exemplary system block diagram according to shown in one or more embodiments of this specification, in this specification In some embodiments, the system for obtaining target criteria information may include that the first acquisition module 202, second obtains module 204, the first determining module 206, the second determining module 208 and target determination module 210.
First acquisition module 202 can be used for the problem of obtaining target user.
Second acquisition module 204 can be used for obtaining the track factor of the target user.In some embodiments, described The track factor reflects at least one behavior of the target user on one or more service platforms.In some embodiments, The second acquisition module 204 is also used to obtain one of target user within the set time in one or more service platforms Or multiple behaviors；One or more codings are generated based on one or more of behaviors；Splice one or more of codings, obtains To the track factor.
First determining module 206 can be used for determining described problem corresponding based on described problem and the first preset algorithm One group of standard information.
Second determining module 208 can be used for determining the first kind and standard based on the track factor and the second preset algorithm The relevant information of information.
Target determination module 210 can be used for based on first group of standard information and the first kind and standard information phase The information of pass determines one or more target criteria information.In some embodiments, the target determination module 210 be also used to Few sequence that the target criteria information is determined based on first kind information relevant to standard information.In some embodiments In, the target determination module 210 is also used to believe first group of standard based on first kind information relevant to standard information Breath is screened, and determines one or more target criteria information.In some embodiments, the target determination module 210 is also used In the public standard information for determining first group of standard information Yu second group of standard information；One is determined from the public standard information A or multiple target criteria information.In some embodiments, the target determination module 210 is also used to: determining standard information；It will The first numerical value and second value of public standard information carry out operation；Based on operation result, from the public standard information really Fixed one or more target criteria information.
In some embodiments, the system can also include target output module 212, for exporting one or more mesh Mark standard information.
In some embodiments, the system can also include: that third obtains module and algorithm optimization module.The third Module is obtained for obtaining user to the feedback of one or more of target criteria information；The algorithm optimization module, is used for First preset algorithm and/or second preset algorithm are updated according to the feedback of user.
It should be appreciated that system shown in Fig. 2 and its module can use various modes to realize.For example, in some implementations In example, system and its module can be realized by the combination of hardware, software or software and hardware.Wherein, hardware components can To be realized using special logic；Software section then can store in memory, by instruction execution system appropriate, for example (,) it is micro- Processor or special designs hardware execute.It will be appreciated by those skilled in the art that meter can be used in above-mentioned method and system It calculation machine executable instruction and/or is included in the processor control code to realize, such as in such as disk, CD or DVD-ROM The programmable memory of mounting medium, such as read-only memory (firmware) or the data of such as optics or electrical signal carrier Such code is provided on carrier.The system and its module of one or more embodiments of this specification can not only have such as The semiconductor of ultra large scale integrated circuit or gate array, logic chip, transistor etc. or such as field-programmable gate array The hardware circuit of the programmable hardware device of column, programmable logic device etc. is realized, can also be used for example by various types of Software realization performed by device is managed, it can also be by combination (for example, firmware) Lai Shixian of above-mentioned hardware circuit and software.
It should be noted that the description of system and its module is shown, determined for candidate item above, only for convenience of description, One or more embodiments of this specification can not be limited within the scope of illustrated embodiment.It is appreciated that for ability It, may be without departing substantially from this principle, to modules after the principle for understanding the system for the technical staff in domain Any combination is carried out, subsystem is perhaps constituted and is connect with other modules or one or more of modules are omitted. For example, disclose in Fig. 2 first obtains module 202, second acquisition module 204, the first determining module 206, the second determining module 208, target determination module 210 and target output module 212 can be the disparate modules in a system, be also possible to a mould Block realizes the function of two or more above-mentioned modules.In some embodiments, target output module 212 can also save Slightly.In some embodiments, the first acquisition module 202, the first determining module 206 can be two modules, be also possible to one Module has the function of acquisition simultaneously and determines.For example, modules can share a memory module, modules can also divide It Ju You not respective memory module.Suchlike deformation, the protection scope in one or more embodiments of this specification Within.
It should be pointed out that " the problem of user " described in this specification, " enquirement of user ", " customer problem ", " use Put question at family " it is the identical meaning.For convenience, this specification some parts will " first kind be relevant to standard information Information " is referred to as " type I information ", they belong to the identical meaning.
Fig. 3 is according to shown in one or more embodiments of this specification for the exemplary of target criteria acquisition of information Main flow chart.As shown in figure 3, a kind of method for obtaining target criteria information may include:
Step 302, the problem of obtaining target user.In some embodiments, step 302 can obtain module by first 202 execute.
The problem of in some embodiments, including but not limited to manually being inputted the problem of the target user, One or more kinds of any groups of the problem of being inputted by voice mode, the problem of taking word mode to input by camera etc. It closes.The problem of the problem of manual mode inputs is inputs textual form, for example, target user can be in customer service dialog box The problem of it wants consulting is described using text；The problem of voice mode inputs, which can be, obtains module acquisition language by voice Sound input, and the problem of convert textual form for the problem of speech form by speech recognition module；The camera takes word side The problem of formula inputs can be the problem of obtaining written form in image by image collection module, and will be literary by Text region module The problem of the problem of word image format is converted to textual form, described image can be text screenshot, are also possible to other and have The picture of word content can also be that other have the picture of specific reference or meaning.For example, the logo figure with text or letter Piece etc. identifies logo picture using camera, so that it may determine problem content.In another example to common non-legible Logo is identified, can identify product related information corresponding with logo.The presentation mode of described problem content can be Complete sentence is also possible to incomplete sentence.Complete sentence can be interrogative sentence, declarative sentence, exclamative sentence, confirmative question etc. The form of any clause, complete sentence are also possible to any nonstandard sentences such as grammatically wrong sentence, wrong sentence.Incomplete sentence can be with It is missing from any one in sentence, phrase, phrase, word, word of primary structure etc. or two or more combinations.
In some embodiments, the problem of user, can be received by user terminal 130, and be sent to clothes by network 140 It is engaged in device 110 (such as first acquisition module 202).By taking payment platform customer service as an example, which can be to input in user and complete simultaneously After confirmation, the data of server 110 are sent to by network 140 by user terminal 130；The problem is also possible to input in user In the process, user terminal 130 is sent to the data of server 110 by network 140 in real time.Wherein, user terminal 130 passes through The data that network is sent to server 110 can be the whole of customer problem, be also possible to a part of customer problem, for example, Text-processing is carried out to customer problem, keyword therein is extracted and is sent to server.
This specification will be illustrated each processing links in this specification as example using " net quotient borrows ", need It is noted that this example only indicates a kind of possible performance of this specification, with the clearer mistake illustrated in this process Journey, the method described in this specification any embodiment or process do not play any restriction effect.
For example, it is " net quotient borrow refund " that user, which by way of voice, inputs one section of voice messaging, user terminal 130 or Person first, which obtains module 202, can be converted to content of text for the voice messaging of " net quotient, which borrows, to refund " by speech recognition module " net quotient, which borrows, to refund ", is handled to put question in follow-up process user.In other embodiments, user can also directly exist The problem of text " net quotient borrow refund " is using as target user is inputted in text box.
Step 304, first group of standard information is determined based on described problem and the first preset algorithm.In some embodiments, Step 304 can be executed by the first determining module 206.
Standard information may include typical problem, be also possible to the corresponding content information of typical problem, the content information It may include answer, can also be information relevant to typical problem.The typical problem referred to as mark is asked, is standardized ask Sentence, can reduce the ambiguity of question sentence.For example, " net quotient borrow how to refund ", " how also net quotient borrows ", " net quotient borrows to refund and how grasp Work " etc. corresponds to mark and asks " net quotient borrows and how to refund ".The corresponding content information of the typical problem may include the answer that mark is asked. The mark asks that the answer asked with mark has determining corresponding relationship, and the corresponded manner can be one-to-one correspondence, be also possible to Many-to-one correspondence can also be one-to-many correspondence.In some embodiments, the presentation mode that standard information is presented to the user May include the form of text, the form of picture, the form of voice, visual form, and link one of form etc. or A variety of combinations.Specifically, presentation mode is the typical problem and its answer of written form, user is obtained by reading word content It must solve the problems, such as its answer.Alternatively, presentation mode is the form of picture, can be used in picture with the step main points of display operation Family is achieved a solution the answer of its problem by checking picture.Or presentation mode is the typical problem of speech form and its answers Case, user are achieved a solution the answer of its problem by listening voice content.Or presentation mode is the form of video, video can Solve the process of customer problem with complete demonstration, user is achieved a solution the answer of its problem by viewing video.Also alternatively, presenting Mode is link form, and user is achieved a solution the answering of its problem by the page that clickthrough access record has user's problem answers Case.
In some embodiments, determine that first group of standard information may include base based on described problem and the first preset algorithm Relevant inquiring is carried out in the database in customer problem and then is obtained and the matched standard information of described problem.It below will be to the party Method is introduced in more detail.
In some embodiments, the problem of being based on target user, carries out the processing such as text-processing and/or semantic analysis, Determine the corresponding keyword of described problem, the keyword may include problem word (such as why, how, what being), Act (such as borrow, go back, registering, nullifying), entity (such as loan, interest, account, the amount of money, amount) and numerical value (including amount Value, temporal information etc.) one of or a variety of combinations.In some embodiments, first preset algorithm can be Data base querying algorithm.After first obtains the problem of module 302 obtains target user, the first determining module 306 asks user Topic carries out semantic analysis, extracts input of the keyword as the first preset algorithm or search algorithm in customer problem, then from Mark asks that corresponding mark is inquired in database asks and/or answer.The search algorithm may include that equivalence inquires (=), range is looked into Ask (>,<, BETWEEN, IN), fuzzy query (LIKE), intersection inquiry (AND), union inquiry (OR) etc. any one or more Combination.Search algorithm can retrieve corresponding standard information as output by inputting information from standard information database. In some embodiments, the standard information in the standard information library can have an id, title or index.Pass through this side Formula can determine first group of standard information.
For example, user inputs enquirement is " how repaying ", after the first acquisition acquisition user of module 302 puts question to, first is determining 306 pairs of enquirements of module carry out semantic analyses, extract enquirement keyword " how " and " repaying ", and keyword is passed through into query operator Method carries out match query in standard information database, these standard information are determined as first group of standard information.For example, marking Match query is carried out in calibration information database, has been obtained mark and has been asked " the refund process that net quotient borrows ", " flower refund process ", " financing Treasured how to refund " etc. and answer.
In some embodiments, the first preset algorithm can be the Processing Algorithm realized by machine learning model, corresponding Ground determines that first group of standard information may include based on customer problem and machine learning based on described problem and the first preset algorithm Model determines standard information corresponding with described problem.In some embodiments, can the problem of user directly as machine The input of learning model obtains standard information corresponding with described problem or standard information and corresponding probability value.In some realities It applies in example, can also be obtained and described problem pair using the semantic analysis result of customer problem as the input of machine learning model The standard information or standard information answered and corresponding probability value.In some embodiments, the machine learning model can be certainly Right Language Processing (NLP) model, the NLP model include but is not limited to Bert model, DSSM model, CNN-DSSM model, LSTM-DSSM model, BCNN model, ABCNN model, Hybrid CNN model etc..In some embodiments, the first determining module 206 can be used the problem of machine learning model matches user and corresponding first group of standard information.It is described below to incite somebody to action The problem of user, which is directly inputted in machine learning model, to be handled, and determines the embodiment of first group of standard information.
In some embodiments, the machine learning model may include one model, such as Bert model.Only conduct Example, the model may include feature extraction layer, classification layer and output layer.Wherein, feature extraction layer can be wrapped further again Include one of word embeding layer, context expression layer or vector presentation layer or a variety of combinations.Module 302 is obtained first to obtain After user puts question to, first, which determines that model 306 puts question to user, is input in model.The feature extraction layer of model can be by user's Problem indicates by word segmentation processing, vector expression (such as word insertion is handled) processing, context and (extracts the context relation of participle) Processing obtains each character/word in customer problem text and has merged the vector expression after full text semantic information.In some embodiments In, feature extraction layer can segment the problem of input, and the result of participle can be embedded into single word and/or short (for example, " withdrawing deposit ", " refund ") in the vector expression of language.It later, can be to according to any two in the expression of at least two vectors A adjacent vector indicates to carry out convolution, extraction contextual information.Can maximum pond convolution information to obtain the language of customer problem Adopted information.For example, can choose vector corresponding with each maximum convolution information indicates.Then each convolution is subjected to layer specification Change (Layer Normalization), and transform coding (Transformer Encode) is carried out (for example, linear to each vector Conversion；In another example vector is converted to the id of an int to represent), last Bert model can turn the customer problem of input Change the expression of the vector after having merged full text semantic information into.
Later, the classification layer in machine learning model indicates to carry out classification processing to the vector of customer problem, obtains one Or multiple standard information, last output layer export one or more of standard information.In some embodiments, classification layer can also To determine that one or more vectors with customer problem indicate matched standard information and the first numerical value, the first numerical value reflection Matching degree or the probability that will be easily selected by a user.For example, user inputs to put question to obtain module 302 for " how refunding ", first After obtaining user's enquirement, problem " how refunding " is input in preparatory trained machine learning model, the feature of model mentions It takes layer that can carry out segmenting to problem " how refunding " and the character/word after participle is converted into term vector.The classification layer of model by this A little vectors are handled, the output layer of last model can export putd question to user the standard information (as mark is asked) to match and its Probability, for example, " the refund process (probability 0.72) that net quotient borrows " " which does the mode of flower refund have? (probability 0.24) ", " how Determine the payback period that net quotient borrows? (probability 0.04) ".And the information of output is determined as first group of standard information.
In some embodiments, the machine learning model also may include multiple models.Only as an example, may include Term vector model (such as Word2Vector) and disaggregated model, wherein term vector model be used for by user's question sentence be converted to Amount expression, disaggregated model (such as Masked LM) be used for customer problem corresponding vector expression handle, obtain one or Multiple standard information.
It should be noted that described determine that first group of standard information may include true based on customer problem based on described problem Fixed corresponding one or more marks ask that the one or more marks that also can include determining that ask corresponding content information, can also be One or more mark asks about the combination of its content information.This specification does not do any restrictions.
Step 306, the track factor of the target user is obtained.In some embodiments, step 306 can be obtained by second Modulus block 204 executes.
In some embodiments, the track factor reflects the target user in one or more service platforms extremely The personal attribute information of a few behavior and/or the target user.
In some embodiments, operation behavior of the track factor reflection target user in one or more service platforms, Service platform herein may include the platform where target criteria acquisition of information and presentation system, also may include except target mark Other platforms other than platform where calibration information obtains and system is presented.For example, target criteria acquisition of information and presentation system institute Platform be shopping platform, other described platforms can be treasury management services' platform etc..Target criteria acquisition of information and presentation system It can have data interaction (for example, the available user of platform A is in platform B between platform and other described platforms where uniting On all or part of behavior record), can also be isolated from each other.In some embodiments, the behavior may include browsing, point Operations and its operation object such as hit, register, downloading, unloading, sharing, transferring accounts, withdrawing deposit, thumbing up, paying.The operation object can be with Service content, fund type of webpage, new registration including access etc..The behavior can also include the parameter information of behavior, Parameter information includes but is not limited to time parameter, numerical parameter (such as transfer amounts, Withdrawal Amount), frequency parameter etc..For example, For user after service platform is transferred accounts, the behavior includes the behavioural information transferred accounts, and further includes the temporal information transferred accounts, transfers accounts Amount information etc..In some embodiments, the personal attribute information of the user may include the essential information of user, such as user Gender, age, educational background, native place, occupation, registion time, preference, anlage or behavioural habits etc..
For example, target user has registered " net quotient borrows " account in certain service platform, the second acquisition module 204 will acquire target The behavior " registration " of user, and the object of action " net quotient borrows " of target user is obtained, and using these information as the track of user The factor determines the track factor based on these information.
In some embodiments, the action trail and/or personal attribute information of user can be obtained by user terminal 130, And server 110 (such as second acquisition module 202) is sent to by network 140.Server 110 is according to the action trail of user And/or personal attribute information obtains the track factor of user.In some embodiments, the action trail of user and/or a Genus Homo Property information be also used as history data store storage equipment in, transferred by server 110.About the acquisition user trajectory factor More descriptions, may refer to the associated description of this specification elsewhere, such as the associated description of Fig. 7.
Step 308, first kind information relevant to standard information is determined based on the track factor and the second preset algorithm. In some embodiments, step 308 can be executed by the second determining module 208.
Type I information may include screening auxiliary information, or may include second group of standard information.Accordingly, one In a little embodiments, determine that type I information may include based on the track factor and second based on the track factor and the second preset algorithm Preset algorithm determines screening conditions；In other embodiments, type I information is determined based on the track factor and the second preset algorithm It may include that the second category calibration information is determined based on the track factor and the second preset algorithm.It below will be for both embodiments point It is not described in detail.
In some embodiments, type I information may include screening auxiliary information, and the screening auxiliary information reflects mesh Mark the screening conditions of standard information.In some embodiments, screen auxiliary information or screening conditions may include range information, Conditionity information etc. plays the information of restriction effect.
In some embodiments, by screening auxiliary information or screening conditions, first group of standard information can be sieved Choosing, and then obtain one or more target criteria information.In some embodiments, obtain target user the track factor it Afterwards, the second preset algorithm can carry out keyword extraction to the track factor, obtain corresponding entity information, and then determine and be somebody's turn to do Keyword or the associated screening conditions of entity information.In some embodiments, the second preset algorithm can to the track because Son is classified, to determine the range condition of screening.
It is still illustrated using " net quotient borrows " as example, for example, target user has registered " net quotient borrows " service, corresponding target The track factor information of user may include " registration " as operation behavior, and as " net quotient borrows " clothes of object of action Business.After system obtains track factor information, the second preset algorithm can carry out relevant treatment to track factor information, extract key Word " net quotient borrows " is then based on the keyword and obtains screening conditions " borrowing related problem with net quotient ".
In some embodiments, type I information can also include second group of standard information corresponding with the track factor.? In some embodiments, it also may include that mark is asked pair that second group of standard information, which may include that mark relevant to the user trajectory factor is asked, The answer information answered can also include that mark is asked and corresponding answer information.It is corresponding, in some embodiments, based on track because Son and the second preset algorithm determine type I information, can also include determining second group based on the track factor and the second preset algorithm Standard information.Detailed description about second group of standard information is referred to the associated description of first group of standard information.
In some embodiments, the second preset algorithm may include machine learning model, for in the first preset algorithm Machine learning model is distinguished, and hereafter referred to as " factor prior model ", but should not be named as the limit to the model System." factor prior model " can be disaggregated model, regression model, be also possible to machine relevant to the first preset algorithm The identical model of learning model type.Accordingly, in some embodiments, machine learning model may include disaggregated model, i.e., Second group of standard information can be exported based on the track factor and machine learning model；In further embodiments, machine learning mould Type may include regression model, it can determine second group of standard information and its correspondence based on the track factor and machine learning model Probability value.Both embodiments will be introduced respectively below.
In some embodiments, described " factor prior model " includes disaggregated model.The disaggregated model includes but unlimited In decision tree, nearest neighbor classifier, Naive Bayes Classifier, bayesian belief network (BBN), neural network, supporting vector Machine (SVM) etc..In some embodiments, the track factor that will acquire is input to preparatory trained machine as input data In learning model, machine learning model can automatically export second group of standard information corresponding with the track factor.About The acquisition or training process of the machine learning model, can elaborate below.
For example, target user clicks on " net quotient borrow " platform and relevant with " promotion amount " link and checked related Information, the second acquisition module 204 will acquire the behavior " click " of target user, " checking ", and obtain the behavior pair of target user As " net quotient borrows " platform, " promoting amount " link and information, and using these information as the track factor of user.By the track because Son is labeled and encodes, and being converted into mark id or vector indicates, and classifies to mark id or vector expression.Pass through Disaggregated model will determine second group of standard information, such as: second group of mark ask information " how to be promoted net quotient borrow amount? ", " net quotient The upper limit of loan is how many? ", " net quotient borrow refund process ".
In some embodiments, described " factor prior model " also may include regression model, and the regression model can be with It is linear regression model (LRM), Logic Regression Models, polynomial regression model, Gradual regression analysis model, ridge regression model, lasso trick recurrence mould Type, resilient model etc..In some embodiments, described " factor prior model " includes but is not limited to Bert model, DSSM Model, CNN-DSSM model, LSTM-DSSM model, BCNN model, ABCNN model, Hybrid CNN model etc..
In some embodiments, the track factor that will acquire is input to preparatory trained engineering as input data It practises in model, machine learning model can automatically export second group of standard information and the corresponding prediction of second group of standard information Probability value.About the acquisition or training process of the machine learning model, can elaborate below.
Step 310, one is determined based on first group of standard information and first kind information relevant to standard information A or multiple target criteria information.In some embodiments, step 310 can be executed by target determination module 210.
In some embodiments, at the information of the track factor of the problem of target criteria information includes based on user and user Accurate one or more standard information that reason finally determines.It in some embodiments, can be finally determining target Standard information regards a combination as, and the combination has a presentation sequence when being presented to the user.In some embodiments, most When determining target criteria information is one eventually, a kind of combination can also be seen as.In some embodiments, described Presentation sequence may include the sequence on position, can also include temporal sequence etc..In some embodiments, the sequence The sequencing that can be exported based on algorithm is determined, is also possible to the matching degree based on each standard information or is clicked determine the probability, It is also possible to random determination.
In some embodiments, the sequence of target criteria information can be determined based on type I information.In some embodiments In, type I information includes screening conditions, is ranked up based on screening conditions to the items in first group of standard information, will be according to Determine that tactic first group of standard information is determined as target criteria information.For example, type I information include screening conditions " with Net quotient borrows related problem ", first group of standard information includes " flower refund process ", " the refund process that net quotient borrows ", " financing is precious How to refund " etc. marks ask about its answer.Therefore, the mark for meeting screening conditions can be asked preferentially according to the type I information It has been shown that, remaining mark are asked rearward.I.e., it is possible to which determining multiple target criteria information are " the refund process that net quotient borrows ", " flower refund stream The marks such as journey ", " how financing treasured refunds " ask about its answer, wherein if above-mentioned multiple target criteria information are regarded as one If combination, which just has a corresponding sequence, " the refund process that net quotient borrows " front in the embodiment, at it In his embodiment, desired sequence can also be carried out to multiple target criteria information according to other ordering rules.
In some embodiments, first group of standard information can be screened based on type I information, to determine one Or multiple target criteria information.Only as an example, type I information include screening conditions " with net quotient borrow related problem ", first Group standard information is asked about it including the marks such as " flower refund process ", " the refund process that net quotient borrows ", " how financing is precious refunds " and is answered Case.Therefore, the mark for meeting screening conditions can be asked and selected according to the type I information, remaining mark is asked, is rejected.That is, It can determine that a target criteria information is " the refund process that net quotient borrows ".
In some embodiments, type I information includes second group of standard information, accordingly, is based on first group of standard information Determine that target criteria information can also include determining based on first group of standard information and second group of standard information with type I information Then public standard information determines target criteria information from public standard information.In some embodiments, type I information is also Including second group of standard information and its corresponding second value, first group of standard information further includes corresponding with standard information first Numerical value accordingly determines that target criteria information can also include first determining public affairs based on first group of standard information and type I information Then corresponding first numerical value of public standard information and second value are carried out operation by standard information altogether, true based on operation result Set the goal standard information.
Determine that more descriptions of target criteria information may refer to about based on first group of standard information and type I information In text elsewhere, for example, the associated description of Fig. 4.
Step 312, one or more target criteria information are exported.In some embodiments, step 312 can be defeated by target Module 212 executes out.
In some embodiments, the presentation mode includes but is not limited to written form, speech form, image format, view Frequency form etc..Output is exactly that these target criteria information are exported to client, allows the user to intuitively receive.
In some embodiments, the target criteria information may include typical problem；Also may include and typical problem Corresponding content information；It can also include the combination of typical problem and its content information.In some embodiments, typical problem Content information can be understood as the corresponding answer of typical problem.
In some embodiments, when exporting target criteria information, one or more standard information can also be worked as Make a combination, this combination is exported out.When this combination the inside includes multiple standard information, a presentation is just had Sequentially.For example, can export multiple marks to user ask about its content information, the multiple mark asks about its content information can be according to The successively output of certain sequence, or in tandem according to certain sequence, the sequence can be determined by each probability value asked of marking, often A mark asks the answer text that can be directed toward the mark and ask, answer diagram or demonstration video, solves corresponding mark for providing a user The method or answer of quasi- problem, user click any one mark and ask and can check its answer content.
It should be noted that the above-mentioned description in relation to process 300 is used for the purpose of example and explanation, without limiting this explanation The scope of application of one or more embodiments of book.To those skilled in the art, in the one or more of this specification Various modifications and variations can be carried out to process 300 under the guidance of embodiment.However, these modifications and variations are still in this specification One or more embodiments within the scope of.For example, in some embodiments, it is convenient to omit step 312.In another example one It, can be with exchange step 302,304 and step 306,308 sequencing in a little embodiments.Step 302,304 and/or 306,308 It can execute, can also execute on different devices on the same device.
Fig. 4 is the sub-process figure schematic diagram that target criteria information is determined according to shown in some embodiments of this specification. Process 400 can be executed by target criteria acquisition of information and presentation system 100, specifically, can be real by target determination module 210 It is existing.The operation of process as shown below is for illustration purposes only.In some embodiments, process 400 can be added when implementing The operation bidirectional that one or more this specification one or more embodiment does not describe, and/or delete this place of one or more The operation of description.In addition, being not limiting as shown in Figure 4 with the sequence of process described below operation.
In some embodiments, the type I information includes second group of standard information, and second group of standard information includes one A or multiple typical problems and/or its corresponding content information.In some embodiments, second group of standard information with first group Standard information has intersection.Intersection herein can be understood as having identical or phase in second group of standard information and first group of standard information Close standard information.
In some embodiments, the close semantic distance that can be understood as is less than given threshold or words and phrases having the same Expression, can determine whether two or more standard information are close based on semantic distance algorithm or text matches algorithm.Some In embodiment, same or similar standard information may include identical typical problem；In some embodiments, phase Same or similar standard information may include the higher typical problem of similarity.The similarity of typical problem can be based on specific Situation is set in advance, for example, setting similarity according to the text registration in typical problem, such as is had several identical Word or word；Or identical field proportion is how many in entire typical problem.In some embodiments, same or similar Standard information may include identical answer content；In some embodiments it is possible to include in similar answer Hold.In some embodiments, similar answer content may include the higher answer content of text registration.Step 402, really Fixed first group of standard information and the public standard information in second group of standard information.In some embodiments, public standard information It can be the intersection of first group of standard information and second group of standard information, the understanding of intersection herein is referred to upper section.Some In embodiment, public standard information can also include carrying out screening it to first group of standard information and second group of standard information respectively The same or similar standard information afterwards.Screening may include the screening based on preset rules.Public standard information may include mark It asks, also may include asking corresponding answer with mark, can also include the combination of the two.In some embodiments, the determination The modes such as first group of standard information can be matched with the public standard information in second group of standard information based on equivalence, intersection is inquired It determines.For example, first group of standard information includes " flower refund process ", " the refund process that net quotient borrows ", " financing is precious how also The marks such as money " ask about its answer, type I information includes " how promoting the amount that net quotient borrows? ", " net quotient borrow the upper limit be more It is few? ", " net quotient borrow refund process ", " how the refund time that net quotient borrows determines? " equal marks ask about its answer, can determine it Intersection include " net quotient borrow refund process ", " how the refund time that net quotient borrows determines? ".
In some embodiments, can be using the option in the intersection as the public standard information, it can also be from institute State in the option of intersection that selected section is as the public standard information, wherein alternative condition can be same or similar degree. For example, in above-mentioned intersection " the refund process that net quotient borrows " than " how the refund time that net quotient borrows determines? " close degree it is high, because " the refund process that net quotient borrows " is used as the public standard information by this.
In some embodiments, first group of standard information includes the first numerical value corresponding with its each standard information, and described Two groups of standard information include second value corresponding with its each standard information.In some embodiments, first group of standard information First numerical value can be probability value, matching value or weighted value.The second value of second group of standard information is also possible to probability Value, matching value or weighted value, these numerical value can be the probability for each standard information being calculated by preset algorithm.It is described Preset algorithm can be the first preset algorithm, be also possible to the second preset algorithm.
Step 404, the first numerical value and second value of public standard information are subjected to operation.The operation can be addition Or it is multiplied.For example, include in public standard information " net quotient borrow refund process ", " how the refund time that net quotient borrows determines? " Two marks are asked, wherein first numerical value of " the refund process that net quotient borrows " from first group of standard information is 0.9, come from second group of mark The second value of calibration information is 0.7, the first numerical value is multiplied with second value, the target value for obtaining public standard information is 0.63." how the refund time that net quotient borrows determines? " the first numerical value from first group of standard information is 0.6, comes from second group The second value of standard information is 0.8, the first numerical value is multiplied with second value, the target value for obtaining public standard information is 0.48。
Step 406, it is based on operation result, one or more target criteria information is determined from the public standard information. In some embodiments, it can be ranked up based on the target value of public standard information, and before recommending ranking in ranking results The standard information of several (such as: first 3) is as target criteria information.For example, " the refund process that quotient borrows can will be netted ", " net quotient How the refund time of loan determines? " two marks ask about its answer as one or more of target criteria information.In some realities It applies in example, or the target value of public standard information sets a threshold value (such as: 0.5), when the target of public standard information When value is greater than threshold value, then target criteria information is determined it as.It is asked about for example, being only capable of this mark by " the refund process that net quotient borrows " Its answer is as one or more of target criteria information.
Fig. 5 is the training of the relevant machine learning model of the first preset algorithm according to shown in some embodiments of this specification Flow chart schematic diagram.The operation of process as shown below is for illustration purposes only.In some embodiments, process 500 is being implemented When can add the operation bidirectional that one or more this specification one or more embodiment do not describe, and/or delete one or Operation described herein above.In addition, being not limiting as shown in Figure 5 with the sequence of process described below operation.
Step 502, the problem of obtaining historical user and corresponding standard information.
The historical user may include the user data in history on service platform.It in some embodiments, can be with base In historical use data determines the historical user the problem of, the problem of historical user includes but is not limited to manually defeated The problem of entering, the problem of being inputted by voice mode, the problem of taking word mode to input by camera etc. it is one or two kinds of with On any combination.The corresponding standard information may include the standard letter that historical user clicks after the data input Breath；It also may include the standard information that historical user thumbs up after the data input；It can also include historical user in input data The most standard information of browsing time afterwards；It can also share after the data input including historical user and/or hop count is most Standard information.Both the standard information may include that mark is asked, also may include asking corresponding answer with mark, can also include Combination.
Step 504, initial first machine learning model of training.The machine learning model can be natural language processing (NLP) model, the NLP model include but is not limited to Bert model, DSSM model, CNN-DSSM model, LSTM-DSSM mould Type, BCNN model, ABCNN model, Hybrid CNN model etc..
In some embodiments, first machine learning model can be trained in the following manner and be obtained: obtain the first instruction Practice sample set；First training sample set includes the input problem of historical user, and mark corresponding with the input problem The problem of historical user obtained in calibration information, that is, step 502 and corresponding standard information.In some embodiments In, standard information corresponding with input problem may include click after historical user's input problem selection typical problem and/or Corresponding answer；It also may include the typical problem and/or corresponding answer that historical user thumbs up after input problem；It can be with The typical problem and/or corresponding answer forwarded after input problem including historical user.Use first training sample set Initial first machine learning model of training obtains trained first machine learning model.
Step 506, trained first machine learning model is obtained.Trained first machine learning model can be with In response to customer problem, in some embodiments, trained first machine learning model can be based on customer problem or user The corresponding one or more corresponding standard information of term vector output of problem, to determine first group of standard information.In some implementations In example, after first obtains the acquisition customer problem of module 202, at least one first group can be determined using the first machine learning model Standard information.First group of standard information includes at least mark and asks, asks one of corresponding answer or combinations thereof with mark.About The application method of first machine learning model can find more detailed description elsewhere in the text.
In some embodiments, (for example, standard information of prediction) and reference standard can be exported according to the prediction of model Between difference reversely adjust model parameter, to achieve the purpose that trained or Optimized model.In some embodiments, can pass through Increase the mode of sample data to optimize or more new model.In some embodiments, when difference meets a certain preset condition, example Such as, the prediction accuracy of model is greater than the small Mr. Yu of value of a certain predetermined accuracy threshold value or loss function (Loss Function) One preset value, training process will stop arriving trained first machine learning model.
In some embodiments, the gap of output valve and reference value can be judged according to the feedback of user, and then passed through Increase sample data to optimize algorithm.In some embodiments, the feedback of user can include but is not limited to user's click And/or number of clicks, browsing and/or browsing time, forwarding and/or hop count, sharing and/or share number, abandon checking Deng.
Fig. 6 is the training of the relevant machine learning model of the second preset algorithm according to shown in some embodiments of this specification Flow chart schematic diagram.The operation of process as shown below is for illustration purposes only.Process 600 can be by target criteria acquisition of information And the second determining module 208 execution of system 100 is presented.The operation of process as shown below is for illustration purposes only.Some In embodiment, process 600 can add the volume that one or more this specification one or more embodiment does not describe when implementing Outer operation, and/or delete one or more operation described herein.In addition, being operated as shown in Figure 6 with process described below Sequence be not limiting.
In some embodiments, step 308 determines the first kind and mark based on the track factor and the second preset algorithm The relevant information of calibration information can be automatically processed by machine algorithm；In some embodiments, machine algorithm can be by engineering Model treatment is practised, so that processing speed is fast, accuracy is high.
Step 602, the track factor and corresponding first kind letter relevant to standard information of historical user are obtained Breath.The track factor reflects at least one behavior of the historical user in one or more service platforms, the history User may include the existing user data on service platform.In some embodiments, described in the track factor also reflects The personal attribute information of historical user.In some embodiments, corresponding first kind information relevant to standard information or Corresponding type I information may include one or more standard information corresponding with the track factor of historical user, may be used also To include other auxiliary informations relevant to the track factor of historical user or screening conditions, screening conditions may include range Information, conditionity information etc. play the information of restriction effect.
Step 604, initial second machine learning model of training.Second machine learning model may include and the first machine The similar model of device learning model.For example, the second machine learning model includes but is not limited to Bert model.
In some embodiments, initial second machine learning model of the training includes: to obtain the second training sample set；Make Trained second machine learning model is obtained with initial second machine learning model of second training sample set training.It is described Second training sample set include the historical user obtained in step 602 the track factor and corresponding type I information.
In some embodiments, it can will be answered with track factor pair using the track factor in sample set as input data Type I information as output data or reference standard, the second machine learning model is trained with this.
Step 606, trained second machine learning model is obtained.Trained second machine learning model can be with In response to the track factor of target user, in some embodiments, trained second machine learning model can be based on target The track factor of user determines screening conditions corresponding with the track factor or standard information, to obtain determining type I information. In some embodiments, it after the second acquisition module 204 obtains the user trajectory factor, can be determined using the second machine learning model At least one first kind information relevant to standard information.In some embodiments, first kind information relevant to standard information Or type I information includes at least mark and asks, asks one of corresponding answer or combinations thereof with mark；In some embodiments, first Category information can also include screening auxiliary information, the screening conditions of the screening auxiliary information reflection target criteria information.One In a little embodiments, the screening conditions can screen first group of standard information, to determine target criteria information service.It closes In the application method of the second machine learning model, more detailed description can be found elsewhere in the text.
In some embodiments, (for example, standard information of prediction) and reference standard can be exported according to the prediction of model Between difference reversely adjust model parameter, to achieve the purpose that trained or Optimized model.In some embodiments, can pass through Increase the mode of sample data to optimize or more new model.In some embodiments, when difference meets a certain preset condition, example Such as, the prediction accuracy of model is greater than the small Mr. Yu of value of a certain predetermined accuracy threshold value or loss function (Loss Function) One preset value, training process will stop arriving trained second machine learning model.
In some embodiments, it is also possible to by obtain user to the feedbacks of one or more of target criteria information come Model parameter is adjusted, to achieve the purpose that trained or Optimized model.The feedback includes but is not limited to that user clicks and/or clicks Number, browsing and/or browsing time, forwarding and/or hop count, sharing and/or sharing number are abandoned checking.
In some embodiments, user can also be obtained to the feedback of one or more of target criteria information；According to The feedback of user updates first preset algorithm and/or second preset algorithm.The feedback of user may include but unlimited It is clicked in user and/or number of clicks, browsing and/or browsing time, forwarding and/or hop count, sharing and/or sharing is secondary Number is abandoned checking.First preset algorithm and/or the second preset algorithm will affect the sequence of target criteria information.In some realities Apply in example, updating the first preset algorithm and the/the second preset algorithm includes the new training sample of acquisition, to the first preset algorithm and/ Or second the relevant machine learning model of preset algorithm carry out retraining, improve its and export accuracy.In some embodiments, more New first preset algorithm and the/the second preset algorithm include adjusting the screening conditions (such as optimal screening range) of standard information, are mentioned The accuracy of height screening.In some embodiments, it updates the first preset algorithm and/the second preset algorithm includes adjustment and the first number Value, the corresponding threshold value of second value improve it and export accuracy.In some embodiments, the first preset algorithm and/or the are updated Two preset algorithms further include adjusting logic or method that keyword extraction is carried out to customer problem, to improve the standard of keyword acquisition True property.
Fig. 7 be in Fig. 3 obtain target user the track factor in some embodiments shown in sub-process figure schematic diagram.It crosses Journey 900 can be executed by target criteria acquisition of information and presentation system 100, can specifically be realized by the second acquisition module 204.With The operation of process shown in lower is for illustration purposes only.In some embodiments, process 700 implement when can add one or The operation bidirectional that above this specification one or more embodiment does not describe, and/or to delete one or more described herein Operation.In addition, being not limiting as shown in Figure 7 with the sequence of process described below operation.
In some embodiments, one or more behaviors of the user within the set time in service platform can be by people Work acquisition and be input in system；It is also possible to obtain by machine and is encoded.Following embodiment will be introduced logical Cross the process that machine obtains user behavior and encoded.
Step 702, target user's one or more behaviors in service platform within the set time are obtained.In some realities It applies in example, the service platform can be the platform where target criteria acquisition of information and presentation system, be also possible to and target Platform where standard information obtains and system is presented is associated, can or have permission the association platform for obtaining relevant information. In some embodiments, the behavior can be browsing, click, registration, downloading, unloading, share, transfers accounts, withdrawing deposit, thumbing up, propping up Pay etc. any user can to service platform carry out operation.The behavior can also include the parameter information of behavior, parameter letter Breath includes but is not limited to time parameter, numerical parameter, frequency parameter etc..After the setting time can be user's consummatory behavior In a certain specific time, for example, in a week, in three days, within one day, within five minutes etc..
Step 704, one or more codings are generated based on one or more of behaviors.In some embodiments, described Coding, which can be, encodes user behavior track, and the purpose of the coding is for being labeled and normalized. The method of the coding can use different methods according to business difference.For example, user has accessed online friend's link, then The link is represented by the id that concatenated coding is an int is changed；It in some serial numbers, is also processed similarly, such as will transfer accounts The amount of money is divided into 0~9.99,10~49.99,50~199.99,200~999.99,1000 or more five grades, by these etc. Grade is also converted into the id of an int to represent.
Step 706, splice one or more of codings, obtain the track factor.The track factor is based on upper The corresponding one or more coding splicings of one or more operation behaviors are stated to obtain.
In some embodiments, in some embodiments, the behavior of the target user can also include target user's Personal attribute information.The personal attribute information may include essential information, personality, preference etc..Accordingly, in some implementations Can also be encoded to individual subscriber attribute information to obtain the corresponding track factor in example.For example, obtaining user's After consumption preferences, this preference of user is encoded to the id of an int to represent；The academic information of user is also encoded to one The id of a int is represented；The id that the income information (in the case where that can obtain) of user is also encoded to an int is carried out into generation Table.
In some embodiments, can also behavior to user and personal information carry out comprehensive coding, it is corresponding to obtain The track factor.Such as: serial user's behavior of lending will be liked and be encoded to the id of an int to represent；By annual income 15 ~20 ten thousand user's behavior of lending is also converted into the id of an int to represent.
The possible beneficial effect of one or more embodiments of this specification includes but is not limited to: this specification one Or multiple embodiments are determining and during export target criteria information, it, can be larger by the action trail information of user Promotion recognition capability, especially promote fuzzy problem identification rate.Customer service question answering system can not only be promoted and answer customer problem Accuracy rate, the information of original " not can recognize " can also be converted to the information of " recognizable ", greatly improve user The ability of question sentence identification.It should be noted that the different issuable beneficial effects of embodiment are different, in different embodiments In, it is possible to create beneficial effect can be the combinations of any of the above one or more, be also possible to other and any possible obtain Beneficial effect.
Basic conception is described above, it is clear that those skilled in the art, above-mentioned detailed disclosure is only As an example, and not constituting the restriction to one or more embodiments of this specification.Although do not clearly state herein, Those skilled in the art may carry out various modifications, improve and correct to one or more embodiments of this specification.Such Modification, improvement and amendment are proposed in one or more embodiments of this specification, so such modification, improvement, amendment are still Belong to the spirit and scope of one or more embodiment example embodiments of this specification.
Meanwhile one or more embodiments of this specification have used particular words to describe the one of this specification or more The embodiment of a embodiment.As " one embodiment ", " embodiment ", and/or " some embodiments " means with this specification The relevant a certain feature of at least one embodiment of one or more embodiments, structure or feature.Therefore, it should be emphasized that and pay attention to It is " embodiment " or " one embodiment " referred to twice or repeatedly in this specification in different location or " one alternative Embodiment " is not necessarily meant to refer to the same embodiment.In addition, certain features, knot in one or more embodiments of this specification Structure or feature can carry out combination appropriate.
In addition, it will be understood by those skilled in the art that the various aspects of one or more embodiments of this specification can lead to It crosses several types with patentability or situation is illustrated and described, including any new and useful process, machine, production The combination of product or substance, or any new and useful improvement to them.Correspondingly, the one or more of this specification implement Example various aspects can completely by hardware execute, can completely by software (including firmware, resident software, microcode etc.) execute, It can also be executed by combination of hardware.Hardware above or software are referred to alternatively as " data block ", " module ", " engine ", " list Member ", " component " or " system ".In addition, the various aspects of one or more embodiments of this specification may show as being located at one Or the computer product in multiple computer-readable mediums, the product include computer-readable program coding.
Computer storage medium may include the propagation data signal containing computer program code in one, such as in base Take or as carrier wave a part.The transmitting signal may there are many forms of expression, including electromagnetic form, light form etc., or Suitable combining form.Computer storage medium can be any computer-readable Jie in addition to computer readable storage medium Matter, the medium can realize communication, propagation or transmission for using by being connected to an instruction execution system, device or equipment Program.Program coding in computer storage medium can be propagated by any suitable medium, including wireless The combination of electricity, cable, fiber optic cables, RF or similar mediums or any of above medium.
Computer program code needed for one or more embodiment each sections operation of this specification can be with any one Kind or multiple programs language write, including Object-Oriented Programming Language such as Java, Scala, Smalltalk, Eiffel, JADE, Emerald, C++, C#, VB.NET, Python etc., conventional procedural programming language for example C language, VisualBasic, Fortran2003, Perl, COBOL2002, PHP, ABAP, dynamic programming language such as Python, Ruby and Groovy or other Programming language etc..The program coding can run on the user computer completely or calculate as independent software package in user Operation or part are run or completely at remote computer or place operation part on the user computer in remote computer on machine It is run in reason equipment.In the latter cases, remote computer can be connect by any latticed form with subscriber computer, such as Local area network (LAN) or wide area network (WAN), or it is connected to outer computer (such as passing through internet), or in cloud computing environment, Or (SaaS) is serviced using such as software as service.
In addition, except clearly stating in non-claimed, processing element described in one or more embodiments of this specification and The sequence of sequence, the use of digital alphabet or the use of other titles, the one or more for being not intended to limit this specification are real Apply the sequence of a process and method.Although discussing some by various examples in above-mentioned disclosure it is now recognized that useful invention is real Example is applied, but it is to be understood that, such details only plays the purpose of explanation, and appended claims are not limited in the reality disclosed Example is applied, on the contrary, claim is intended to cover the amendment of all one or more embodiment spirit and scopes for meeting this specification And equivalent combinations.Although can also only pass through for example, system component described above can be realized by hardware device The solution of software is achieved, and described system is installed such as in existing processing equipment or mobile device.
Similarly, it is noted that in order to simplify the statement that one or more embodiments of this specification disclose, to help The understanding to one or more inventive embodiments is helped, above in the description of one or more embodiments of this specification, sometimes Meeting will be in various features merger to one embodiment, attached drawing or descriptions thereof.But this disclosure method is not meant to this The feature referred in aspect ratio claim required for one or more embodiment objects of specification is more.In fact, implementing The feature of example will be less than whole features of the single embodiment of above-mentioned disclosure.
The number of description ingredient, number of attributes is used in some embodiments, it should be appreciated that such to be used for embodiment The number of description has used qualifier " about ", " approximation " or " generally " to modify in some instances.Unless in addition saying It is bright, " about ", " approximation " or " generally " show the variation that the number allows to have ± 20%.Correspondingly, in some embodiments In, numerical parameter used in description and claims is approximation, approximation feature according to needed for separate embodiment It can change.In some embodiments, numerical parameter is considered as defined significant digit and using the reservation of general digit Method.Although for confirming Numerical Range and the parameter of its range range for approximation in one or more embodiments of this specification Value, in a particular embodiment, being set in for such numerical value is reported as precisely as possible in feasible region.
Each patent, patent application, patent application publication object for one or more embodiments reference of this specification And other materials, such as article, books, specification, publication, document, entire contents are incorporated to the one of this specification hereby A or multiple embodiments are as reference.It is inconsistent or generate the Shen that conflicts with the content of one or more embodiments of this specification (this explanation please currently or be later additional to the conditional file of this specification claim widest scope except history file In one or more embodiments of book) also except.It should be noted that if one or more embodiments of this specification are attached Belong to material in description, definition, and/or the use of term and this specification one or more embodiments described in have it is different Cause or conflict place, be subject to this specification one or more embodiments description, definition and/or the use of term.
Finally, it will be understood that embodiment described in one or more embodiments of this specification is only to illustrate this The principle of one or more embodiments of specification.Others deformation may also belong to one or more embodiments of this specification Range.Therefore, as an example, not a limit, the alternative configuration of one or more embodiments of this specification can be considered and this theory The introduction of one or more embodiments of bright book is consistent.Correspondingly, this specification embodiment is not limited only to one of this specification Or the embodiment that multiple embodiments are clearly introduced and described.

Claims (21)

1. a kind of method for obtaining target criteria information, which is characterized in that the described method includes:
The problem of obtaining target user；
First group of standard information is determined based on described problem and the first preset algorithm；
Obtain the track factor of the target user；The track factor reflects that the target user is flat in one or more service At least one behavior on platform；
First kind information relevant to standard information is determined based on the track factor and the second preset algorithm；
One or more targets are determined based on first group of standard information and first kind information relevant to standard information Standard information.
2. the method according to claim 1, wherein further including exporting one or more target criteria information.
3. the method according to claim 1, wherein further include:
User is obtained to the feedback of one or more of target criteria information；
First preset algorithm and/or second preset algorithm are updated according to the feedback of user.
4. the method according to claim 1, wherein first preset algorithm and/or the second pre- imputation Method includes machine learning model, and the machine learning model includes Bert model.
5. the method according to claim 1, wherein based on first group of standard information and the first kind with The relevant information of standard information determines that one or more target criteria information include:
First group of standard information is screened based on first kind information relevant to standard information, is determined one or more Target criteria information；First kind information relevant to standard information includes screening auxiliary information, the screening auxiliary information Reflect the screening conditions of target criteria information.
6. the method according to claim 1, wherein first kind information relevant to standard information includes the Two groups of standard information；Second group of standard information includes one or more standard information and/or its corresponding content information.
7. according to the method described in claim 6, it is characterized in that, described be based on first group of standard information and described first Class information relevant to markup information determines that one or more target criteria information include:
Determine the public standard information of first group of standard information Yu second group of standard information；
One or more target criteria information is determined from the public standard information.
8. according to the method described in claim 6, it is characterized in that, first group of standard information further includes believing with its each standard Corresponding first numerical value of manner of breathing；Second group of standard information further includes second value corresponding with its each standard information.
9. according to the method described in claim 8, it is characterized in that, based on first group of standard information and the first kind with The relevant information of markup information determines that one or more target criteria information include:
Determine the public standard information of first group of standard information Yu second group of standard information；
The first numerical value and second value of public standard information are subjected to operation；
Based on operation result, one or more target criteria information is determined from the public standard information.
10. according to the method described in claim 9, it is characterized in that, the operation includes being multiplied or being added.
11. a kind of system for obtaining target criteria information, which is characterized in that the system comprises:
First obtains module, the problem of for obtaining target user；
First determining module determines first group of standard information based on described problem and the first preset algorithm；
Second obtains module, for obtaining the track factor of the target user；The track factor reflects the target user At least one behavior on one or more service platforms；
Second determining module determines first kind letter relevant to standard information based on the track factor and the second preset algorithm Breath；
Target determination module determines one based on first group of standard information and first kind information relevant to standard information A or multiple target criteria information.
12. system according to claim 11, which is characterized in that the system also includes:
Target output module, for exporting one or more target criteria information.
13. system according to claim 11, which is characterized in that the system also includes:
Third obtains module, for obtaining user to the feedback of one or more of target criteria information；
Algorithm optimization module, for updating first preset algorithm and/or second preset algorithm according to the feedback of user.
14. system according to claim 11, which is characterized in that first preset algorithm and/or described second default Algorithm includes machine learning model, and the machine learning model includes Bert model.
15. system according to claim 11, which is characterized in that the target determination module is also used to based on described first Class information relevant to standard information screens first group of standard information, determines one or more target criteria information；Institute Stating first kind information relevant to standard information includes screening auxiliary information, and the screening auxiliary information reflects target criteria information Screening conditions.
16. system according to claim 11, which is characterized in that first kind information relevant to standard information includes Second group of standard information；Second group of standard information includes one or more standard information and/or its corresponding content information.
17. system according to claim 16, which is characterized in that the target determination module is also used to determine first group of mark The public standard information of calibration information and second group of standard information；
One or more target criteria information is determined from the public standard information.
18. system according to claim 16, which is characterized in that first group of standard information further include and its each standard Corresponding first numerical value of information；Second group of standard information further includes second value corresponding with its each standard information.
19. system according to claim 18, which is characterized in that the target determination module is also used to: determining first group The public standard information of standard information and second group of standard information；
The first numerical value and second value of public standard information are subjected to operation；
Based on operation result, one or more target criteria information is determined from the public standard information.
20. system according to claim 19, which is characterized in that the operation includes being multiplied or being added.
21. a kind of for obtaining the device of target criteria information, described device include at least one processor and at least one Memory；
At least one processor is for storing computer instruction；
At least one described processor be used to execute at least partly instruction in the computer instruction with realize claim 1~ Operation described in any one of 10.
CN201910595218.9A 2019-07-03 2019-07-03 Method and system for acquiring target standard information Active CN110222167B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910595218.9A CN110222167B (en) 2019-07-03 2019-07-03 Method and system for acquiring target standard information

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910595218.9A CN110222167B (en) 2019-07-03 2019-07-03 Method and system for acquiring target standard information

Publications (2)

Publication Number Publication Date
CN110222167A true CN110222167A (en) 2019-09-10
CN110222167B CN110222167B (en) 2023-04-07

Family

ID=67815860

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910595218.9A Active CN110222167B (en) 2019-07-03 2019-07-03 Method and system for acquiring target standard information

Country Status (1)

Country Link
CN (1) CN110222167B (en)

Cited By (14)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110704586A (en) * 2019-09-30 2020-01-17 支付宝(杭州)信息技术有限公司 Information processing method and system
CN110955755A (en) * 2019-11-29 2020-04-03 支付宝(杭州)信息技术有限公司 Method and system for determining target standard information
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
CN111709746A (en) * 2020-06-09 2020-09-25 支付宝(杭州)信息技术有限公司 Risk processing method and device and electronic equipment
CN111859094A (en) * 2020-08-10 2020-10-30 广州驰兴通用技术研究有限公司 Information analysis method and system based on cloud computing
CN111881274A (en) * 2020-07-13 2020-11-03 北京捷通华声科技股份有限公司 Method, device and processor for determining answer of question
CN111914553A (en) * 2020-08-11 2020-11-10 民生科技有限责任公司 Financial information negative subject judgment method based on machine learning
CN113094491A (en) * 2021-05-18 2021-07-09 支付宝(杭州)信息技术有限公司 Service obstacle identification method and system
CN113110887A (en) * 2021-03-31 2021-07-13 联想(北京)有限公司 Information processing method and device, electronic equipment and storage medium
CN113139058A (en) * 2021-05-11 2021-07-20 支付宝(杭州)信息技术有限公司 User obstacle identification method and system
WO2022095354A1 (en) * 2020-11-03 2022-05-12 平安科技（深圳）有限公司 Bert-based text classification method and apparatus, computer device, and storage medium
CN114548103A (en) * 2020-11-25 2022-05-27 马上消费金融股份有限公司 Training method of named entity recognition model and recognition method of named entity
US11468239B2 (en) 2020-05-22 2022-10-11 Capital One Services, Llc Joint intent and entity recognition using transformer models
CN111881274B (en) * 2020-07-13 2024-06-04 北京捷通华声科技股份有限公司 Method, device and processor for determining answers to questions

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107329967A (en) * 2017-05-12 2017-11-07 北京邮电大学 Question answering system and method based on deep learning
WO2018213996A1 (en) * 2017-05-22 2018-11-29 Beijing Didi Infinity Technology And Development Co., Ltd. Systems and methods for determining estimated time of arrival
CN108984658A (en) * 2018-06-28 2018-12-11 阿里巴巴集团控股有限公司 A kind of intelligent answer data processing method and device
CN109783632A (en) * 2019-02-15 2019-05-21 腾讯科技（深圳）有限公司 Customer service information-pushing method, device, computer equipment and storage medium

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107329967A (en) * 2017-05-12 2017-11-07 北京邮电大学 Question answering system and method based on deep learning
WO2018213996A1 (en) * 2017-05-22 2018-11-29 Beijing Didi Infinity Technology And Development Co., Ltd. Systems and methods for determining estimated time of arrival
CN108984658A (en) * 2018-06-28 2018-12-11 阿里巴巴集团控股有限公司 A kind of intelligent answer data processing method and device
CN109783632A (en) * 2019-02-15 2019-05-21 腾讯科技（深圳）有限公司 Customer service information-pushing method, device, computer equipment and storage medium

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
李恒超等: "一种用于构建用户画像的二级融合算法框架", 《计算机科学》 *

Cited By (17)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110704586A (en) * 2019-09-30 2020-01-17 支付宝(杭州)信息技术有限公司 Information processing method and system
CN110955755A (en) * 2019-11-29 2020-04-03 支付宝(杭州)信息技术有限公司 Method and system for determining target standard information
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
US11468239B2 (en) 2020-05-22 2022-10-11 Capital One Services, Llc Joint intent and entity recognition using transformer models
CN111709746A (en) * 2020-06-09 2020-09-25 支付宝(杭州)信息技术有限公司 Risk processing method and device and electronic equipment
CN111881274A (en) * 2020-07-13 2020-11-03 北京捷通华声科技股份有限公司 Method, device and processor for determining answer of question
CN111881274B (en) * 2020-07-13 2024-06-04 北京捷通华声科技股份有限公司 Method, device and processor for determining answers to questions
CN111859094A (en) * 2020-08-10 2020-10-30 广州驰兴通用技术研究有限公司 Information analysis method and system based on cloud computing
CN111914553B (en) * 2020-08-11 2023-10-31 民生科技有限责任公司 Financial information negative main body judging method based on machine learning
CN111914553A (en) * 2020-08-11 2020-11-10 民生科技有限责任公司 Financial information negative subject judgment method based on machine learning
WO2022095354A1 (en) * 2020-11-03 2022-05-12 平安科技（深圳）有限公司 Bert-based text classification method and apparatus, computer device, and storage medium
CN114548103A (en) * 2020-11-25 2022-05-27 马上消费金融股份有限公司 Training method of named entity recognition model and recognition method of named entity
CN114548103B (en) * 2020-11-25 2024-03-29 马上消费金融股份有限公司 Named entity recognition model training method and named entity recognition method
CN113110887A (en) * 2021-03-31 2021-07-13 联想(北京)有限公司 Information processing method and device, electronic equipment and storage medium
CN113110887B (en) * 2021-03-31 2023-07-21 联想(北京)有限公司 Information processing method, device, electronic equipment and storage medium
CN113139058A (en) * 2021-05-11 2021-07-20 支付宝(杭州)信息技术有限公司 User obstacle identification method and system
CN113094491A (en) * 2021-05-18 2021-07-09 支付宝(杭州)信息技术有限公司 Service obstacle identification method and system

Also Published As

Publication number Publication date
CN110222167B (en) 2023-04-07

Similar Documents

Publication Publication Date Title
CN110222167A (en) A kind of method and system obtaining target criteria information
US11748555B2 (en) Systems and methods for machine content generation
Burgess The Executive Guide to Artificial Intelligence: How to identify and implement applications for AI in your organization
US11694040B2 (en) Using communicative discourse trees to detect a request for an explanation
US11775494B2 (en) Multi-service business platform system having entity resolution systems and methods
JP7086993B2 (en) Enable rhetorical analysis by using a discourse tree for communication
US20220206993A1 (en) Multi-service business platform system having custom object systems and methods
US11907274B2 (en) Hyper-graph learner for natural language comprehension
CN109002464B (en) Method and system for automatic report analysis and distribution of suggestions using a conversational interface
US20230057760A1 (en) Constructing conclusive answers for autonomous agents
US11829420B2 (en) Summarized logical forms for controlled question answering
CN109792402A (en) Automated to respond to the request of user
US20230252224A1 (en) Systems and methods for machine content generation
US20230410195A1 (en) Dynamically determining real-time offers
CN114547475B (en) Resource recommendation method, device and system
CN110264330A (en) Credit index calculating method, device, computer readable storage medium
CN110362594A (en) A kind of information processing method and system
US20230418793A1 (en) Multi-service business platform system having entity resolution systems and methods
CN117291722A (en) Object management method, related device and computer readable medium
CN115905472A (en) Business opportunity service processing method, business opportunity service processing device, business opportunity service processing server and computer readable storage medium
KR20190108246A (en) Method for providing contract maker service based on artificial intelligence chatbot
CN110929528B (en) Method, device, server and storage medium for analyzing emotion of sentence
CN112836113B (en) Information processing method, apparatus, device, storage medium, and computer program
US20230316186A1 (en) Multi-service business platform system having entity resolution systems and methods
Li Developing an intelligent assistant for the audit plan brainstorming session

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: 20200923

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.

Effective date of registration: 20200923

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.

TA01 Transfer of patent application right
GR01 Patent grant
GR01 Patent grant