WO2019024162A1 - 意图获取方法、电子装置及计算机可读存储介质 - Google Patents

意图获取方法、电子装置及计算机可读存储介质 Download PDF

Info

Publication number
WO2019024162A1
WO2019024162A1 PCT/CN2017/100056 CN2017100056W WO2019024162A1 WO 2019024162 A1 WO2019024162 A1 WO 2019024162A1 CN 2017100056 W CN2017100056 W CN 2017100056W WO 2019024162 A1 WO2019024162 A1 WO 2019024162A1
Authority
WO
WIPO (PCT)
Prior art keywords
intent
node
information
mode
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.)
Ceased
Application number
PCT/CN2017/100056
Other languages
English (en)
French (fr)
Inventor
周俊琨
官民
严江浩
彭小明
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.)
Ping An Technology Shenzhen Co Ltd
Original Assignee
Ping An Technology Shenzhen Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Ping An Technology Shenzhen Co Ltd filed Critical Ping An Technology Shenzhen Co Ltd
Priority to US16/337,929 priority Critical patent/US10423725B2/en
Priority to SG11201902848QA priority patent/SG11201902848QA/en
Publication of WO2019024162A1 publication Critical patent/WO2019024162A1/zh
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F40/00—Handling natural language data
    • G06F40/30—Semantic analysis
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/24—Querying
    • G06F16/245—Query processing
    • G06F16/2455—Query execution
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; 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
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F40/00—Handling natural language data
    • G06F40/20—Natural language analysis
    • G06F40/279—Recognition of textual entities
    • G06F40/284—Lexical analysis, e.g. tokenisation or collocates
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/042—Knowledge-based neural networks; Logical representations of neural networks
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N5/00—Computing arrangements using knowledge-based models
    • G06N5/02—Knowledge representation; Symbolic representation
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N5/00—Computing arrangements using knowledge-based models
    • G06N5/02—Knowledge representation; Symbolic representation
    • G06N5/022—Knowledge engineering; Knowledge acquisition

Definitions

  • the present invention relates to the field of artificial intelligence technologies, and in particular, to an intent acquisition method, an electronic device, and a computer readable storage medium.
  • the main object of the present invention is to provide an intent acquisition method, an electronic device, and a computer readable storage medium, which improve migration efficiency and human-computer interaction response speed, and have little impact on global results.
  • the present invention provides an electronic device including a memory, a processor, and an intent acquisition system stored on the memory and operable on the processor, the intent acquisition system
  • the following steps are implemented when executed by the processor:
  • the intent node after the complement slot is an egress node in the intent node layer, obtaining, according to the intent information corresponding to the intent node after the complement slot, within a specified range in the detailed intent layer of the knowledge map The detailed intent information associated with the intent node after the slotting, and determining the intent of the user according to the obtained detailed intent information;
  • the user's intent is determined according to a preset challenge mode.
  • the present invention also provides an intent acquisition method, which is applied to an electronic device, and the method includes:
  • the intent node after the complement slot is an egress node in the intent node layer, obtaining, according to the intent information corresponding to the intent node after the complement slot, within a specified range in the detailed intent layer of the knowledge map The detailed intent information associated with the intent node after the slotting, and determining the intent of the user according to the obtained detailed intent information;
  • the user's intent is determined according to a preset challenge mode.
  • the present invention also provides a computer readable storage medium storing an intent acquisition system, the intent acquisition system being executable by at least one processor to implement the following steps :
  • the intent node after the complement slot is an egress node in the intent node layer, obtaining, according to the intent information corresponding to the intent node after the complement slot, within a specified range in the detailed intent layer of the knowledge map The detailed intent information associated with the intent node after the slotting, and determining the intent of the user according to the obtained detailed intent information;
  • the user's intent is determined according to a preset challenge mode.
  • the intent acquisition method, the electronic device and the computer readable storage medium provided by the invention only acquire the user's intention information within a specified small range, and the human-computer interaction response speed is faster, even if there is inaccurate knowledge in the knowledge base, It will not affect the overall situation. If the user's intention is not clear, the user's intention is further determined through the intelligent questioning mode, and the fully automatic process of the user's intention to acquire the server is realized. Further, the present invention takes into account the situation of different channel types in semantic complementation and intent questioning. When the system migrates between different channels, it can be carried out without interruption, reducing the learning cost of the migration work, and only need to learn to use the migration tool. Yes, not easy to make mistakes, high migration efficiency and safe and reliable.
  • FIG. 1 is a schematic diagram of an optional hardware architecture of an electronic device of the present invention
  • FIG. 2 is a schematic diagram of functional modules of various embodiments of an intent acquisition system in an electronic device according to the present invention
  • FIG. 3 is a schematic flowchart of an implementation process of an embodiment of an intent acquisition method according to the present invention.
  • FIG. 4 is a diagram showing an example of a knowledge map set in advance in the present invention.
  • FIG. 5 is a diagram showing an example of word semantic combination of an intent node layer in the knowledge map of FIG. 4;
  • FIG. 6 is a schematic diagram of an implementation architecture of the present invention.
  • FIG. 7 is a schematic diagram of a planning architecture of the present invention.
  • Electronic device 2 Intention acquisition system 20 Memory twenty one processor twenty two Network Interface twenty three Semantic understanding module 200 Word segmentation module 201 Intent acquisition module 202 Intent knowledge acquisition module 203 Training module 204 Post processing module 205 Training data module 206 Test data module 207 Module above 208 Knowledge map module 209 Process step S31-S36
  • first, second and the like in the present invention are for the purpose of description only, and are not to be construed as indicating or implying their relative importance or implicitly indicating the number of indicated technical features. .
  • features defining “first” and “second” may include at least one of the features, either explicitly or implicitly.
  • the technical solutions between the various embodiments may be combined with each other, but must be based on the realization of those skilled in the art, and when the combination of the technical solutions is contradictory or impossible to implement, it should be considered that the combination of the technical solutions does not exist. Not within the scope of protection required by the present invention Inside.
  • the present invention proposes an electronic device 2.
  • FIG. 1 there is shown a schematic diagram of an optional hardware architecture of the electronic device of the present invention.
  • the electronic device 2 may include, but is not limited to, a memory 21, a processor 22, and a network interface 23 that are communicably connected to each other through a system bus.
  • the electronic device 2 may be a computing device such as a rack server, a blade server, a tower server, or a rack server.
  • the electronic device 2 may be an independent server or a server cluster composed of multiple servers. . It is pointed out that FIG. 1 only shows the electronic device 2 with the components 21-23, but it should be understood that not all illustrated components are required to be implemented, and more or fewer components may be implemented instead.
  • the memory 21 includes at least one type of readable storage medium including a flash memory, a hard disk, a multimedia card, a card type memory (eg, SD or DX memory, etc.), and a random access memory (RAM). , static random access memory (SRAM), read only memory (ROM), electrically erasable programmable read only memory (EEPROM), programmable read only memory (PROM), magnetic memory, magnetic disk, optical disk, and the like.
  • the memory 21 may be an internal storage unit of the electronic device 2, such as a hard disk or a memory of the electronic device 2.
  • the memory 21 may also be an external storage device of the electronic device 2, such as a plug-in hard disk equipped on the electronic device 2, a smart memory card (SMC), and a secure digital device. (Secure Digital, SD) card, flash card, etc.
  • the memory 21 can also include both the internal storage unit of the electronic device 2 and its external storage device.
  • the memory 21 is generally used to store an operating system installed in the electronic device 2 and various types of application software, such as program code of the intent acquisition system 20. Further, the memory 21 can also be used to temporarily store various types of data that have been output or are to be output.
  • the processor 22 may be a Central Processing Unit (CPU), controller, microcontroller, microprocessor, or other data processing chip in some embodiments.
  • the processor 22 is typically used to control the overall operation of the electronic device 2, such as performing control and processing associated with data interaction or communication with the electronic device 2.
  • the processor 22 is configured to run program code or process data stored in the memory 21, such as running the intent acquisition system 20 and the like.
  • the network interface 23 may comprise a wireless network interface or a wired network interface, which is typically used to establish a communication connection between the electronic device 2 and other electronic devices.
  • the network interface 23 can be used to connect the electronic device 2 to other electronic devices or clients over a communication network.
  • the communication network may be an intranet, an Internet, a Global System of Mobile communication (GSM), a Wideband Code Division Multiple Access (WCDMA), a 4G network, 5G network Wireless or wired networks such as network, Bluetooth, Wi-Fi, etc.
  • the intent acquisition system 20 may be divided into one or more modules, and the one or more modules are stored in the memory 21 and are composed of one or more processors (this embodiment) This is performed by the processor 22) to complete the present invention.
  • the intent acquisition system 20 can be divided into a component word module 201, an intent acquisition module 202, an intent knowledge acquisition module 203, a training module 204, a post-processing module 205, a training data module 206, and a test data module. 207. Module 208 above and knowledge map module 209.
  • the intent acquisition module 202 and the intent knowledge acquisition module 203 may be combined into a semantic understanding module 200.
  • the functional modules referred to in the present invention refer to a series of computer program instruction segments capable of performing a specific function, and are more suitable than the program to describe the execution process of the intent acquisition system 20 in the electronic device 2.
  • the functions of the respective function modules 201-209 will be described in detail below.
  • modules that implement the core functions of the present invention are the word segmentation module 201 and the intent acquisition module 202, and the other modules 203-209 are complementary and further improved to implement the core functions of the present invention.
  • the word segmentation module 201 is configured to obtain a text sentence of a current conversation (ie, a current conversation) of the user, and decompose the text sentence into a plurality of words by using a preset word segmentation algorithm.
  • a current conversation ie, a current conversation
  • the voice information of the user is first converted into a text sentence by a voice recognition algorithm such as a MATLAB algorithm or a DTW algorithm.
  • the intent acquisition module 202 is configured to map the decomposed plurality of words to the preset keywords according to the synonym mode and the part of speech of the words, and obtain the words after the text sentence mapping.
  • the text sentence of the current conversation of the user is: asking for a credit card bill, and the plurality of words broken down into: an inquiry, a credit card, a bill, if the preset keywords include: an inquiry, a credit card, and a bill. Since the part of the word "inquiry” is the same as the part of the keyword “query” (both verbs) and the meaning is similar, the word “inquiry” is mapped to the keyword “query”. Similarly, the word “credit card” is mapped to the keyword “credit card”, and the word “bill” is mapped to the keyword “bill”, that is, the words after the text sentence mapping include: query, credit card, bill.
  • the intent acquisition module 202 is further configured to: determine an intent node that is closest to the semantics of the mapped word in a preset knowledge map.
  • the preset knowledge map includes, but is not limited to, an intent node layer (or a knowledge system layer, a knowledge node layer), a detailed intent layer, and a knowledge base layer.
  • the intent node layer comprises a multi-level node composed of different types of word elements
  • the detailed intent layer storing an egress node associated with the intent node layer (or referred to as a "last level node")
  • the knowledge base layer stores detailed intent information filtered out from the detailed intent layer, and records the filtered detailed intent information as specific knowledge.
  • the different types of word elements include, but are not limited to, a series of elements (such as a credit card), a target element (such as a bill or card), and an action element (such as a query).
  • the intent node layer includes a three-level node: a first-level node (such as a credit card node) composed of a series of elements, and a second-level node composed of target elements. (such as billing nodes or card nodes), a third-level node consisting of action elements (such as query nodes or replenishment nodes).
  • “Yes” and “No” in the parentheses are used to record whether a node is an exit node, "Yes" represents an exit node, and "No” represents a non-export node.
  • the intent node layer includes a three-level node, and not all the egress nodes are third-level nodes, and some egress nodes may be second-level nodes, as shown in FIG. 4 .
  • the card node (second level node) is the exit node.
  • the intent node layer may also be a four-level node or other multi-level nodes set according to different application environments.
  • the intent information output by the query node is “query credit card current bill” and “query credit card history bill”, and the intent information output by the card node (export node) is “credit card”.
  • the detailed intent information in the detailed intent layer includes: “query credit card current bill, query credit card history bill, and credit card card” and the like. Filtering out the detailed detailed intent information from the detailed intent layer, "Query credit card current bill, query credit card history bill” (excluding meaningless detailed intent information "credit card”), and filter out the above meaningful meaning Detailed intent information is stored as specific knowledge to the knowledge base layer.
  • the determining step of the semantically closest intent node comprises: traversing an intent node layer in the knowledge map, and determining one in the intent node layer according to a synonym pattern and a part of speech of a word An intent node that is most similar to the semantics of the mapped words.
  • the intent information of all the egress nodes in the intent node layer can be obtained by differently combining the word semantics of all the nodes of the intent node layer in FIG.
  • the intent information included in the query node (egress node) is: "credit card, bill, query (yes)”
  • the intent information contained in the replenishment node is: “credit card, bill, replenishment (yes)”.
  • “Yes” and “No” in the parentheses are used to record whether a node is an exit node, "Yes” represents an exit node, and "No” represents a non-export node.
  • the intent acquisition module 202 is further configured to perform semantic complementation on the closest intent node according to a preset buffering manner to obtain an intent node after the slot is added.
  • the preset buffering manner includes, but is not limited to, a channel information complementing slot method, a user information complementing slot manner, and a knowledge map information complementing slot manner, and the specific slotting process may adopt one of the supplemental slots.
  • the groove method may be carried out separately or in combination with a plurality of grooves, and is not limited herein.
  • the channel information includes: credit card channel information (on behalf of the user through the contact channel of the credit card center for current conversation), and information on the property insurance channel (on behalf of the user through the contact channel of the property insurance sales center for the current conversation), and Phone channel information (on behalf of the user through the self-service channel to conduct the current conversation).
  • the user information includes: an account type of the user, an operation authority of each type of account, and the like.
  • the knowledge map information is described in FIG. 4, and details are not described herein again.
  • the slotting mode is the channel information complementing mode
  • the missing channel information is supplemented in the closest intent node. For example, if the closest intent node is “query, billing” and the user conducts the current conversation through the credit card contact channel, the “credit card” is added to the closest intent node “inquiry, bill”. The channel information, the intent node "inquiry, credit card, bill” after the slot is added.
  • the complementing mode is a user information filling method, supplementing in the closest intent node Fill in missing user information. For example, if the closest intent node is "query, bill", and the account type in the user information includes only a credit card, then the "credit card” is added to the closest intent node “inquiry, bill”. User information, get the intent node "query, credit card, bill” after the slot.
  • the intent node is supplemented with missing feature information (such as series feature information). For example, if the closest intent node is a "card” and all the element information corresponding to the card node path includes “credit card, card”, the missing element information is added to the closest intent node "card”. "Credit card”, get the intent node "credit card, card” after the slot.
  • the intent acquisition module 202 is further configured to: if the intent node after the complement slot is an egress node in the intent node layer, according to the intent information corresponding to the intent node after the complement slot, in the knowledge map
  • the detailed intent information associated with the intent node after the complement is obtained within a specified range in the detailed intent layer, and the intent of the user is determined according to the obtained detailed intent information, that is, under the intent node after the patching Knowledge is searched according to the corpus of the user's intention to obtain the corresponding knowledge. It should be noted that if there is only one knowledge under the intent node after the padding, the knowledge is directly returned as the intent information of the user.
  • the detailed intent information associated with the query node in the detailed intent layer includes: “Querying the credit card current bill And querying the credit card history bill, determining that the user's intention is to query the credit card current bill or query the credit card history bill, and output the determined user intention to the display unit or the client device of the electronic device, and the user performs final confirmation.
  • the present invention acquires the intent information of the user only within a specified smaller range (within a small range associated with the egress node), the intent information of the user is not found in the entire detailed intent layer, and therefore, the present invention Machine interaction is faster.
  • the search is only performed on a small scale, even if there is inaccurate knowledge in the knowledge base, it will not affect the overall situation.
  • the intent acquisition module 202 is further configured to: if the intent node after the slot is not an egress node (non-egress node) in the intent node layer, determine the intent of the user according to a preset challenge mode.
  • the preset challenge mode includes, but is not limited to, an enumerated mode and a feature mode (or referred to as an "open challenge mode").
  • the enumerated mode determines different enumeration strategies according to different channel types, and outputs corresponding enumeration prompt information according to different enumeration strategies.
  • the feature type mode outputs corresponding element prompt information according to the missing element information in the intent node after the slot is added.
  • the enumeration prompt information or the element prompt information is output to a display unit or a client device of the electronic device.
  • the invention can determine the intention of the user by using a single questioning mode, and can also determine the intention of the user by combining a plurality of questioning modes.
  • the preset tracking mode is an enumerated mode
  • the channel type includes a credit card channel (such as online credit card inquiry), a property insurance channel (such as online property insurance consulting), and Telephone channel (such as bank customer service phone).
  • a credit card channel such as online credit card inquiry
  • a property insurance channel such as online property insurance consulting
  • Telephone channel such as bank customer service phone.
  • the channel type is a credit card channel, outputting a first predetermined number (eg, up to 6) of enumeration prompt information
  • the channel type is a property insurance channel, outputting a second predetermined number (eg, up to 4) of enumerations Prompt message
  • the channel type is phone
  • the channel outputs a third predetermined number (for example, up to 2) of enumeration prompt information.
  • the output enumeration prompt information may be:
  • the output enumeration prompt information may be:
  • the feature mode outputs corresponding element prompt information according to the missing element information in the intent node after the slotting.
  • the output amount element prompt information may be: performing credit card billing What kind of operation?
  • the present invention is further ambiguous when the user's intention is unclear (ie, the intent node after the complement is not the egress node in the intent node layer), it is further determined by a preset challenge mode (enumeration mode or feature mode)
  • a preset challenge mode award mode or feature mode
  • the intention of the user so that the user intends to obtain the process of the server, without the intervention of the customer service personnel, realizes the automatic process of the user's intention to obtain the server.
  • the present invention considers different channel types in further questioning user intent (determining different enumeration strategies according to different channel types, and outputting corresponding enumeration prompt information), and in the most similar intentions
  • nodes perform semantic replenishment, they also consider the different slot types (the credit card channel replenishment method, the production insurance channel replenishment method, and the telephone channel replenishment method). Therefore, when the system migrates between different channels, It can be carried out without interruption, which reduces the learning cost of the migration work. It only needs to learn to use the migration tool, it is not easy to make mistakes, and the migration efficiency is high and safe.
  • the intent acquisition module 202 is further configured to: if the intent node after the complement slot is not an egress node in the intent node layer, output information that is intended to acquire failure to the electronic The display unit or the client device of the device prompts the user to re-convers.
  • the step of semantic padding may also be removed.
  • the intent acquisition module 202 is further configured to: if the determined intent node (ie, the closest intent node) For the egress node in the intent node layer, obtaining detailed intent information associated with the determined intent node within a specified range in the detailed intent layer of the knowledge map according to the intent information corresponding to the determined intent node And determining the intention of the user according to the obtained detailed intent information, that is, searching for the knowledge under the determined intent node according to the corpus of the user's intention to obtain the corresponding knowledge.
  • the intent acquisition module 202 is further configured to: if the determined intent node is not an egress node in the intent node layer, according to a preset The challenge mode determines the intention of the user, or outputs the information intended to acquire the failure to the display unit or the client device of the electronic device, prompting the user to re-synchronize the conversation.
  • the intent acquisition system 20 further includes an intent knowledge acquisition module 203, and the intent knowledge acquisition module 203 is configured to:
  • the determined user intent is output, and the determined user intent is stored to the knowledge base layer of the knowledge map to obtain the user's intention knowledge.
  • the system first calculates an angle cosine between the word vector in the determined user intent and the word vector stored in the knowledge base layer, and obtains the word vector in the determined user intent and the knowledge base layer in advance.
  • the similarity value between the stored word vectors (similarity matching). If the similarity value is greater than a preset threshold (80%), the determined user intent is stored to the knowledge base layer of the knowledge map.
  • the intent acquisition system 20 further includes a training module 204, configured to: perform a word segmentation operation on the training corpus, a word vector model operation, a keyword similar word vector operation, and a first Sub-manual screening, secondary word vector operation, and second manual screening, obtain training data for keyword mapping.
  • a training module 204 configured to: perform a word segmentation operation on the training corpus, a word vector model operation, a keyword similar word vector operation, and a first Sub-manual screening, secondary word vector operation, and second manual screening, obtain training data for keyword mapping.
  • the semantic understanding module 200 further includes:
  • Identifying specific words (such as credit cards, bills, etc.) from the plurality of words that are decomposed according to a preset named entity recognition algorithm (such as a deep neural network based named entity recognition algorithm);
  • the intent acquisition system 20 further includes a post-processing module 205, configured to: the acquired user intent knowledge according to user portrait information (or user attribute information, etc.) Different processing is performed and a semantic result feedback response is performed.
  • the user attribute information includes, but is not limited to, user gender (male and female), user age, user level, and the like. For example, if the attribute information of the user is a VIP user, the credit card billing information of the user for a long time (for example, 100 days) is provided, or when the intent information of the user cannot be obtained, the manual service is automatically switched.
  • the attribute information of the user is a non-VIP user
  • the credit card billing information of the user for a short time for example, 30 days
  • the user's intent information cannot be obtained, the user is prompted to re-synchronize the conversation.
  • the intent acquisition system 20 further includes a training data module 206 for storing a custom dictionary for performing word segmentation training and a stop word list, and storing for semantic recognition training. Online text data and semantic configuration models.
  • the intent acquisition system 20 further includes a test data module 207 for:
  • a word segmentation test data storing a verification word segmentation effect, and semantic test data storing a verification semantic understanding effect
  • Verifying the effect of the word segmentation results using pre-set word segmentation test data and word segmentation analysis algorithms (such as algorithms for calculating accuracy, recall, and F values);
  • the semantic learning effect of the acquired user intent knowledge is verified by using preset semantic test data and semantic analysis algorithms (such as single-step algorithm), and the semantic understanding effect is stored in the knowledge base layer.
  • the intent acquisition system 20 further includes the above module 208, the module 208 is configured to: record the last round of the conversation when acquiring the intention of the user's current conversation (ie, the current round of dialogue) Semantic information (ie, last round semantics) and electronic device status information (such as robot status).
  • Semantic information ie, last round semantics
  • electronic device status information such as robot status
  • the intent acquisition system 20 further includes a knowledge map module 209 for storing the pre-set knowledge map.
  • the intent acquisition system 20 proposed by the present invention acquires the user's intention information only within a specified small range, and the human-computer interaction response speed is faster, even if there is inaccurate knowledge in the knowledge base. It will not affect the overall situation. If the user's intention is not clear, the user's intention is further determined through the intelligent questioning mode, and the fully automatic process of the user's intention to acquire the server is realized. Further, the present invention takes into account the situation of different channel types in semantic complementation and intent questioning. When the system migrates between different channels, it can be carried out without interruption, reducing the learning cost of the migration work, and only need to learn to use the migration tool. Yes, not easy to make mistakes, high migration efficiency and safe and reliable.
  • the present invention also proposes an intent acquisition method.
  • FIG. 3 it is a schematic flowchart of an implementation process of an embodiment of the present invention.
  • the order of execution of the steps in the flowchart shown in FIG. 3 may be changed according to different requirements, and some steps may be omitted.
  • Step S31 Obtain a text sentence of the current conversation (ie, the current conversation) of the user, and decompose the text sentence into a plurality of words by using a preset word segmentation algorithm.
  • the voice information of the user is first converted into a text sentence by a voice recognition algorithm such as a MATLAB algorithm or a DTW algorithm.
  • Step S32 mapping the decomposed plurality of words to the preset keywords according to the synonym mode and the part of speech of the words, and obtaining the words after the text sentence mapping.
  • the text sentence of the current conversation of the user is: asking for a credit card bill, and the plurality of words broken down into: an inquiry, a credit card, a bill, if the preset keywords include: an inquiry, a credit card, and a bill. Since the part of the word "inquiry" is the same as the part of the keyword “query” (both verbs) and the meaning is similar, the word “inquiry” is mapped to the keyword “query”. Similarly, the word “credit card” is mapped to the keyword “credit card”, and the word “bill” is mapped to the keyword “bill”, that is, the words after the text sentence mapping include: query, credit card, bill.
  • Step S33 determining an intent node that is closest to the semantics of the mapped word in a preset knowledge map.
  • the preset knowledge map includes, but is not limited to, an intent node layer (or a knowledge system layer, a knowledge node layer), a detailed intent layer, and a knowledge base layer.
  • the intent node layer comprises a multi-level node composed of different types of word elements
  • the detailed intent layer storing an egress node associated with the intent node layer (or referred to as a "last level node")
  • the knowledge base layer stores detailed intent information filtered out from the detailed intent layer, and records the filtered detailed intent information as specific knowledge.
  • the different types of word elements include, but are not limited to, a series of elements (such as a credit card), a target element (such as a bill or card), and an action element (such as a query).
  • the intent node layer includes a three-level node: a first-level node (such as a credit card node) composed of a series of elements, and a second-level node composed of target elements. (such as billing nodes or card nodes), a third-level node consisting of action elements (such as query nodes or replenishment nodes).
  • “Yes” and “No” in the parentheses are used to record whether a node is an exit node, "Yes" represents an exit node, and "No” represents a non-export node.
  • the intent node layer includes a three-level node, and not all the egress nodes are third-level nodes, and some egress nodes may be second-level nodes, as shown in FIG. 4 .
  • the card node (second level node) is the exit node.
  • the intent node layer may also be a four-level node or other multi-level nodes set according to different application environments.
  • the intent information output by the query node is “query credit card current bill” and “query credit card history bill”, and the intent information output by the card node (export node) is “credit card”.
  • the detailed intent information in the detailed intent layer includes: “query credit card current bill, query credit card history bill, and credit card card” and the like. Filtering out the detailed detailed intent information from the detailed intent layer, "Query credit card current bill, query credit card history bill” (excluding meaningless detailed intent information "credit card”), and filter out the above meaningful meaning Detailed intent information is stored as specific knowledge to the knowledge base layer.
  • the determining step of the semantically closest intent node comprises: traversing an intent node layer in the knowledge map, and determining one in the intent node layer according to a synonym pattern and a part of speech of a word An intent node that is most similar to the semantics of the mapped words.
  • the intent information of all the egress nodes in the intent node layer can be obtained by differently combining the word semantics of all the nodes of the intent node layer in FIG.
  • the intent information included in the query node (egress node) is: "credit card, bill, query (yes)”
  • the intent information contained in the replenishment node is: “credit card, bill, replenishment (yes)”.
  • “Yes” and “No” in the parentheses are used to record whether a node is an exit node, "Yes” represents an exit node, and "No” represents a non-export node.
  • Step S34 Perform semantic encapsulation on the closest intent node according to a preset complementing manner to obtain an intent node after the slot is added.
  • the preset buffering manner includes, but is not limited to, a channel information complementing slot method, a user information complementing slot manner, and a knowledge map information complementing slot manner, and the specific slotting process may adopt one of the supplemental slots.
  • the groove method may be carried out separately or in combination with a plurality of grooves, and is not limited herein.
  • the channel information includes: credit card channel information (on behalf of the user through the contact channel of the credit card center for current conversation), and information on the property insurance channel (on behalf of the user through the contact channel of the property insurance sales center for the current conversation), and Phone channel information (on behalf of the user through the self-service channel to conduct the current conversation).
  • the user information includes: an account type of the user, an operation authority of each type of account, and the like.
  • the knowledge map information is described in FIG. 4, and details are not described herein again.
  • the slotting mode is the channel information complementing mode
  • the missing channel information is supplemented in the closest intent node. For example, if the closest intent node is “query, billing” and the user conducts the current conversation through the credit card contact channel, the “credit card” is added to the closest intent node “inquiry, bill”. The channel information, the intent node "inquiry, credit card, bill” after the slot is added.
  • the complementing mode is a user information filling method, supplementing in the closest intent node Fill in missing user information. For example, if the closest intent node is "query, bill", and the account type in the user information includes only a credit card, then the "credit card” is added to the closest intent node “inquiry, bill”. User information, get the intent node "query, credit card, bill” after the slot.
  • the intent node is supplemented with missing feature information (such as series feature information). For example, if the closest intent node is a "card” and all the element information corresponding to the card node path includes “credit card, card”, the missing element information is added to the closest intent node "card”. "Credit card”, get the intent node "credit card, card” after the slot.
  • Step S35 if the intent node after the complement slot is an egress node in the intent node layer, the specified range in the detailed intent layer of the knowledge map according to the intent information corresponding to the intent node after the complement slot
  • the detailed intent information associated with the intent node after the slot is obtained, and the intention of the user is determined according to the obtained detailed intent information, that is, the knowledge under the intent node after the slot is performed according to the corpus of the user's intention Search for the corresponding knowledge. It should be noted that if there is only one knowledge under the intent node after the padding, the knowledge is directly returned as the intent information of the user.
  • the detailed intent information associated with the query node in the detailed intent layer includes: “Querying the credit card current bill And querying the credit card history bill, determining that the user's intention is to query the credit card current bill or query the credit card history bill, and output the determined user intention to the display unit or the client device of the electronic device, and the user performs final confirmation.
  • the present invention acquires the intent information of the user only within a specified smaller range (within a small range associated with the egress node), the intent information of the user is not found in the entire detailed intent layer, and therefore, the present invention Machine interaction is faster.
  • the search is only performed on a small scale, even if there is inaccurate knowledge in the knowledge base, it will not affect the overall situation.
  • Step S36 If the intent node after the slot is not an egress node (non-egress node) in the intent node layer, determine the intent of the user according to a preset challenge mode.
  • the preset challenge mode includes, but is not limited to, an enumerated mode and a feature mode (or referred to as an "open challenge mode").
  • the enumerated mode determines different enumeration strategies according to different channel types, and outputs corresponding enumeration prompt information according to different enumeration strategies.
  • the feature type mode outputs corresponding element prompt information according to the missing element information in the intent node after the slot is added.
  • the enumeration prompt information or the element prompt information is output to a display unit or a client device of the electronic device.
  • the invention can determine the intention of the user by using a single questioning mode, and can also determine the intention of the user by combining a plurality of questioning modes.
  • the preset tracking mode is an enumerated mode
  • the channel type includes a credit card channel (such as online credit card inquiry), a property insurance channel (such as online property insurance consulting), and Telephone channel (such as bank customer service phone).
  • a credit card channel such as online credit card inquiry
  • a property insurance channel such as online property insurance consulting
  • Telephone channel such as bank customer service phone.
  • the channel type is a credit card channel, outputting a first predetermined number (eg, up to 6) of enumeration prompt information
  • the channel type is a property insurance channel, outputting a second predetermined number (eg, up to 4) of enumerations Prompt message
  • the channel type is phone
  • the channel outputs a third predetermined number (for example, up to 2) of enumeration prompt information.
  • the output enumeration prompt information may be:
  • the output enumeration prompt information may be:
  • the feature mode outputs corresponding element prompt information according to the missing element information in the intent node after the slotting.
  • the output amount element prompt information may be: performing credit card billing What kind of operation?
  • the present invention is further ambiguous when the user's intention is unclear (ie, the intent node after the complement is not the egress node in the intent node layer), it is further determined by a preset challenge mode (enumeration mode or feature mode)
  • a preset challenge mode award mode or feature mode
  • the intention of the user so that the user intends to obtain the process of the server, without the intervention of the customer service personnel, realizes the automatic process of the user's intention to obtain the server.
  • the present invention considers different channel types in further questioning user intent (determining different enumeration strategies according to different channel types, and outputting corresponding enumeration prompt information), and in the most similar intentions
  • nodes perform semantic replenishment, they also consider the different slot types (the credit card channel replenishment method, the production insurance channel replenishment method, and the telephone channel replenishment method). Therefore, when the system migrates between different channels, It can be carried out without interruption, which reduces the learning cost of the migration work. It only needs to learn to use the migration tool, it is not easy to make mistakes, and the migration efficiency is high and safe.
  • the step S36 may be further configured to: if the intent node after the complement slot is not an egress node in the intent node layer, output information that the intent acquisition fails to the electronic device.
  • the display unit or the client device prompts the user to re-convers.
  • the step S34 may also be removed.
  • the step S35 includes: if the intent node determined in the step S33 (ie, the closest intent node) is in the intent node layer. Exporting node, according to the intent information corresponding to the determined intent node, acquiring detailed intent information associated with the determined intent node within a specified range in the detailed intent layer of the knowledge map, and according to the detailed intent obtained The information determines the intention of the user, that is, the knowledge under the determined intent node is searched according to the corpus of the user's intention to obtain the corresponding knowledge.
  • step S36 in this case includes: if the determined intent node is not an egress node in the intent node layer, determining the user's intention according to a preset challenge mode, or The output intention is to obtain the failed information to the display unit or the client device of the electronic device, prompting the user to re-synchronize the conversation.
  • the method further comprises an intent knowledge acquisition step:
  • the determined user intent is output, and the determined user intent is stored to the knowledge base layer of the knowledge map to obtain the user's intention knowledge.
  • the system first calculates an angle cosine between the word vector in the determined user intent and the word vector stored in the knowledge base layer, and obtains the word vector in the determined user intent and the knowledge base layer in advance.
  • the similarity value between the stored word vectors (similarity matching). If the similarity value is greater than a preset threshold (80%), the determined user intent is stored to the knowledge base layer of the knowledge map.
  • the method further includes a training step of sequentially performing a word segmentation operation on the training corpus, a word vector model operation, a keyword similar word vector operation, a first manual screening, a secondary word vector operation, and a Secondary artificial screening to obtain training data for keyword mapping.
  • Identifying specific words (such as credit cards, bills, etc.) from the plurality of words that are decomposed according to a preset named entity recognition algorithm (such as a deep neural network based named entity recognition algorithm);
  • the method further includes a post-processing step of differently processing the acquired user intent knowledge according to user portrait information (or user attribute information, etc.) and performing a semantic result feedback response.
  • the user attribute information includes, but is not limited to, user gender (male and female), user age, user level, and the like. For example, if the attribute information of the user is a VIP user, the credit card billing information of the user for a long time (for example, 100 days) is provided, or when the intent information of the user cannot be obtained, the manual service is automatically switched.
  • the attribute information of the user is a non-VIP user
  • the credit card billing information of the user for a short time for example, 30 days
  • the user's intent information cannot be obtained, the user is prompted to re-synchronize the conversation.
  • the method further comprises the step of training data: storing a custom dictionary for performing word segmentation training and a stop word list, and storing online text data and a semantic configuration model for performing semantic recognition training.
  • the method further comprises the step of testing data:
  • a word segmentation test data storing a verification word segmentation effect, and semantic test data storing a verification semantic understanding effect
  • Verifying the effect of the word segmentation results using pre-set word segmentation test data and word segmentation analysis algorithms (such as algorithms for calculating accuracy, recall, and F values);
  • the semantic learning effect of the acquired user intent knowledge is verified by using preset semantic test data and semantic analysis algorithms (such as single-step algorithm), and the semantic understanding effect is stored in the knowledge base layer.
  • the method further comprises the step of recording: recording the semantic information (ie, the last round semantics) of the last round of dialogue and the state of the electronic device when acquiring the intention of the user's current conversation (ie, the current round of dialogue) Information (such as robot status) and so on.
  • the semantic information ie, the last round semantics
  • the state of the electronic device when acquiring the intention of the user's current conversation (ie, the current round of dialogue) Information (such as robot status) and so on.
  • the intent acquisition method proposed by the present invention only acquires the user's intention information within a specified small range, and the human-computer interaction response speed is faster, even if there is inaccurate knowledge in the knowledge base. Will not affect the overall situation. If the user's intention is not clear, then through the intelligent questioning mode The intention of the user is further determined, and the fully automatic process of the user's intention to acquire the server is realized. Further, the present invention takes into account the situation of different channel types in semantic complementation and intent questioning. When the system migrates between different channels, it can be carried out without interruption, reducing the learning cost of the migration work, and only need to learn to use the migration tool. Yes, not easy to make mistakes, high migration efficiency and safe and reliable.
  • the present invention also provides a computer readable storage medium (such as a ROM/RAM, a magnetic disk, an optical disk) storing the intent acquisition system 20, the intent acquisition system 20 may be executed by at least one processor 22 to cause the at least one processor to perform the steps of the intent acquisition method as described above.
  • a computer readable storage medium such as a ROM/RAM, a magnetic disk, an optical disk
  • the foregoing embodiment method can be implemented by means of software plus a necessary general hardware platform, and can also be implemented by hardware, but in many cases, the former is A better implementation.
  • the technical solution of the present invention which is essential or contributes to the prior art, may be embodied in the form of a software product stored in a storage medium (such as ROM/RAM, disk,
  • the optical disc includes a number of instructions for causing a terminal device (which may be a cell phone, a computer, a server, an air conditioner, or a network device, etc.) to perform the methods described in various embodiments of the present invention.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Computational Linguistics (AREA)
  • General Physics & Mathematics (AREA)
  • Artificial Intelligence (AREA)
  • Mathematical Physics (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Data Mining & Analysis (AREA)
  • Audiology, Speech & Language Pathology (AREA)
  • Software Systems (AREA)
  • Computing Systems (AREA)
  • Evolutionary Computation (AREA)
  • Databases & Information Systems (AREA)
  • Human Computer Interaction (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Molecular Biology (AREA)
  • Machine Translation (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

一种意图获取方法、电子装置及计算机可读存储介质，该方法包括步骤：将文本句子分解成多个词语；将所述分解成的多个词语映射至预设关键词；于一预先设置的知识图谱中确定一个与映射后的词语语义最相近的意图节点；根据预设的补槽方式对所述最相近的意图节点进行语义补槽，得到补槽后的意图节点；若所述补槽后的意图节点为出口节点，则根据该补槽后的意图节点对应的意图信息，在所述知识图谱的详细意图层中的指定范围内获取与该补槽后的意图节点相关联的详细意图信息，并根据获取的详细意图信息确定该用户的意图；若为非出口节点，则根据预设的追问模式确定该用户的意图。本发明提高了迁移效率和人机交互响应速度。

Description

意图获取方法、电子装置及计算机可读存储介质
本申请申明享有2017年8月4日递交的申请号为201710660484.6、名称为“意图获取方法、电子装置及计算机可读存储介质”的中国专利申请的优先权，该中国专利申请的整体内容以参考的方式结合在本申请中。
技术领域
本发明涉及人工智能技术领域，尤其涉及一种意图获取方法、电子装置及计算机可读存储介质。
背景技术
目前，业务同事在维护知识库时，不太专业，没有考虑知识分类的鉴别性，导致鉴别性差的知识添加到知识库后有可能会影响全局。另外，目前的大部分客服机器人对多轮对话交互与对话管理不支持，或者就算支持也需要人工介入和配置。而且在线或者电话等不同渠道需要对多轮对话进行定制化管理，导致系统迁移性较差。
发明内容
本发明的主要目的在于提供一种意图获取方法、电子装置及计算机可读存储介质，提高了迁移效率和人机交互响应速度，对全局结果影响小。
首先，为实现上述目的，本发明提出一种电子装置，所述电子装置包括存储器、处理器及存储在所述存储器上并可在所述处理器上运行的意图获取系统，所述意图获取系统被所述处理器执行时实现如下步骤：
获取用户当前对话的文本句子，通过预设的分词算法将所述文本句子分解成多个词语；
根据近义词模式以及词语的词性，将所述分解成的多个词语映射至预设关键词，获得所述文本句子映射后的词语；
于一预先设置的知识图谱中确定一个与所述映射后的词语语义最相近的意图节点；
根据预设的补槽方式对所述最相近的意图节点进行语义补槽，得到补槽后的意图节点；
若所述补槽后的意图节点为所述意图节点层中的出口节点，则根据该补槽后的意图节点对应的意图信息，在所述知识图谱的详细意图层中的指定范围内获取与该补槽后的意图节点相关联的详细意图信息，并根据获取的详细意图信息确定该用户的意图；及
若所述补槽后的意图节点不是所述意图节点层中的出口节点，则根据预设的追问模式确定该用户的意图。
此外，为实现上述目的，本发明还提供一种意图获取方法，该方法应用于电子装置，所述方法包括：
获取用户当前对话的文本句子，通过预设的分词算法将所述文本句子分解成多个词语；
根据近义词模式以及词语的词性，将所述分解成的多个词语映射至预设关键词，获得所述文本句子映射后的词语；
于一预先设置的知识图谱中确定一个与所述映射后的词语语义最相近的意图节点；
根据预设的补槽方式对所述最相近的意图节点进行语义补槽，得到补槽后的意图节点；
若所述补槽后的意图节点为所述意图节点层中的出口节点，则根据该补槽后的意图节点对应的意图信息，在所述知识图谱的详细意图层中的指定范围内获取与该补槽后的意图节点相关联的详细意图信息，并根据获取的详细意图信息确定该用户的意图；及
若所述补槽后的意图节点不是所述意图节点层中的出口节点，则根据预设的追问模式确定该用户的意图。
进一步地，为实现上述目的，本发明还提供一种计算机可读存储介质，所述计算机可读存储介质存储有意图获取系统，所述意图获取系统可被至少一个处理器执行，以实现如下步骤：
获取用户当前对话的文本句子，通过预设的分词算法将所述文本句子分解成多个词语；
根据近义词模式以及词语的词性，将所述分解成的多个词语映射至预设关键词，获得所述文本句子映射后的词语；
于一预先设置的知识图谱中确定一个与所述映射后的词语语义最相近的意图节点；
根据预设的补槽方式对所述最相近的意图节点进行语义补槽，得到补槽后的意图节点；
若所述补槽后的意图节点为所述意图节点层中的出口节点，则根据该补槽后的意图节点对应的意图信息，在所述知识图谱的详细意图层中的指定范围内获取与该补槽后的意图节点相关联的详细意图信息，并根据获取的详细意图信息确定该用户的意图；及
若所述补槽后的意图节点不是所述意图节点层中的出口节点，则根据预设的追问模式确定该用户的意图。
本发明提出的意图获取方法、电子装置及计算机可读存储介质，只在指定的较小范围内获取用户的意图信息，人机交互响应速度会更快，即使知识库中存在不准确的知识，也不会影响全局。若用户意图不明确，则通过智能追问模式进一步确定该用户的意图，实现了用户意图获取服务端的全自动流程。进一步地，本发明在语义补槽和意图追问时考虑到了不同渠道类型的情形，当系统在不同渠道之间迁移时，可以无间断进行，降低了迁移工作的学习成本，只需要学会使用迁移工具即可，不易出错，迁移效率高且安全可靠。
附图说明
图1为本发明电子装置一可选的硬件架构的示意图；
图2为本发明电子装置中意图获取系统各实施例的功能模块示意图；
图3为本发明意图获取方法一实施例的实施流程示意图；
图4为本发明中所预先设置的知识图谱的示例图；
图5为图4所述知识图谱中的意图节点层的词语语义组合示例图；
图6为本发明的实现架构示意图；
图7为本发明的规划架构示意图。
附图标记：
电子装置 2
意图获取系统 20
存储器 21
处理器 22
网络接口 23
语义理解模块 200
分词模块 201
意图获取模块 202
意图知识获取模块 203
训练模块 204
后处理模块 205
训练数据模块 206
测试数据模块 207
上文模块 208
知识图谱模块 209
流程步骤 S31-S36
本发明目的的实现、功能特点及优点将结合实施例，参照附图做进一步说明。
具体实施方式
为了使本发明的目的、技术方案及优点更加清楚明白，以下结合附图及实施例，对本发明进行进一步详细说明。应当理解，此处所描述的具体实施例仅用以解释本发明，并不用于限定本发明。基于本发明中的实施例，本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例，都属于本发明保护的范围。
需要说明的是，在本发明中涉及“第一”、“第二”等的描述仅用于描述目的，而不能理解为指示或暗示其相对重要性或者隐含指明所指示的技术特征的数量。由此，限定有“第一”、“第二”的特征可以明示或者隐含地包括至少一个该特征。另外，各个实施例之间的技术方案可以相互结合，但是必须是以本领域普通技术人员能够实现为基础，当技术方案的结合出现相互矛盾或无法实现时应当认为这种技术方案的结合不存在，也不在本发明要求的保护范围之 内。
进一步需要说明的是，在本文中，术语“包括”、“包含”或者其任何其他变体意在涵盖非排他性的包含，从而使得包括一系列要素的过程、方法、物品或者装置不仅包括那些要素，而且还包括没有明确列出的其他要素，或者是还包括为这种过程、方法、物品或者装置所固有的要素。在没有更多限制的情况下，由语句“包括一个……”限定的要素，并不排除在包括该要素的过程、方法、物品或者装置中还存在另外的相同要素。
首先，本发明提出一种电子装置2。
参阅图1所示，是本发明电子装置一可选的硬件架构的示意图。
本实施例中，所述电子装置2可包括，但不限于，可通过系统总线相互通信连接存储器21、处理器22、网络接口23。其中，所述电子装置2可以是机架式服务器、刀片式服务器、塔式服务器或机柜式服务器等计算设备，该电子装置2可以是独立的服务器，也可以是多个服务器所组成的服务器集群。需要指出的是，图1仅示出了具有组件21-23的电子装置2，但是应理解的是，并不要求实施所有示出的组件，可以替代的实施更多或者更少的组件。
其中，所述存储器21至少包括一种类型的可读存储介质，所述可读存储介质包括闪存、硬盘、多媒体卡、卡型存储器(例如，SD或DX存储器等)、随机访问存储器(RAM)、静态随机访问存储器(SRAM)、只读存储器(ROM)、电可擦除可编程只读存储器(EEPROM)、可编程只读存储器(PROM)、磁性存储器、磁盘、光盘等。在一些实施例中，所述存储器21可以是所述电子装置2的内部存储单元，例如该电子装置2的硬盘或内存。在另一些实施例中，所述存储器21也可以是所述电子装置2的外部存储设备，例如该电子装置2上配备的插接式硬盘，智能存储卡(Smart Media Card,SMC)，安全数字(Secure Digital,SD)卡，闪存卡(Flash Card)等。当然，所述存储器21还可以既包括所述电子装置2的内部存储单元也包括其外部存储设备。本实施例中，所述存储器21通常用于存储安装于所述电子装置2的操作系统和各类应用软件，例如所述意图获取系统20的程序代码等。此外，所述存储器21还可以用于暂时地存储已经输出或者将要输出的各类数据。
所述处理器22在一些实施例中可以是中央处理器(Central Processing Unit，CPU)、控制器、微控制器、微处理器、或其他数据处理芯片。该处理器22通常用于控制所述电子装置2的总体操作，例如执行与所述电子装置2进行数据交互或者通信相关的控制和处理等。本实施例中，所述处理器22用于运行所述存储器21中存储的程序代码或者处理数据，例如运行所述的意图获取系统20等。
所述网络接口23可包括无线网络接口或有线网络接口，该网络接口23通常用于在所述电子装置2与其他电子设备之间建立通信连接。例如，所述网络接口23可以用于通过通信网络将所述电子装置2与其它电子设备或客户端相连。所述通信网络可以是企业内部网(Intranet)、互联网(Internet)、全球移动通讯系统(Global System of Mobile communication，GSM)、宽带码分多址(Wideband Code Division Multiple Access，WCDMA)、4G网络、5G网 络、蓝牙(Bluetooth)、Wi-Fi等无线或有线网络。
至此，己经详细介绍了本发明相关设备的硬件结构和功能。下面，将基于上述相关设备，提出本发明的各个实施例。
参阅图2所示，是本发明电子装置2中意图获取系统20各实施例的功能模块图。本实施例中，所述的意图获取系统20可以被分割成一个或多个模块，所述一个或者多个模块被存储于所述存储器21中，并由一个或多个处理器(本实施例中为所述处理器22)所执行，以完成本发明。例如，在图2中，所述的意图获取系统20可以被分割成分词模块201、意图获取模块202、意图知识获取模块203、训练模块204、后处理模块205、训练数据模块206、测试数据模块207、上文模块208、以及知识图谱模块209。其中，所述意图获取模块202及意图知识获取模块203可以合并组合成语义理解模块200。本发明所称的功能模块是指能够完成特定功能的一系列计算机程序指令段，比程序更适合于描述所述意图获取系统20在所述电子装置2中的执行过程。以下将就各功能模块201-209的功能进行详细描述。
需要说明的是，实现本发明目的核心功能的模块为分词模块201和意图获取模块202，其它模块203-209是实现本发明核心功能的补充和进一步完善。
所述分词模块201，用于获取用户当前对话(即本轮对话)的文本句子，通过预设的分词算法将所述文本句子分解成多个词语。优选地，如果用户当前对话的内容为语音信息，则先通过语音识别算法(如MATLAB算法或DTW算法)将用户的语音信息转化成文本句子。
所述意图获取模块202，用于根据近义词模式以及词语的词性，将所述分解成的多个词语映射至预设关键词，获得所述文本句子映射后的词语。例如，用户当前对话的文本句子为：询问信用卡账单，则分解成的多个词语包括：询问、信用卡、账单，如果预设关键词包括：查询、信用卡、账单。由于词语“询问”的词性与关键词“查询”的词性相同(皆为动词)且意义相近，则词语“询问”映射至关键词“查询”。同理，词语“信用卡”映射至关键词“信用卡”，词语“账单”映射至关键词“账单”，即所述文本句子映射后的词语包括：查询、信用卡、账单。
所述意图获取模块202，还用于：于一预先设置的知识图谱中确定一个与所述映射后的词语语义最相近的意图节点。
优选地，在本实施例中，所述预先设置的知识图谱包括，但不限于，意图节点层(或称其为知识体系层、知识节点层)、详细意图层、及知识库层。其中，所述意图节点层包括根据不同类型的词语要素组成的多级节点，所述详细意图层存储与所述意图节点层的出口节点(或称其为“最后一级节点”)相关联的详细意图信息，所述知识库层存储从所述详细意图层中筛选出的详细意图信息，将筛选出的详细意图信息记录为具体知识。
优选地，在本实施例中，所述不同类型的词语要素包括，但不限于，系列要素(如信用卡)、目标要素(如账单或卡片)、及动作要素(如查询)。举例而言，参阅图4所示，在本实施例中，所述意图节点层包括三级节点：由系列要素组成的第一级节点(如信用卡节点)、由目标要素组成的第二级节点 (如账单节点或卡片节点)、由动作要素组成的第三级节点(如查询节点或补寄节点)。其中，所述括号内标注的“是”和“否”用于记录某节点是否为出口节点，“是”代表出口节点，“否”代表非出口节点。
需要说明的是，在本实施例中，所述意图节点层包括三级节点并不是指所有的出口节点都是第三级节点，某些出口节点可能是第二级节点，如图4中的卡片节点(第二级节点)即是出口节点。进一步地，在其它实施例中，根据不同的应用环境，所述意图节点层也可以是四级节点或其它根据不同应用环境设置的多级节点。
进一步参阅图4所示，查询节点(出口节点)输出的意图信息为“查询信用卡本期账单”、及“查询信用卡历史账单”，卡片节点(出口节点)输出的意图信息为“信用卡卡片”。相应地，所述详细意图层中的详细意图信息包括：“查询信用卡本期账单、查询信用卡历史账单、及信用卡卡片”等。从所述详细意图层中筛选出有意义的详细意图信息“查询信用卡本期账单、查询信用卡历史账单”(剔除无意义的详细意图信息“信用卡卡片”)，并将筛选出的上述有意义的详细意图信息作为具体知识存储至所述知识库层。
优选地，在本实施例中，所述语义最相近的意图节点的确定步骤包括：遍历所述知识图谱中的意图节点层，根据近义词模式以及词语的词性，于所述意图节点层中确定一个与所述映射后的词语语义最相近的意图节点。举例而言，参阅图5所示，通过对图4中的意图节点层所有节点的词语语义进行不同组合，可以得到所述意图节点层中所有的出口节点的意图信息。例如，查询节点(出口节点)包含的意图信息为：“信用卡、账单、查询(是)”，补寄节点包含的意图信息为：“信用卡、账单、补寄(是)”。其中，所述括号内标注的“是”和“否”用于记录某节点是否为出口节点，“是”代表出口节点，“否”代表非出口节点。
所述意图获取模块202，还用于：根据预设的补槽方式对所述最相近的意图节点进行语义补槽，得到补槽后的意图节点。
在本实施例中，所述预设的补槽方式包括，但不限于，渠道信息补槽方式、用户信息补槽方式、及知识图谱信息补槽方式，具体补槽过程可以采用其中一种补槽方式单独进行，也可以多种补槽方式结合进行，在此不作限定。在本实施例中，所述渠道信息包括：信用卡渠道信息(代表用户通过信用卡中心的联系渠道进行当前对话)、产险渠道信息(代表用户通过产险销售中心的联系渠道进行当前对话)、及电话渠道信息(代表用户通过自助客服电话渠道进行当前对话)。所述用户信息包括：用户的帐号类型、每个类型帐号的操作权限等。所述知识图谱信息参阅图4所述，在此不再赘述。
若所述补槽方式为渠道信息补槽方式，则在所述最相近的意图节点中补充缺少的渠道信息。举例而言，如果所述最相近的意图节点为“查询、账单”，且该用户系通过信用卡联系渠道进行当前对话，则在所述最相近的意图节点“查询、账单”中补充“信用卡”渠道信息，得到补槽后的意图节点“查询、信用卡、账单”。
若所述补槽方式为用户信息补槽方式，则在所述最相近的意图节点中补 充缺少的用户信息。举例而言，如果所述最相近的意图节点为“查询、账单”，且该用户信息中的帐号类型只包括信用卡，则在所述最相近的意图节点“查询、账单”中补充“信用卡”用户信息，得到补槽后的意图节点“查询、信用卡、账单”。
若所述补槽方式为知识图谱信息补槽方式，则在所述知识图谱的意图节点层中确定所述最相近的意图节点所在的节点路径对应的所有要素信息，并在所述最相近的意图节点中补充缺少的要素信息(如系列要素信息)。举例而言，如果最相近的意图节点为“卡片”，且卡片节点路径对应的所有要素信息包括“信用卡、卡片”，则在所述最相近的意图节点“卡片”中补充缺少的要素信息“信用卡”，得到补槽后的意图节点“信用卡、卡片”。
所述意图获取模块202，还用于：若所述补槽后的意图节点为所述意图节点层中的出口节点，则根据该补槽后的意图节点对应的意图信息，在所述知识图谱的详细意图层中的指定范围内获取与该补槽后的意图节点相关联的详细意图信息，并根据获取的详细意图信息确定该用户的意图，即对所述补槽后的意图节点下的知识根据用户产生意图的语料进行搜索得到相应知识。需要说明的是，如果所述补槽后的意图节点下的知识只有一条，则直接返回该知识作为该用户的意图信息。
举例而言，参阅图4所示，若所述补槽后的意图节点为查询节点(出口节点)，所述详细意图层中与查询节点相关联的详细意图信息包括：“查询信用卡本期账单、查询信用卡历史账单”，则确定该用户的意图为查询信用卡本期账单或查询信用卡历史账单，并输出该确定的用户意图至电子装置的显示单元或用户端设备，由该用户进行最终确认。
由于本发明只在指定的较小范围内(与出口节点相关联的小范围内)获取该用户的意图信息，不会在整个详细意图层中寻找该用户的意图信息，因此，本发明的人机交互响应速度会更快。另外，由于搜索只在较小范围内进行，因此，即使知识库中存在不准确的知识，也不会影响全局。
所述意图获取模块202，还用于：若所述补槽后的意图节点不是所述意图节点层中的出口节点(非出口节点)，则根据预设的追问模式确定该用户的意图。在本实施例中，所述预设的追问模式包括，但不限于，枚举型模式和要素型模式(或称之为“开放式追问模式”)。其中，所述枚举型模式根据不同渠道类型确定不同的枚举策略，并根据不同的枚举策略输出相应的枚举提示信息。所述要素型模式根据所述补槽后的意图节点中缺少的要素信息，输出相应的要素提示信息。例如，输出所述枚举提示信息或要素提示信息至电子装置的显示单元或用户端设备。本发明可以采用单一的追问模式确定该用户的意图，也可以采用多种追问模式相结合的方式确定该用户的意图。
优选地，在其它实施例中，若所述预设的追问模式为枚举型模式，且所述渠道类型包括信用卡渠道(如在线信用卡查询)、产险渠道(如在线产险咨询)、及电话渠道(如银行客服电话)。若该渠道类型为信用卡渠道，则输出第一预定数量(如最多6个)的枚举提示信息；若该渠道类型为产险渠道，则输出第二预定数量(如最多4个)的枚举提示信息；若该渠道类型为电话 渠道，则输出第三预定数量(如最多2个)的枚举提示信息。
举例而言，当渠道类型为电话渠道，且所述补槽后的意图节点为账单节点(非出口节点)，则输出的枚举提示信息可以是：
查询信用卡账单？还是
补寄信用卡账单？
当渠道类型为信用卡渠道，且所述补槽后的意图节点为信用卡节点(非出口节点)，则输出的枚举提示信息可以是：
查询信用卡账单？
补寄信用卡账单？还是
信用卡卡片？
优选地，在其它实施例中，若所述预设的追问模式为要素型模式，所述要素型模式根据所述补槽后的意图节点中缺少的要素信息，输出相应的要素提示信息。举例而言，若所述补槽后的意图节点为账单节点，且该账单节点中缺少的要素信息包括动作要素信息(如查询或补寄)，则输出额要素提示信息可以是：进行信用卡账单的何种操作？
由于本发明在用户意图不明确时(即所述补槽后的意图节点不是所述意图节点层中的出口节点时)，通过预设的追问模式(枚举型模式或要素型模式)进一步确定该用户的意图，从而使得用户意图获取在服务端的过程中无需客服人员介入，实现了用户意图获取服务端的全自动流程。
进一步地，由于本发明在进一步追问用户意图时考虑到了不同渠道类型的情形(根据不同渠道类型确定不同的枚举策略，并输出相应的枚举提示信息)，且在对所述最相近的意图节点进行语义补槽时也考虑到了不同渠道类型的补槽方式(信用卡渠道补槽方式、产险渠道补槽方式、电话渠道补槽方式)，因此，当系统在不同渠道之间进行迁移时，可以无间断进行，降低了迁移工作的学习成本，只需要学会使用迁移工具即可，不易出错，迁移效率高且安全可靠。
需要说明的是，在其它实施例中，所述意图获取模块202还用于：若所述补槽后的意图节点不是所述意图节点层中的出口节点，则输出意图获取失败的信息至电子装置的显示单元或用户端设备，提示用户重新进行对话。
进一步地，在其它实施例中，所述语义补槽的步骤也可以去除，此种情形下所述意图获取模块202，还用于：若所述确定的意图节点(即最相近的意图节点)为所述意图节点层中的出口节点，则根据该确定的意图节点对应的意图信息，在所述知识图谱的详细意图层中的指定范围内获取与该确定的意图节点相关联的详细意图信息，并根据获取的详细意图信息确定该用户的意图，即对该确定的意图节点下的知识根据用户产生意图的语料进行搜索得到相应知识。
进一步地，若所述语义补槽的步骤去除，此种情形下所述意图获取模块202还用于：若所述确定的意图节点不是所述意图节点层中的出口节点，则根据预设的追问模式确定该用户的意图，或者输出意图获取失败的信息至电子装置的显示单元或用户端设备，提示用户重新进行对话。
优选地，在其它实施例中，该意图获取系统20还包括意图知识获取模块203，所述意图知识获取模块203用于：
输出该确定的用户意图，将该确定的用户意图存储至所述知识图谱的知识库层，以获取用户的意图知识。具体而言，系统先计算该确定的用户意图中的词向量与知识库层中预先存储的词向量之间的夹角余弦值，得到该确定的用户意图中的词向量与知识库层中预先存储的词向量之间的相似度值(相似度匹配)。若该相似度值大于预设阀值(80％)，则将该确定的用户意图存储至所述知识图谱的知识库层。
优选地，在其它实施例中，该意图获取系统20还包括训练模块204，所述训练模块204用于：对训练语料依次进行分词操作、词向量模型操作、关键词相似词向量操作、第一次人工筛选、二次词向量操作、及第二次人工筛选，得到用于关键词映射的训练数据。
优选地，在其它实施例中，所述语义理解模块200还包括：
根据预设的命名实体识别算法(如基于深层神经网络的命名实体识别算法)，从所述分解成的多个词语中识别出特定词语(如信用卡、账单等)；
根据预设的集外词检测算法(如两阶段集外词查询算法、词格修正算法等)，针对所述识别出的特定词语进行关键词扩展，并将扩展得到的词语添加至所述分解成的多个词语中。
优选地，在其它实施例中，该意图获取系统20还包括后处理模块205，所述后处理模块205用于：根据用户画像信息(或用户属性信息等)，对所述获取的用户意图知识进行不同处理，并进行语义结果反馈应答。在本实施例中，所述用户属性信息包括，但不限于，用户性别(男和女)、用户年龄、用户等级等。例如，如果该用户的属性信息为VIP用户，则提供该用户查询较长时间(如100天)的信用卡账单信息，或者当无法获取该用户的意图信息时，自动切换至人工服务。如果该用户的属性信息为非VIP用户，则提供该用户查询较短时间(如30天)的信用卡账单信息，或者当无法获取该用户的意图信息时，提示用户重新进行对话等。
优选地，在其它实施例中，该意图获取系统20还包括训练数据模块206，所述训练数据模块206用于：存储进行分词训练的自定义词典及停用词表，以及存储进行语义识别训练的线上文本数据及语义配置模型。
优选地，在其它实施例中，该意图获取系统20还包括测试数据模块207，所述测试数据模块207用于：
存储验证分词效果的分词测试数据，以及存储验证语义理解效果的语义测试数据；
利用预设的分词测试数据和分词分析算法(如计算准确率、召回率、F值的算法)，验证所述分词结果的效果；及
利用预设的语义测试数据和语义分析算法(如单步算法)，验证所述获取的用户意图知识的语义理解效果，并将该语义理解效果存储至知识库层。
优选地，在其它实施例中，该意图获取系统20还包括上文模块208，所述上文模块208用于：在获取用户当前对话(即本轮对话)的意图时，记录上轮对话的语义信息(即上轮语义)及电子装置状态信息(如机器人状态)等。
优选地，在其它实施例中，该意图获取系统20还包括知识图谱模块209，所述知识图谱模块209用于存储所述预先设置的知识图谱。
通过上述功能模块201-209，本发明所提出的意图获取系统20，只在指定的较小范围内获取用户的意图信息，人机交互响应速度会更快，即使知识库中存在不准确的知识，也不会影响全局。若用户意图不明确，则通过智能追问模式进一步确定该用户的意图，实现了用户意图获取服务端的全自动流程。进一步地，本发明在语义补槽和意图追问时考虑到了不同渠道类型的情形，当系统在不同渠道之间迁移时，可以无间断进行，降低了迁移工作的学习成本，只需要学会使用迁移工具即可，不易出错，迁移效率高且安全可靠。
此外，本发明还提出一种意图获取方法。
参阅图3所示，是本发明意图获取方法一实施例的实施流程示意图。在本实施例中，根据不同的需求，图3所示的流程图中的步骤的执行顺序可以改变，某些步骤可以省略。
步骤S31，获取用户当前对话(即本轮对话)的文本句子，通过预设的分词算法将所述文本句子分解成多个词语。优选地，如果用户当前对话的内容为语音信息，则先通过语音识别算法(如MATLAB算法或DTW算法)将用户的语音信息转化成文本句子。
步骤S32，根据近义词模式以及词语的词性，将所述分解成的多个词语映射至预设关键词，获得所述文本句子映射后的词语。例如，用户当前对话的文本句子为：询问信用卡账单，则分解成的多个词语包括：询问、信用卡、账单，如果预设关键词包括：查询、信用卡、账单。由于词语“询问”的词性与关键词“查询”的词性相同(皆为动词)且意义相近，则词语“询问”映射至关键词“查询”。同理，词语“信用卡”映射至关键词“信用卡”，词语“账单”映射至关键词“账单”，即所述文本句子映射后的词语包括：查询、信用卡、账单。
步骤S33，于一预先设置的知识图谱中确定一个与所述映射后的词语语义最相近的意图节点。
优选地，在本实施例中，所述预先设置的知识图谱包括，但不限于，意图节点层(或称其为知识体系层、知识节点层)、详细意图层、及知识库层。其中，所述意图节点层包括根据不同类型的词语要素组成的多级节点，所述详细意图层存储与所述意图节点层的出口节点(或称其为“最后一级节点”)相关联的详细意图信息，所述知识库层存储从所述详细意图层中筛选出的详细意图信息，将筛选出的详细意图信息记录为具体知识。
优选地，在本实施例中，所述不同类型的词语要素包括，但不限于，系列要素(如信用卡)、目标要素(如账单或卡片)、及动作要素(如查询)。举例而言，参阅图4所示，在本实施例中，所述意图节点层包括三级节点：由系列要素组成的第一级节点(如信用卡节点)、由目标要素组成的第二级节点 (如账单节点或卡片节点)、由动作要素组成的第三级节点(如查询节点或补寄节点)。其中，所述括号内标注的“是”和“否”用于记录某节点是否为出口节点，“是”代表出口节点，“否”代表非出口节点。
需要说明的是，在本实施例中，所述意图节点层包括三级节点并不是指所有的出口节点都是第三级节点，某些出口节点可能是第二级节点，如图4中的卡片节点(第二级节点)即是出口节点。进一步地，在其它实施例中，根据不同的应用环境，所述意图节点层也可以是四级节点或其它根据不同应用环境设置的多级节点。
进一步参阅图4所示，查询节点(出口节点)输出的意图信息为“查询信用卡本期账单”、及“查询信用卡历史账单”，卡片节点(出口节点)输出的意图信息为“信用卡卡片”。相应地，所述详细意图层中的详细意图信息包括：“查询信用卡本期账单、查询信用卡历史账单、及信用卡卡片”等。从所述详细意图层中筛选出有意义的详细意图信息“查询信用卡本期账单、查询信用卡历史账单”(剔除无意义的详细意图信息“信用卡卡片”)，并将筛选出的上述有意义的详细意图信息作为具体知识存储至所述知识库层。
优选地，在本实施例中，所述语义最相近的意图节点的确定步骤包括：遍历所述知识图谱中的意图节点层，根据近义词模式以及词语的词性，于所述意图节点层中确定一个与所述映射后的词语语义最相近的意图节点。举例而言，参阅图5所示，通过对图4中的意图节点层所有节点的词语语义进行不同组合，可以得到所述意图节点层中所有的出口节点的意图信息。例如，查询节点(出口节点)包含的意图信息为：“信用卡、账单、查询(是)”，补寄节点包含的意图信息为：“信用卡、账单、补寄(是)”。其中，所述括号内标注的“是”和“否”用于记录某节点是否为出口节点，“是”代表出口节点，“否”代表非出口节点。
步骤S34，根据预设的补槽方式对所述最相近的意图节点进行语义补槽，得到补槽后的意图节点。
在本实施例中，所述预设的补槽方式包括，但不限于，渠道信息补槽方式、用户信息补槽方式、及知识图谱信息补槽方式，具体补槽过程可以采用其中一种补槽方式单独进行，也可以多种补槽方式结合进行，在此不作限定。在本实施例中，所述渠道信息包括：信用卡渠道信息(代表用户通过信用卡中心的联系渠道进行当前对话)、产险渠道信息(代表用户通过产险销售中心的联系渠道进行当前对话)、及电话渠道信息(代表用户通过自助客服电话渠道进行当前对话)。所述用户信息包括：用户的帐号类型、每个类型帐号的操作权限等。所述知识图谱信息参阅图4所述，在此不再赘述。
若所述补槽方式为渠道信息补槽方式，则在所述最相近的意图节点中补充缺少的渠道信息。举例而言，如果所述最相近的意图节点为“查询、账单”，且该用户系通过信用卡联系渠道进行当前对话，则在所述最相近的意图节点“查询、账单”中补充“信用卡”渠道信息，得到补槽后的意图节点“查询、信用卡、账单”。
若所述补槽方式为用户信息补槽方式，则在所述最相近的意图节点中补 充缺少的用户信息。举例而言，如果所述最相近的意图节点为“查询、账单”，且该用户信息中的帐号类型只包括信用卡，则在所述最相近的意图节点“查询、账单”中补充“信用卡”用户信息，得到补槽后的意图节点“查询、信用卡、账单”。
若所述补槽方式为知识图谱信息补槽方式，则在所述知识图谱的意图节点层中确定所述最相近的意图节点所在的节点路径对应的所有要素信息，并在所述最相近的意图节点中补充缺少的要素信息(如系列要素信息)。举例而言，如果最相近的意图节点为“卡片”，且卡片节点路径对应的所有要素信息包括“信用卡、卡片”，则在所述最相近的意图节点“卡片”中补充缺少的要素信息“信用卡”，得到补槽后的意图节点“信用卡、卡片”。
步骤S35，若所述补槽后的意图节点为所述意图节点层中的出口节点，则根据该补槽后的意图节点对应的意图信息，在所述知识图谱的详细意图层中的指定范围内获取与该补槽后的意图节点相关联的详细意图信息，并根据获取的详细意图信息确定该用户的意图，即对所述补槽后的意图节点下的知识根据用户产生意图的语料进行搜索得到相应知识。需要说明的是，如果所述补槽后的意图节点下的知识只有一条，则直接返回该知识作为该用户的意图信息。
举例而言，参阅图4所示，若所述补槽后的意图节点为查询节点(出口节点)，所述详细意图层中与查询节点相关联的详细意图信息包括：“查询信用卡本期账单、查询信用卡历史账单”，则确定该用户的意图为查询信用卡本期账单或查询信用卡历史账单，并输出该确定的用户意图至电子装置的显示单元或用户端设备，由该用户进行最终确认。
由于本发明只在指定的较小范围内(与出口节点相关联的小范围内)获取该用户的意图信息，不会在整个详细意图层中寻找该用户的意图信息，因此，本发明的人机交互响应速度会更快。另外，由于搜索只在较小范围内进行，因此，即使知识库中存在不准确的知识，也不会影响全局。
步骤S36，若所述补槽后的意图节点不是所述意图节点层中的出口节点(非出口节点)，则根据预设的追问模式确定该用户的意图。在本实施例中，所述预设的追问模式包括，但不限于，枚举型模式和要素型模式(或称之为“开放式追问模式”)。其中，所述枚举型模式根据不同渠道类型确定不同的枚举策略，并根据不同的枚举策略输出相应的枚举提示信息。所述要素型模式根据所述补槽后的意图节点中缺少的要素信息，输出相应的要素提示信息。例如，输出所述枚举提示信息或要素提示信息至电子装置的显示单元或用户端设备。本发明可以采用单一的追问模式确定该用户的意图，也可以采用多种追问模式相结合的方式确定该用户的意图。
优选地，在其它实施例中，若所述预设的追问模式为枚举型模式，且所述渠道类型包括信用卡渠道(如在线信用卡查询)、产险渠道(如在线产险咨询)、及电话渠道(如银行客服电话)。若该渠道类型为信用卡渠道，则输出第一预定数量(如最多6个)的枚举提示信息；若该渠道类型为产险渠道，则输出第二预定数量(如最多4个)的枚举提示信息；若该渠道类型为电话 渠道，则输出第三预定数量(如最多2个)的枚举提示信息。
举例而言，当渠道类型为电话渠道，且所述补槽后的意图节点为账单节点(非出口节点)，则输出的枚举提示信息可以是：
查询信用卡账单？还是
补寄信用卡账单？
当渠道类型为信用卡渠道，且所述补槽后的意图节点为信用卡节点(非出口节点)，则输出的枚举提示信息可以是：
查询信用卡账单？
补寄信用卡账单？还是
信用卡卡片？
优选地，在其它实施例中，若所述预设的追问模式为要素型模式，所述要素型模式根据所述补槽后的意图节点中缺少的要素信息，输出相应的要素提示信息。举例而言，若所述补槽后的意图节点为账单节点，且该账单节点中缺少的要素信息包括动作要素信息(如查询或补寄)，则输出额要素提示信息可以是：进行信用卡账单的何种操作？
由于本发明在用户意图不明确时(即所述补槽后的意图节点不是所述意图节点层中的出口节点时)，通过预设的追问模式(枚举型模式或要素型模式)进一步确定该用户的意图，从而使得用户意图获取在服务端的过程中无需客服人员介入，实现了用户意图获取服务端的全自动流程。
进一步地，由于本发明在进一步追问用户意图时考虑到了不同渠道类型的情形(根据不同渠道类型确定不同的枚举策略，并输出相应的枚举提示信息)，且在对所述最相近的意图节点进行语义补槽时也考虑到了不同渠道类型的补槽方式(信用卡渠道补槽方式、产险渠道补槽方式、电话渠道补槽方式)，因此，当系统在不同渠道之间进行迁移时，可以无间断进行，降低了迁移工作的学习成本，只需要学会使用迁移工具即可，不易出错，迁移效率高且安全可靠。
需要说明的是，在其它实施例中，所述步骤S36也可以设置为：若所述补槽后的意图节点不是所述意图节点层中的出口节点，则输出意图获取失败的信息至电子装置的显示单元或用户端设备，提示用户重新进行对话。
进一步地，在其它实施例中，所述步骤S34也可以去除，此种情形下步骤S35包括：若所述步骤S33中确定的意图节点(即最相近的意图节点)为所述意图节点层中的出口节点，则根据该确定的意图节点对应的意图信息，在所述知识图谱的详细意图层中的指定范围内获取与该确定的意图节点相关联的详细意图信息，并根据获取的详细意图信息确定该用户的意图，即对该确定的意图节点下的知识根据用户产生意图的语料进行搜索得到相应知识。
进一步地，若所述步骤S34去除，此种情形下步骤S36包括：若所述确定的意图节点不是所述意图节点层中的出口节点，则根据预设的追问模式确定该用户的意图，或者输出意图获取失败的信息至电子装置的显示单元或用户端设备，提示用户重新进行对话。
优选地，在其它实施例中，该方法还包括意图知识获取步骤：
输出该确定的用户意图，将该确定的用户意图存储至所述知识图谱的知识库层，以获取用户的意图知识。具体而言，系统先计算该确定的用户意图中的词向量与知识库层中预先存储的词向量之间的夹角余弦值，得到该确定的用户意图中的词向量与知识库层中预先存储的词向量之间的相似度值(相似度匹配)。若该相似度值大于预设阀值(80％)，则将该确定的用户意图存储至所述知识图谱的知识库层。
优选地，在其它实施例中，该方法还包括训练步骤：对训练语料依次进行分词操作、词向量模型操作、关键词相似词向量操作、第一次人工筛选、二次词向量操作、及第二次人工筛选，得到用于关键词映射的训练数据。
优选地，在其它实施例中，步骤S31之后、步骤S32之前还包括如下步骤：
根据预设的命名实体识别算法(如基于深层神经网络的命名实体识别算法)，从所述分解成的多个词语中识别出特定词语(如信用卡、账单等)；
根据预设的集外词检测算法(如两阶段集外词查询算法、词格修正算法等)，针对所述识别出的特定词语进行关键词扩展，并将扩展得到的词语添加至所述分解成的多个词语中。
优选地，在其它实施例中，该方法还包括后处理步骤：根据用户画像信息(或用户属性信息等)，对所述获取的用户意图知识进行不同处理，并进行语义结果反馈应答。在本实施例中，所述用户属性信息包括，但不限于，用户性别(男和女)、用户年龄、用户等级等。例如，如果该用户的属性信息为VIP用户，则提供该用户查询较长时间(如100天)的信用卡账单信息，或者当无法获取该用户的意图信息时，自动切换至人工服务。如果该用户的属性信息为非VIP用户，则提供该用户查询较短时间(如30天)的信用卡账单信息，或者当无法获取该用户的意图信息时，提示用户重新进行对话等。
优选地，在其它实施例中，该方法还包括训练数据步骤：存储进行分词训练的自定义词典及停用词表，以及存储进行语义识别训练的线上文本数据及语义配置模型。
优选地，在其它实施例中，该方法还包括测试数据步骤：
存储验证分词效果的分词测试数据，以及存储验证语义理解效果的语义测试数据；
利用预设的分词测试数据和分词分析算法(如计算准确率、召回率、F值的算法)，验证所述分词结果的效果；及
利用预设的语义测试数据和语义分析算法(如单步算法)，验证所述获取的用户意图知识的语义理解效果，并将该语义理解效果存储至知识库层。
优选地，在其它实施例中，该方法还包括上文记录步骤：在获取用户当前对话(即本轮对话)的意图时，记录上轮对话的语义信息(即上轮语义)及电子装置状态信息(如机器人状态)等。
通过上述步骤S31-S36，本发明所提出的意图获取方法，只在指定的较小范围内获取用户的意图信息，人机交互响应速度会更快，即使知识库中存在不准确的知识，也不会影响全局。若用户意图不明确，则通过智能追问模式 进一步确定该用户的意图，实现了用户意图获取服务端的全自动流程。进一步地，本发明在语义补槽和意图追问时考虑到了不同渠道类型的情形，当系统在不同渠道之间迁移时，可以无间断进行，降低了迁移工作的学习成本，只需要学会使用迁移工具即可，不易出错，迁移效率高且安全可靠。
进一步地，为实现上述目的，本发明还提供一种计算机可读存储介质(如ROM/RAM、磁碟、光盘)，所述计算机可读存储介质存储有意图获取系统20，所述意图获取系统20可被至少一个处理器22执行，以使所述至少一个处理器执行如上所述的意图获取方法的步骤。
通过以上的实施方式的描述，本领域的技术人员可以清楚地了解到上述实施例方法可借助软件加必需的通用硬件平台的方式来实现，当然也可以通过硬件来实现，但很多情况下前者是更佳的实施方式。基于这样的理解，本发明的技术方案本质上或者说对现有技术做出贡献的部分可以以软件产品的形式体现出来，该计算机软件产品存储在一个存储介质(如ROM/RAM、磁碟、光盘)中，包括若干指令用以使得一台终端设备(可以是手机，计算机，服务器，空调器，或者网络设备等)执行本发明各个实施例所述的方法。
以上参照附图说明了本发明的优选实施例，并非因此局限本发明的权利范围。上述本发明实施例序号仅仅为了描述，不代表实施例的优劣。另外，虽然在流程图中示出了逻辑顺序，但是在某些情况下，可以以不同于此处的顺序执行所示出或描述的步骤。
本领域技术人员不脱离本发明的范围和实质，可以有多种变型方案实现本发明，比如作为一个实施例的特征可用于另一实施例而得到又一实施例。凡是利用本发明说明书及附图内容所作的等效结构或等效流程变换，或直接或间接运用在其他相关的技术领域，均同理包括在本发明的专利保护范围内。

Claims (20)

  1. 一种电子装置，其特征在于，所述电子装置包括存储器、处理器及存储在所述存储器上并可在所述处理器上运行的意图获取系统，所述意图获取系统被所述处理器执行时实现如下步骤：
    获取用户当前对话的文本句子，通过预设的分词算法将所述文本句子分解成多个词语；
    根据近义词模式以及词语的词性，将所述分解成的多个词语映射至预设关键词，获得所述文本句子映射后的词语；
    于一预先设置的知识图谱中确定一个与所述映射后的词语语义最相近的意图节点；
    根据预设的补槽方式对所述最相近的意图节点进行语义补槽，得到补槽后的意图节点；
    若所述补槽后的意图节点为所述意图节点层中的出口节点，则根据该补槽后的意图节点对应的意图信息，在所述知识图谱的详细意图层中的指定范围内获取与该补槽后的意图节点相关联的详细意图信息，并根据获取的详细意图信息确定该用户的意图；及
    若所述补槽后的意图节点不是所述意图节点层中的出口节点，则根据预设的追问模式确定该用户的意图。
  2. 如权利要求1所述的电子装置，其特征在于，所述预先设置的知识图谱包括：意图节点层、详细意图层、及知识库层，所述意图节点层包括根据不同类型的词语要素组成的多级节点，所述详细意图层存储与所述意图节点层的出口节点相关联的详细意图信息，所述知识库层存储从所述详细意图层中筛选出的详细意图信息，将筛选出的详细意图信息记录为具体知识。
  3. 如权利要求2所述的电子装置，其特征在于，所述不同类型的词语要素包括：系列要素、目标要素、及动作要素，所述意图节点层包括三级节点：由系列要素组成的第一级节点、由目标要素组成的第二级节点、及由动作要素组成的第三级节点。
  4. 如权利要求2所述的电子装置，其特征在于，所述预设的补槽方式包括：渠道信息补槽方式、用户信息补槽方式、及知识图谱信息补槽方式，所述渠道信息包括：信用卡渠道信息、产险渠道信息、及电话渠道信息，所述用户信息包括：用户的帐号类型、每个类型帐号的操作权限。
  5. 如权利要求4所述的电子装置，其特征在于，所述根据预设的补槽方式对所述最相近的意图节点进行语义补槽包括：
    若所述补槽方式为渠道信息补槽方式，则在所述最相近的意图节点中补充缺少的渠道信息；
    若所述补槽方式为用户信息补槽方式，则在所述最相近的意图节点中补充缺少的用户信息；及
    若所述补槽方式为知识图谱信息补槽方式，则在所述知识图谱的意图节点层中确定所述最相近的意图节点所在的节点路径对应的所有要素信息，并 在所述最相近的意图节点中补充缺少的要素信息。
  6. 如权利要求4所述的电子装置，其特征在于，所述预设的追问模式包括枚举型模式和要素型模式，所述枚举型模式根据不同渠道类型确定不同的枚举策略，并根据不同的枚举策略输出相应的枚举提示信息，所述要素型模式根据所述补槽后的意图节点中缺少的要素信息，输出相应的要素提示信息。
  7. 如权利要求6所述的电子装置，其特征在于，所述根据预设的追问模式确定该用户的意图包括：
    若所述预设的追问模式为枚举型模式，且所述渠道类型为信用卡渠道，则输出第一预定数量的枚举提示信息；
    若所述预设的追问模式为枚举型模式，且所述渠道类型为产险渠道，则输出第二预定数量的枚举提示信息；及
    若所述预设的追问模式为枚举型模式，且所述渠道类型为电话渠道，则输出第三预定数量的枚举提示信息。
  8. 一种意图获取方法，应用于电子装置，其特征在于，所述方法包括：
    获取用户当前对话的文本句子，通过预设的分词算法将所述文本句子分解成多个词语；
    根据近义词模式以及词语的词性，将所述分解成的多个词语映射至预设关键词，获得所述文本句子映射后的词语；
    于一预先设置的知识图谱中确定一个与所述映射后的词语语义最相近的意图节点；
    根据预设的补槽方式对所述最相近的意图节点进行语义补槽，得到补槽后的意图节点；
    若所述补槽后的意图节点为所述意图节点层中的出口节点，则根据该补槽后的意图节点对应的意图信息，在所述知识图谱的详细意图层中的指定范围内获取与该补槽后的意图节点相关联的详细意图信息，并根据获取的详细意图信息确定该用户的意图；及
    若所述补槽后的意图节点不是所述意图节点层中的出口节点，则根据预设的追问模式确定该用户的意图。
  9. 如权利要求8所述的意图获取方法，其特征在于，所述预先设置的知识图谱包括：意图节点层、详细意图层、及知识库层，所述意图节点层包括根据不同类型的词语要素组成的多级节点，所述详细意图层存储与所述意图节点层的出口节点相关联的详细意图信息，所述知识库层存储从所述详细意图层中筛选出的详细意图信息，将筛选出的详细意图信息记录为具体知识。
  10. 如权利要求9所述的意图获取方法，其特征在于，所述不同类型的词语要素包括：系列要素、目标要素、及动作要素，所述意图节点层包括三级节点：由系列要素组成的第一级节点、由目标要素组成的第二级节点、及由动作要素组成的第三级节点。
  11. 如权利要求9所述的意图获取方法，其特征在于，所述预设的补槽方式包括：渠道信息补槽方式、用户信息补槽方式、及知识图谱信息补槽方 式，所述渠道信息包括：信用卡渠道信息、产险渠道信息、及电话渠道信息，所述用户信息包括：用户的帐号类型、每个类型帐号的操作权限。
  12. 如权利要求11所述的意图获取方法，其特征在于，所述根据预设的补槽方式对所述最相近的意图节点进行语义补槽包括：
    若所述补槽方式为渠道信息补槽方式，则在所述最相近的意图节点中补充缺少的渠道信息；
    若所述补槽方式为用户信息补槽方式，则在所述最相近的意图节点中补充缺少的用户信息；及
    若所述补槽方式为知识图谱信息补槽方式，则在所述知识图谱的意图节点层中确定所述最相近的意图节点所在的节点路径对应的所有要素信息，并在所述最相近的意图节点中补充缺少的要素信息。
  13. 如权利要求11所述的意图获取方法，其特征在于，所述预设的追问模式包括枚举型模式和要素型模式，所述枚举型模式根据不同渠道类型确定不同的枚举策略，并根据不同的枚举策略输出相应的枚举提示信息，所述要素型模式根据所述补槽后的意图节点中缺少的要素信息，输出相应的要素提示信息。
  14. 如权利要求13所述的意图获取方法，其特征在于，所述根据预设的追问模式确定该用户的意图包括：
    若所述预设的追问模式为枚举型模式，且所述渠道类型为信用卡渠道，则输出第一预定数量的枚举提示信息；
    若所述预设的追问模式为枚举型模式，且所述渠道类型为产险渠道，则输出第二预定数量的枚举提示信息；及
    若所述预设的追问模式为枚举型模式，且所述渠道类型为电话渠道，则输出第三预定数量的枚举提示信息。
  15. 一种计算机可读存储介质，其特征在于，所述计算机可读存储介质存储有意图获取系统，所述意图获取系统可被至少一个处理器执行，以实现如下步骤：
    获取用户当前对话的文本句子，通过预设的分词算法将所述文本句子分解成多个词语；
    根据近义词模式以及词语的词性，将所述分解成的多个词语映射至预设关键词，获得所述文本句子映射后的词语；
    于一预先设置的知识图谱中确定一个与所述映射后的词语语义最相近的意图节点；
    根据预设的补槽方式对所述最相近的意图节点进行语义补槽，得到补槽后的意图节点；
    若所述补槽后的意图节点为所述意图节点层中的出口节点，则根据该补槽后的意图节点对应的意图信息，在所述知识图谱的详细意图层中的指定范围内获取与该补槽后的意图节点相关联的详细意图信息，并根据获取的详细意图信息确定该用户的意图；及
    若所述补槽后的意图节点不是所述意图节点层中的出口节点，则根据预 设的追问模式确定该用户的意图。
  16. 如权利要求15所述的计算机可读存储介质，其特征在于，所述预先设置的知识图谱包括：意图节点层、详细意图层、及知识库层，所述意图节点层包括根据不同类型的词语要素组成的多级节点，所述详细意图层存储与所述意图节点层的出口节点相关联的详细意图信息，所述知识库层存储从所述详细意图层中筛选出的详细意图信息，将筛选出的详细意图信息记录为具体知识。
  17. 如权利要求16所述的计算机可读存储介质，其特征在于，所述不同类型的词语要素包括：系列要素、目标要素、及动作要素，所述意图节点层包括三级节点：由系列要素组成的第一级节点、由目标要素组成的第二级节点、及由动作要素组成的第三级节点。
  18. 如权利要求16所述的计算机可读存储介质，其特征在于，所述预设的补槽方式包括：渠道信息补槽方式、用户信息补槽方式、及知识图谱信息补槽方式，所述渠道信息包括：信用卡渠道信息、产险渠道信息、及电话渠道信息，所述用户信息包括：用户的帐号类型、每个类型帐号的操作权限。
  19. 如权利要求18所述的计算机可读存储介质，其特征在于，所述根据预设的补槽方式对所述最相近的意图节点进行语义补槽包括：
    若所述补槽方式为渠道信息补槽方式，则在所述最相近的意图节点中补充缺少的渠道信息；
    若所述补槽方式为用户信息补槽方式，则在所述最相近的意图节点中补充缺少的用户信息；及
    若所述补槽方式为知识图谱信息补槽方式，则在所述知识图谱的意图节点层中确定所述最相近的意图节点所在的节点路径对应的所有要素信息，并在所述最相近的意图节点中补充缺少的要素信息。
  20. 如权利要求18所述的计算机可读存储介质，其特征在于，所述预设的追问模式包括枚举型模式和要素型模式，所述枚举型模式根据不同渠道类型确定不同的枚举策略，并根据不同的枚举策略输出相应的枚举提示信息，所述要素型模式根据所述补槽后的意图节点中缺少的要素信息，输出相应的要素提示信息；
    所述根据预设的追问模式确定该用户的意图包括：
    若所述预设的追问模式为枚举型模式，且所述渠道类型为信用卡渠道，则输出第一预定数量的枚举提示信息；
    若所述预设的追问模式为枚举型模式，且所述渠道类型为产险渠道，则输出第二预定数量的枚举提示信息；及
    若所述预设的追问模式为枚举型模式，且所述渠道类型为电话渠道，则输出第三预定数量的枚举提示信息。
PCT/CN2017/100056 2017-08-04 2017-08-31 意图获取方法、电子装置及计算机可读存储介质 Ceased WO2019024162A1 (zh)

Priority Applications (2)

Application Number Priority Date Filing Date Title
US16/337,929 US10423725B2 (en) 2017-08-04 2017-08-31 Intention acquisition method, electronic device and computer-readable storage medium
SG11201902848QA SG11201902848QA (en) 2017-08-04 2017-08-31 Intention acquisition method, electronic device and computer-readable storage medium

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201710660484.6A CN107688614B (zh) 2017-08-04 2017-08-04 意图获取方法、电子装置及计算机可读存储介质
CN201710660484.6 2017-08-04

Publications (1)

Publication Number Publication Date
WO2019024162A1 true WO2019024162A1 (zh) 2019-02-07

Family

ID=61153164

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2017/100056 Ceased WO2019024162A1 (zh) 2017-08-04 2017-08-31 意图获取方法、电子装置及计算机可读存储介质

Country Status (4)

Country Link
US (1) US10423725B2 (zh)
CN (1) CN107688614B (zh)
SG (1) SG11201902848QA (zh)
WO (1) WO2019024162A1 (zh)

Cited By (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110069631A (zh) * 2019-04-08 2019-07-30 腾讯科技（深圳）有限公司 一种文本处理方法、装置以及相关设备
CN110502608A (zh) * 2019-07-05 2019-11-26 平安科技（深圳）有限公司 基于知识图谱的人机对话方法及人机对话装置
CN111125379A (zh) * 2019-12-26 2020-05-08 科大讯飞股份有限公司 知识库扩充方法、装置、电子设备和存储介质
CN113672738A (zh) * 2020-05-15 2021-11-19 罗伯特·博世有限公司 用于确定知识图谱的至少一部分的设备和方法
CN114398903A (zh) * 2022-01-21 2022-04-26 平安科技（深圳）有限公司 意图识别方法、装置、电子设备及存储介质
CN114817576A (zh) * 2022-06-28 2022-07-29 北京邮电大学 模型训练及专利知识图谱补全方法、装置及存储介质
CN116911389A (zh) * 2023-08-15 2023-10-20 电子科技大学 一种基于知识图谱的辐射源工作模式和意图推理方法
CN119292086A (zh) * 2024-09-20 2025-01-10 北京世纪互联宽带数据中心有限公司 设备控制方法、映射关系建立方法及相关设备
CN119831114A (zh) * 2025-03-17 2025-04-15 浙江浙石油综合能源销售有限公司 一种基于知识图谱的加油站顾客行为预测方法及装置

Families Citing this family (53)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP6893186B2 (ja) * 2018-02-16 2021-06-23 日本電信電話株式会社 フレーズベクトル学習装置、方法、及びプログラム
CN108920497B (zh) * 2018-05-23 2021-10-15 北京奇艺世纪科技有限公司 一种人机交互方法及装置
US10726204B2 (en) * 2018-05-24 2020-07-28 International Business Machines Corporation Training data expansion for natural language classification
CN109033223B (zh) * 2018-06-29 2021-09-07 北京百度网讯科技有限公司 用于跨类型对话的方法、装置、设备以及计算机可读存储介质
JP6988715B2 (ja) * 2018-06-29 2022-01-05 日本電信電話株式会社 回答文選択装置、方法、およびプログラム
CN109271557B (zh) 2018-08-31 2022-03-22 北京字节跳动网络技术有限公司 用于输出信息的方法和装置
CN109388802B (zh) * 2018-10-11 2022-11-25 北京轮子科技有限公司 一种基于深度学习的语义理解方法和装置
CN109597993B (zh) * 2018-11-30 2021-11-05 深圳前海微众银行股份有限公司 语句分析处理方法、装置、设备以及计算机可读存储介质
CN110032724B (zh) * 2018-12-19 2022-11-25 阿里巴巴集团控股有限公司 用于识别用户意图的方法及装置
CN109815326B (zh) * 2019-01-24 2021-09-10 网易（杭州）网络有限公司 对话控制方法及装置
CN109871543B (zh) * 2019-03-12 2023-10-24 广东小天才科技有限公司 一种意图获取方法及系统
CN110086826B (zh) * 2019-05-13 2021-09-17 西安点告网络科技有限公司 信息处理方法
CN110263180B (zh) * 2019-06-13 2021-06-04 北京百度网讯科技有限公司 意图知识图谱生成方法、意图识别方法及装置
CN110211577B (zh) * 2019-07-19 2021-06-04 宁波方太厨具有限公司 终端设备及其语音交互方法
CN110442700A (zh) * 2019-08-12 2019-11-12 珠海格力电器股份有限公司 用于人机交互的人机多轮对话方法及系统、智能设备
CN110516046A (zh) * 2019-08-30 2019-11-29 北京泰迪熊移动科技有限公司 一种负面标记号码识别方法、设备及计算机存储介质
CN110598215A (zh) * 2019-09-11 2019-12-20 海南大学 关系定义一切语义的存在计算模型中的语义识别和推导方法
CN110825880A (zh) * 2019-09-18 2020-02-21 平安科技（深圳）有限公司 案件胜率确定方法、装置、设备及计算机可读存储介质
CN110704703A (zh) * 2019-09-27 2020-01-17 北京百度网讯科技有限公司 人机对话方法及装置
CN110909159B (zh) * 2019-09-29 2022-09-16 珠海格力电器股份有限公司 一种用户意图识别方法、装置、终端及存储介质
CN110909169A (zh) * 2019-09-30 2020-03-24 珠海格力电器股份有限公司 基于知识图谱的控制方法、系统、电子设备和存储介质
CN110941694A (zh) * 2019-10-14 2020-03-31 珠海格力电器股份有限公司 知识图谱的搜索定位方法、系统、电子设备和存储介质
CN110738044B (zh) * 2019-10-17 2023-09-22 杭州涂鸦信息技术有限公司 控制意图识别方法及装置、电子设备和存储介质
CN110909138B (zh) * 2019-10-31 2023-03-07 杭州未名信科科技有限公司 一种用户意图识别方法及系统
CN111046667B (zh) * 2019-11-14 2024-02-06 深圳市优必选科技股份有限公司 一种语句识别方法、语句识别装置及智能设备
CN111078855A (zh) * 2019-12-19 2020-04-28 联想(北京)有限公司 信息处理方法、装置、电子设备及存储介质
CN111125309A (zh) * 2019-12-23 2020-05-08 中电云脑(天津)科技有限公司 自然语言处理方法、装置及计算设备、存储介质
CN111126553B (zh) * 2019-12-25 2024-04-30 平安银行股份有限公司 智能机器人面试方法、设备、存储介质及装置
CN111048088A (zh) * 2019-12-26 2020-04-21 北京蓦然认知科技有限公司 一种多应用程序的语音交互方法和装置
CN111241257B (zh) * 2020-01-03 2023-07-21 联想(北京)有限公司 信息处理方法和电子设备
CN111341311A (zh) * 2020-02-21 2020-06-26 深圳前海微众银行股份有限公司 一种语音对话方法及装置
CN111339745A (zh) * 2020-03-06 2020-06-26 京东方科技集团股份有限公司 一种随访报告生成方法、设备、电子设备和存储介质
CN111400480B (zh) * 2020-04-21 2023-05-12 支付宝(杭州)信息技术有限公司 针对多轮对话的用户意图识别方法和装置
US11386132B2 (en) 2020-05-08 2022-07-12 International Business Machines Corporation System and method for retrieving results and responses with context based exclusion criteria
CN111539741B (zh) * 2020-05-15 2023-02-24 支付宝(杭州)信息技术有限公司 识别操作风险的方法、系统和非暂时性存储介质
CN111625633A (zh) * 2020-05-22 2020-09-04 广东飞企互联科技股份有限公司 基于知识图谱的企业制度问答意图识别方法及装置
CN111611350B (zh) * 2020-05-26 2024-04-09 北京妙医佳健康科技集团有限公司 基于健康知识的应答方法、装置及电子设备
CN111680144A (zh) * 2020-06-03 2020-09-18 湖北亿咖通科技有限公司 多轮对话语音交互的方法及系统、存储介质、电子设备
CN111737423B (zh) * 2020-08-26 2021-01-01 北京声智科技有限公司 领域识别方法、装置、电子设备及存储介质
CN112104734B (zh) * 2020-09-15 2022-09-02 北京百度网讯科技有限公司 用于推送信息的方法、装置、设备以及存储介质
CN112115276B (zh) * 2020-09-18 2024-05-24 平安科技（深圳）有限公司 基于知识图谱的智能客服方法、装置、设备及存储介质
CN112905774A (zh) * 2021-02-22 2021-06-04 武汉市聚联科软件有限公司 一种基于事理图谱的人机对话深度意图理解方法
CN113407667B (zh) * 2021-06-11 2024-03-22 上海适享文化传播有限公司 基于语音固定条件下的查询方法
CN113420124B (zh) * 2021-06-25 2024-03-22 上海适享文化传播有限公司 解决语音检索多条件下的冲突方法
CN113657083B (zh) * 2021-08-30 2023-04-18 海南大学 面向意图计算与推理的dikw资源交互填充系统
CN113807063B (zh) * 2021-09-01 2023-09-05 海南大学 面向dikw内容的意图驱动交互填表方法
CN114548119A (zh) * 2022-01-07 2022-05-27 科大讯飞华南人工智能研究院（广州）有限公司 测试集的生成方法、测试方法、装置、设备及介质
CN114519005A (zh) * 2022-02-16 2022-05-20 平安普惠企业管理有限公司 应答文本的测试方法、装置、设备及存储介质
CN114708047B (zh) * 2022-06-06 2022-09-09 上海荣数信息技术有限公司 一种基于知识图谱的外呼策略运营方法与系统
CN115328956B (zh) * 2022-08-11 2025-07-22 北京百度网讯科技有限公司 基于人工智能的数据查询方法、装置和存储介质
CN116049425B (zh) * 2022-11-04 2023-09-22 合肥中科前沿科技有限公司 一种基于rcs架构的机器人交互方法
CN115906846B (zh) * 2022-11-10 2026-03-24 山西大学 一种基于双图的层次特征融合的文档级命名实体识别方法
CN116016782A (zh) * 2023-01-04 2023-04-25 京东科技信息技术有限公司 一种咨询意图的识别方法、装置、电子设备及存储介质

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105843875A (zh) * 2016-03-18 2016-08-10 北京光年无限科技有限公司 一种面向智能机器人的问答数据处理方法及装置
CN106447346A (zh) * 2016-08-29 2017-02-22 北京中电普华信息技术有限公司 一种智能电力客服系统的构建方法及系统

Family Cites Families (24)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JPH0816620A (ja) * 1994-04-28 1996-01-19 Fujitsu Ltd データ分類装置／方法、データ分類木生成装置／方法、派生語抽出装置／方法、シソーラス構築装置／方法、データ処理システム
US5956740A (en) * 1996-10-23 1999-09-21 Iti, Inc. Document searching system for multilingual documents
US6243670B1 (en) * 1998-09-02 2001-06-05 Nippon Telegraph And Telephone Corporation Method, apparatus, and computer readable medium for performing semantic analysis and generating a semantic structure having linked frames
US8407281B2 (en) * 2000-07-25 2013-03-26 Newton Howard Intention-based automated conflict prediction and notification system
US7983997B2 (en) * 2007-11-02 2011-07-19 Florida Institute For Human And Machine Cognition, Inc. Interactive complex task teaching system that allows for natural language input, recognizes a user's intent, and automatically performs tasks in document object model (DOM) nodes
CN101431573B (zh) * 2007-11-08 2013-02-20 上海撼世网络科技有限公司 通过人机交互技术实现自动客户服务的方法和设备
US9524291B2 (en) * 2010-10-06 2016-12-20 Virtuoz Sa Visual display of semantic information
US9092802B1 (en) * 2011-08-15 2015-07-28 Ramakrishna Akella Statistical machine learning and business process models systems and methods
US8873813B2 (en) * 2012-09-17 2014-10-28 Z Advanced Computing, Inc. Application of Z-webs and Z-factors to analytics, search engine, learning, recognition, natural language, and other utilities
US9916538B2 (en) * 2012-09-15 2018-03-13 Z Advanced Computing, Inc. Method and system for feature detection
US20140236570A1 (en) * 2013-02-18 2014-08-21 Microsoft Corporation Exploiting the semantic web for unsupervised spoken language understanding
US10191999B2 (en) * 2014-04-30 2019-01-29 Microsoft Technology Licensing, Llc Transferring information across language understanding model domains
US20160041986A1 (en) * 2014-08-08 2016-02-11 Cuong Duc Nguyen Smart Search Engine
JP6614152B2 (ja) * 2014-09-05 2019-12-04 日本電気株式会社 テキスト処理システム、テキスト処理方法、及び、コンピュータ・プログラム
JP6433209B2 (ja) * 2014-09-18 2018-12-05 大和ハウス工業株式会社 対話用通信システム及び対話用通信方法
CN104731895B (zh) * 2015-03-18 2018-09-18 北京京东尚科信息技术有限公司 自动应答的方法和装置
CN104866593B (zh) * 2015-05-29 2018-05-22 中国电子科技集团公司第二十八研究所 一种基于知识图谱的数据库搜索方法
US9818405B2 (en) * 2016-03-15 2017-11-14 SAESTEK Ses ve Iletisim Bilgisayar Tekn. San. Ve Tic. A.S. Dialog management system
US10606952B2 (en) * 2016-06-24 2020-03-31 Elemental Cognition Llc Architecture and processes for computer learning and understanding
CN106486120B (zh) * 2016-10-21 2019-11-12 上海智臻智能网络科技股份有限公司 交互式语音应答方法及应答系统
CN106682190B (zh) * 2016-12-29 2020-12-15 北京奇虎科技有限公司 标签知识库的构建方法、装置、应用搜索方法和服务器
CN106709040B (zh) * 2016-12-29 2021-02-19 北京奇虎科技有限公司 一种应用搜索方法和服务器
CN106919655B (zh) * 2017-01-24 2020-05-19 网易（杭州）网络有限公司 一种答案提供方法和装置
US20190012388A1 (en) * 2017-07-07 2019-01-10 Jordi Torras Method and system for a semantic search engine using an underlying knowledge base

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105843875A (zh) * 2016-03-18 2016-08-10 北京光年无限科技有限公司 一种面向智能机器人的问答数据处理方法及装置
CN106447346A (zh) * 2016-08-29 2017-02-22 北京中电普华信息技术有限公司 一种智能电力客服系统的构建方法及系统

Cited By (13)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110069631B (zh) * 2019-04-08 2022-11-29 腾讯科技（深圳）有限公司 一种文本处理方法、装置以及相关设备
CN110069631A (zh) * 2019-04-08 2019-07-30 腾讯科技（深圳）有限公司 一种文本处理方法、装置以及相关设备
CN110502608B (zh) * 2019-07-05 2023-10-10 平安科技（深圳）有限公司 基于知识图谱的人机对话方法及人机对话装置
CN110502608A (zh) * 2019-07-05 2019-11-26 平安科技（深圳）有限公司 基于知识图谱的人机对话方法及人机对话装置
CN111125379A (zh) * 2019-12-26 2020-05-08 科大讯飞股份有限公司 知识库扩充方法、装置、电子设备和存储介质
CN111125379B (zh) * 2019-12-26 2022-12-06 科大讯飞股份有限公司 知识库扩充方法、装置、电子设备和存储介质
CN113672738A (zh) * 2020-05-15 2021-11-19 罗伯特·博世有限公司 用于确定知识图谱的至少一部分的设备和方法
CN114398903A (zh) * 2022-01-21 2022-04-26 平安科技（深圳）有限公司 意图识别方法、装置、电子设备及存储介质
CN114398903B (zh) * 2022-01-21 2023-06-20 平安科技（深圳）有限公司 意图识别方法、装置、电子设备及存储介质
CN114817576A (zh) * 2022-06-28 2022-07-29 北京邮电大学 模型训练及专利知识图谱补全方法、装置及存储介质
CN116911389A (zh) * 2023-08-15 2023-10-20 电子科技大学 一种基于知识图谱的辐射源工作模式和意图推理方法
CN119292086A (zh) * 2024-09-20 2025-01-10 北京世纪互联宽带数据中心有限公司 设备控制方法、映射关系建立方法及相关设备
CN119831114A (zh) * 2025-03-17 2025-04-15 浙江浙石油综合能源销售有限公司 一种基于知识图谱的加油站顾客行为预测方法及装置

Also Published As

Publication number Publication date
CN107688614A (zh) 2018-02-13
CN107688614B (zh) 2018-08-10
US10423725B2 (en) 2019-09-24
US20190228069A1 (en) 2019-07-25
SG11201902848QA (en) 2019-05-30

Similar Documents

Publication Publication Date Title
WO2019024162A1 (zh) 意图获取方法、电子装置及计算机可读存储介质
US11580119B2 (en) System and method for automatic persona generation using small text components
WO2020082569A1 (zh) 文本分类方法、装置、计算机设备和存储介质
CN110765763A (zh) 语音识别文本的纠错方法、装置、计算机设备和存储介质
CN111160017A (zh) 关键词抽取方法、话术评分方法以及话术推荐方法
WO2021003819A1 (zh) 基于知识图谱的人机对话方法及人机对话装置
CN107943998A (zh) 一种基于知识图谱的人机对话控制系统及方法
WO2020114100A1 (zh) 一种信息处理方法、装置和计算机存储介质
CN107015983A (zh) 一种用于智能问答中提供知识信息的方法与设备
WO2019062010A1 (zh) 语义识别方法、电子装置及计算机可读存储介质
US11966704B1 (en) Techniques for verifying a veracity of machine learning outputs
CN113436614A (zh) 语音识别方法、装置、设备、系统及存储介质
US12386794B2 (en) Predictive recommendations for schema mapping
CN106649739B (zh) 多轮交互信息继承识别方法、装置以及交互系统
US20210377035A1 (en) Intelligent hashing of sensitive information
CN114239602B (zh) 会话方法、装置和计算机程序产品
CN117634865A (zh) 工作流创建方法、装置、设备及存储介质
WO2019041528A1 (zh) 新闻情感方向判断方法、电子设备及计算机可读存储介质
CN110232108B (zh) 人机对话方法以及对话系统
CN119441616B (zh) 一种基于多维信息和事件因果梳理的潜在风险隐患识别方法、装置、设备、介质及产品
CN120745639A (zh) 对话交互处理方法、装置、设备、介质和产品
CN114036267B (zh) 对话方法及系统
CN110020181A (zh) 一种推荐信息的处理方法、装置及计算机可读存储介质
CN115455164B (zh) 意愿级别的识别方法、装置、计算机设备和存储介质
CN119067103A (zh) 电力应急预案文本生成方法、装置、计算机设备和存储介质

Legal Events

Date Code Title Description
NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 17919731

Country of ref document: EP

Kind code of ref document: A1

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 10/08/2020)

122 Ep: pct application non-entry in european phase

Ref document number: 17919731

Country of ref document: EP

Kind code of ref document: A1