WO2021017173A1 - 自然语言处理的方法、装置及设备 - Google Patents

自然语言处理的方法、装置及设备 Download PDF

Info

Publication number
WO2021017173A1
WO2021017173A1 PCT/CN2019/110894 CN2019110894W WO2021017173A1 WO 2021017173 A1 WO2021017173 A1 WO 2021017173A1 CN 2019110894 W CN2019110894 W CN 2019110894W WO 2021017173 A1 WO2021017173 A1 WO 2021017173A1
Authority
WO
WIPO (PCT)
Prior art keywords
word slot
recognition result
feedback information
slot recognition
bilstm
Prior art date
Application number
PCT/CN2019/110894
Other languages
English (en)
French (fr)
Inventor
钱庄
Original Assignee
北京小米智能科技有限公司
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 北京小米智能科技有限公司 filed Critical 北京小米智能科技有限公司
Priority to RU2019141376A priority Critical patent/RU2726739C1/ru
Priority to JP2019564011A priority patent/JP7101706B2/ja
Priority to KR1020197035003A priority patent/KR102330061B1/ko
Publication of WO2021017173A1 publication Critical patent/WO2021017173A1/zh

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F40/00Handling natural language data
    • G06F40/30Semantic analysis
    • G06F40/35Discourse or dialogue representation
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F40/00Handling natural language data
    • G06F40/30Semantic analysis
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/30Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
    • G06F16/33Querying
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/30Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
    • G06F16/33Querying
    • G06F16/332Query formulation
    • G06F16/3329Natural language query formulation or dialogue systems
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/30Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
    • G06F16/33Querying
    • G06F16/3331Query processing
    • G06F16/334Query execution
    • G06F16/3344Query execution using natural language analysis
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/90Details of database functions independent of the retrieved data types
    • G06F16/903Querying
    • G06F16/9032Query formulation
    • G06F16/90332Natural language query formulation or dialogue systems
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F40/00Handling natural language data
    • G06F40/20Natural language analysis
    • G06F40/279Recognition of textual entities
    • G06F40/289Phrasal analysis, e.g. finite state techniques or chunking
    • G06F40/295Named entity recognition
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00Machine learning
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/004Artificial life, i.e. computing arrangements simulating life
    • G06N3/006Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/04Architecture, e.g. interconnection topology
    • G06N3/044Recurrent networks, e.g. Hopfield networks
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/04Architecture, e.g. interconnection topology
    • G06N3/047Probabilistic or stochastic networks
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/08Learning methods

Definitions

  • the embodiments of the present disclosure relate to the field of man-machine dialogue technology, and in particular to methods, devices, and equipment for natural language processing.
  • Natural language processing is a science that integrates linguistics, computer science, and mathematics. It studies the theories and methods for realizing effective communication between humans and computers in natural language.
  • sequence annotation model is a commonly used model and is widely used in text processing and other related fields.
  • HMM hidden Markov models
  • CRFs conditional random fields
  • RNNs recurrent neural networks
  • the embodiments of the present disclosure provide a natural language processing method, device and equipment.
  • a natural language processing method which is applied to a dialogue robot in a human-machine dialogue system, and the method includes:
  • the man-machine dialogue system further includes a central control module
  • the method further includes:
  • the dialogue robot outputs the word slot recognition result output by the BiLSTM-CRF model to the central control module;
  • the word slot recognition result set includes the word slot recognition result output by the BiLSTM-CRF model and the word slot recognition result output by other dialogue robots; the target word slot recognition result is used as a pair of the human-machine dialogue system The user's reply result is output.
  • the determining feedback information based on the word slot recognition result and user feedback on the word slot recognition result includes:
  • the feedback information is determined according to the user's feedback operation on the reply result.
  • the determining feedback information according to a user's feedback operation on the reply result includes:
  • the positive feedback rate is determined based on the user's feedback operation on the reply result within a period of time.
  • the performing model enhancement learning according to the feedback information includes:
  • the feedback information is fed back to the CRF layer in the BiLSTM-CRF model, so that the CRF layer performs model enhancement training according to the feedback information.
  • a natural language processing device which is applied to a dialog robot in a human-machine dialog system, and the device includes:
  • the word slot recognition result determination module is configured to determine the word slot recognition result output after the bi-directional long short-term memory network algorithm and the conditional random field algorithm BiLSTM-CRF model used to recognize the word slot of the dialogue data input by the user;
  • a feedback information determining module configured to determine feedback information based on the word slot recognition result and the user's feedback on the word slot recognition result
  • the model enhancement learning module is configured to perform enhancement learning on the BiLSTM-CRF model according to the feedback information.
  • the man-machine dialogue system further includes a central control module
  • the device also includes:
  • a word slot recognition result output module configured to output the word slot recognition result output by the BiLSTM-CRF model to the central control module
  • the target word slot recognition result determination module is configured to obtain the target word slot recognition result determined by the central control module from the received word slot recognition result set for the dialogue data;
  • the word slot recognition result set includes the word slot recognition result output by the BiLSTM-CRF model and the word slot recognition result output by other dialogue robots; the target word slot recognition result is used as a pair of the human-machine dialogue system The user's reply result is output.
  • the feedback information determining module includes:
  • the first feedback information determining sub-module is configured to determine that the inconsistent feedback information is negative feedback information in response to the inconsistency between the target word slot recognition result and the word slot recognition result output by the BiLSTM-CRF model;
  • the second feedback information determining sub-module is configured to respond to the target word slot recognition result being consistent with the word slot recognition result output by the BiLSTM-CRF model, and determine the feedback information according to the user's feedback operation on the reply result.
  • the second feedback information determining submodule is specifically configured as:
  • the positive feedback rate is determined based on a user's feedback operation on the reply result within a period of time.
  • model enhancement learning module is specifically configured as:
  • the feedback information is fed back to the CRF layer in the BiLSTM-CRF model, so that the CRF layer performs model enhancement training according to the feedback information.
  • a man-machine dialogue device includes a dialogue robot, and the man-machine dialogue device includes:
  • a memory configured to store executable instructions of the processor
  • the processor is configured to:
  • a computer-readable storage medium on which a computer program is stored, and when the program is executed by a processor, the steps of the foregoing method are implemented.
  • the BiLSTM-CRF model is used as the basic framework. After the BiLSTM-CRF model outputs the word slot recognition result to the outside, the dialogue robot can obtain corresponding feedback information according to the word slot recognition result, and make a pair according to the feedback information.
  • the BiLSTM-CRF model performs enhanced learning to realize dynamic self-learning of the model to reduce the manual labeling process and improve the efficiency and accuracy of word slot recognition.
  • Fig. 1 is a step flowchart of an embodiment of a natural language processing method according to an exemplary embodiment of the present disclosure
  • Fig. 2 is a flow chart showing the steps of another embodiment of a natural language processing method according to an exemplary embodiment of the present disclosure
  • Fig. 3 is a schematic diagram showing a BiLSTM-CRF model according to an exemplary embodiment of the present disclosure
  • Fig. 4 is a block diagram showing an embodiment of a natural language processing apparatus according to an exemplary embodiment of the present disclosure
  • Fig. 5 is a block diagram showing a human-computer interaction device according to an exemplary embodiment of the present disclosure.
  • first, second, third, etc. may be used to describe various information in the embodiments of the present disclosure, the information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other.
  • first information may also be referred to as second information, and similarly, the second information may also be referred to as first information.
  • word “if” as used herein can be interpreted as "when” or "when” or "in response to determination”.
  • FIG. 1 is a step flow chart of an embodiment of a natural language processing method according to an exemplary embodiment of the present disclosure.
  • the method of the embodiment of the present disclosure can be applied to a dialogue robot in a human-machine dialogue system, which may specifically include the following step:
  • Step 101 Determine the word slot recognition result output after the used BiLSTM-CRF model performs word slot recognition on the dialogue data input by the user.
  • the dialogue robot when the dialogue robot receives the dialogue data input by the user, it can use BiLSTM (Bi-directional Long Short-Term Memory, two-way long and short-term memory network algorithm)-CRF (Conditional Random Field algorithm, conditional random field algorithm) )
  • the model performs word slot recognition on the dialogue data, and obtains the word slot recognition result output by the BiLSTM-CRF model.
  • word slot recognition can be understood as a sequence labeling task to solve the sequence labeling problem.
  • Step 102 Determine feedback information based on the word slot recognition result and the user's feedback on the word slot recognition result.
  • the dialogue robot after the dialogue robot obtains the word slot recognition result, it can further determine the feedback information of the word slot recognition result.
  • the feedback information may include the user's feedback on the word slot recognition result.
  • Step 103 Perform reinforcement learning on the BiLSTM-CRF model according to the feedback information.
  • the dialogue robot can perform enhanced learning on the BiLSTM-CRF model based on the obtained feedback information of the word slot recognition result, thereby realizing dynamic self-learning of the model, reducing the manual labeling process, and improving the efficiency and efficiency of word slot recognition Accuracy.
  • FIG. 2 is a step flow chart of another natural language processing method embodiment according to an exemplary embodiment of the present disclosure.
  • the method of the embodiment of the present disclosure may be applied to a dialogue robot in a human-machine dialogue system, which may specifically include The following steps:
  • Step 201 Determine the word slot recognition result output after the used BiLSTM-CRF model performs word slot recognition on the dialogue data input by the user.
  • NLU Natural Language Understanding
  • the NLU module is the core part of the human-machine dialogue system.
  • the functions of the entire NLU module mainly include the following two: The understanding of user intent and the analysis of the core slot (Slot) in the sentence expressed by the user.
  • Intent is a classifier that determines the type of sentence expressed by the user, and then the program corresponding to the determined type (ie, Bot (voice robot)) performs special analysis. For example, the user says: "Put me a happy song.” At this time, it can be judged that the user's intention classification is music, so a music bot (Bot) can be called to recommend a song to the user, and the user feels that it is not right. At that time, say: "change another song", or this music robot will continue to serve the user until the user expresses other problems and the intention is no longer music, and then switch to another robot to serve the user.
  • Bot Voice robot
  • the man-machine dialogue system also includes a central control module that communicates with the NLU module.
  • the central control module can send the user’s dialogue sentence to a statement that can handle the user’s intent.
  • Bot ClickBot, dialogue robot
  • these bots return the results they have processed.
  • the Bot needs to understand the content of the dialogue sentence. For the sake of simplicity, you can select only the core and important parts to understand, and ignore other non-core content, and those core and important parts are called Slots, namely Word slot.
  • the dialog robot Bot in the embodiment of the present disclosure can use the BiLSTM-CRF model to recognize the word slot of the dialog data input by the user.
  • word slot recognition can be understood as a sequence labeling task to solve the sequence labeling problem.
  • x i represents the id of the i-th word of the sentence in the dictionary, and then the one-hot encoding (one-hot) vector of each word can be obtained, and the dimension is the dictionary size.
  • the first layer of the model is the look-up layer, which uses a pre-trained or randomly initialized embedding matrix to map each word in the sentence from a one-hot vector to a low-dimensional dense word vector.
  • set dropout random deactivation, which is a method of optimizing neural networks with deep structure.
  • part of the weight or output of the hidden layer is randomly reset to zero to reduce the interdependence between nodes.
  • the second layer of the model is the bidirectional LSTM (Long Short-Term Memory, long short-term memory network) layer, which includes a forward LSTM and a backward long and short-term memory network (backward LSTM).
  • LSTM Long Short-Term Memory, long short-term memory network
  • backward LSTM backward long and short-term memory network
  • the third layer of the model is the CRF layer, which performs sentence-level sequence labeling.
  • the parameter of the CRF layer is a matrix A of (k+2)*(k+2), and A ij represents from the i-th label to the j-th
  • the transfer score of each label can then be used when labeling a position.
  • the label that has been labeled before can be used.
  • the reason for adding 2 is to add a starting state for the beginning of the sentence and a termination state for the end of the sentence.
  • B-PER and I-PER respectively indicate the first character of a person’s name and the non-first character of a person’s name
  • B-LOC and I-LOC indicate the first character of a place name and the non-first word of a place name respectively
  • B-ORG, I-ORG Respectively indicate the first word of the organization name and the non-first word of the organization name
  • O means that the word is not part of the named entity.
  • various entity types in the field will be customized accordingly, such as movie type (video), weather type (weather) and other types.
  • Step 202 The dialog robot outputs the word slot recognition result output by the BiLSTM-CRF model to the central control module.
  • Step 203 Obtain the target word slot recognition result determined by the central control module from the received word slot recognition result set for the dialogue data.
  • the dialogue robot after the dialogue robot obtains the word slot recognition result output by the BiLSTM-CRF model, it can output the word slot recognition result to the central control module.
  • the central control module For the central control module, it can receive word slot recognition results for the same dialogue data from different dialogue robots Bot to form a word slot recognition result set, then the word slot recognition result set can include the output of the BiLSTM-CRF model Word slot recognition results and word slot recognition results output by other dialogue robots.
  • the central control module can make a decision, decide one intention from a variety of intentions as the user intention, and determine at least one corresponding to the user’s intention
  • a Bot performs analysis and processing separately to obtain a set of processed word slot recognition results.
  • the central control module can decide the target word slot recognition result that best matches the user's intention from the word slot recognition result set, and the target word slot recognition result can be used as the response result output of the human-machine dialogue system to the user.
  • Step 204 In response to the target word slot recognition result being inconsistent with the word slot recognition result output by the BiLSTM-CRF model, determine the inconsistent feedback information as negative feedback information.
  • the target word slot result and the comparison result of the word slot recognition result output by the BiLSTM-CRF model can be combined to determine the feedback information of the word slot recognition result output by the BiLSTM-CRF model.
  • the feedback information may include positive feedback information and negative feedback information.
  • positive feedback information can be represented by a value of 1
  • negative feedback information can be represented by a value of -1.
  • the feedback information can be determined as negative feedback information.
  • Step 205 In response to the target word slot recognition result being consistent with the word slot recognition result output by the BiLSTM-CRF model, determine feedback information according to the user's feedback operation on the reply result.
  • the feedback information can be determined by combining the user's feedback operation on the reply result.
  • the feedback information can be determined by combining the user's feedback operation on the reply result, because the user is the real judge of whether the reply result is reasonable. .
  • step 205 may include the following sub-steps:
  • sub-step S11 in response to the user's positive feedback rate being greater than or equal to a preset threshold, the feedback information is determined as positive feedback information.
  • the feedback information in response to the positive feedback rate being less than a preset threshold, the feedback information is determined to be negative feedback information.
  • the positive feedback rate is determined based on the user's feedback operation on the reply result within a period of time.
  • multiple users may send the same or similar dialogue data to the human-machine dialogue system.
  • the word slot recognition result output by the BiLSTM-CRF model can Count the feedback operations of multiple users on the response results (such as click or like operations) during this time period. If the positive feedback rate is greater than or equal to the preset threshold, it means that the user's feedback is positive feedback.
  • the feedback information can also be stored in the buffer area for subsequent use.
  • Step 206 Feed the feedback information to the CRF layer in the BiLSTM-CRF model, so that the CRF layer performs model enhancement training according to the feedback information.
  • the recorded feedback information can be fed back to the CRF layer of the BiLSTM-CRF model, and the CRF layer will identify the results of each word slot and the corresponding
  • the feedback information is used as training data for the model's enhanced learning training, and the closed loop of the entire learning is completed through the enhanced learning, so that the trained BiLSTM-CRF model can obtain more accurate word slot recognition results.
  • the four elements of BiLSTM-CRF model for reinforcement learning can include:
  • the central control module and the user's feedback on the word slot recognition result can be combined to determine the feedback information, so as to avoid the trouble of user tagging.
  • the information is returned to the CRF layer of the BiLSTM-CRF model for enhanced learning, which can improve the accuracy of the BiLSTM-CRF model.
  • the dialogue data input by the user is "Tomorrow's Weather". Since “Tomorrow's Weather” is a movie name, after the BiLSTM-CRF model recognizes the word slot of "Tomorrow's Weather", the output word slot recognition result is: Tomorrow's weather/video. After the central control module receives the word slot recognition results sent by multiple bots, the final decision of the user's intention of the dialogue data is the weather, so the final target word slot recognition result is "tomorrow/dateweather/weather”.
  • the Reward Calculator After the Reward Calculator obtains the target word slot recognition result of the decision of the central control module, it matches it with the word slot recognition result output by the BiLSTM-CRF model, and judges that the two are inconsistent (one is the result of video, the other is the result of weather), so ,
  • the final target word slot recognition result is "tomorrow's weather/video".
  • the Reward Calculator obtains the target word slot recognition result of the decision made by the central control module, it matches it with the word slot recognition result output by the BiLSTM-CRF model, and judges that the two are consistent (both are video results).
  • the Reward Calculator will count The click-through rate of the user's response result corresponding to the target word slot recognition result (for example, the central control module replies to the user the movie resource of tomorrow’s weather) within a period of time.
  • the click-through rate is relatively low (below the preset threshold)
  • the reply result does not satisfy the user
  • the click rate is relatively high (higher than the preset threshold)
  • the reply result can be considered to satisfy
  • the labeling result is the labeling result of weather
  • the output recognition result is "tomorrow/dateweather/weather", thus achieving the purpose of learning.
  • the embodiment of the present disclosure also provides an embodiment of a natural language processing apparatus.
  • FIG. 4 is a block diagram of an embodiment of a natural language processing apparatus according to an exemplary embodiment of the present disclosure.
  • the apparatus of the embodiment of the present disclosure is applied to a dialogue robot in a human-machine dialogue system. Specifically, it can include the following modules:
  • the word slot recognition result determination module 401 is configured to determine the word slot recognition result output after the used BiLSTM-CRF model performs word slot recognition on the dialogue data input by the user;
  • the feedback information determining module 402 is configured to determine feedback information based on the word slot recognition result and user feedback on the word slot recognition result;
  • the model enhancement learning module 403 is configured to perform enhancement learning on the BiLSTM-CRF model according to the feedback information.
  • the embodiment of the present disclosure obtains the slot recognition result output by the BiLSTM-CRF model through the word slot recognition result determination module 401, and uses the feedback information determination module 402 based on the word slot recognition result and the user's recognition result of the word slot The feedback to determine the feedback information to reduce the amount of manual annotation. Then, the model enhancement learning module 403 performs enhancement learning on the BiLSTM-CRF model according to the feedback information to realize dynamic self-learning of the model, thereby improving the accuracy of the word slot recognition of the model.
  • the human-machine dialogue system further includes a central control module; the device further includes the following modules:
  • a word slot recognition result output module configured to output the word slot recognition result output by the BiLSTM-CRF model to the central control module
  • the target word slot recognition result determination module is configured to obtain the target word slot recognition result determined by the central control module from the received word slot recognition result set for the dialogue data;
  • the word slot recognition result set includes the word slot recognition result output by the BiLSTM-CRF model and the word slot recognition result output by other dialogue robots; the target word slot recognition result is used as a pair of the human-machine dialogue system The user's reply result is output.
  • the feedback information determining module 402 may include the following submodules:
  • the first feedback information determining sub-module is configured to determine that the inconsistent feedback information is negative feedback information in response to the inconsistency between the target word slot recognition result and the word slot recognition result output by the BiLSTM-CRF model;
  • the second feedback information determining sub-module is configured to respond to the target word slot recognition result being consistent with the word slot recognition result output by the BiLSTM-CRF model, and determine the feedback information according to the user's feedback operation on the reply result.
  • the feedback information may include the feedback information of the central processing module and the feedback information of the user, which enriches the certain dimension of the feedback information and improves the accuracy of labeling.
  • the second feedback information determining submodule is specifically configured as:
  • the positive feedback rate is determined based on a user's feedback operation on the reply result within a period of time.
  • the central control module and the user's feedback information can be combined for analysis, so that the accuracy of the marking can be improved.
  • model enhancement learning module 403 is specifically configured as:
  • the feedback information is fed back to the CRF layer in the BiLSTM-CRF model, so that the CRF layer performs model enhancement training according to the feedback information.
  • model enhancement learning module 403 performs model enhancement learning according to the feedback information, which can realize the dynamic self-learning of the model and improve the accuracy of the word slot recognition of the model.
  • the relevant part can refer to the part of the description of the system embodiment.
  • the device embodiments described above are merely illustrative.
  • the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in One place, or it can be distributed to multiple network units.
  • Some or all of the modules may be selected according to actual needs to achieve the objectives of the solutions of the embodiments of the present disclosure. Those of ordinary skill in the art can understand and implement it without creative work.
  • FIG. 5 is a block diagram of a human-machine dialogue device 500 according to an exemplary embodiment of the present disclosure.
  • the device 500 may include one or more of the following components: a processing component 502, a memory 504, a power supply component 506, a multimedia component 508, an audio component 510, an input/output (I/O) interface 512, a sensor component 514, and Communication component 516.
  • the processing component 502 generally controls the overall operation of the device 500, and the processing component 502 may include one or more processors 520 to execute instructions to complete all or part of the steps of the aforementioned method.
  • the processing component 502 may include one or more modules to facilitate the interaction between the processing component 502 and other components.
  • the processing component 502 may include a multimedia module to facilitate the interaction between the multimedia component 508 and the processing component 502.
  • the memory 504 is configured to store various types of data to support operations in the device 500. Examples of such data include instructions for any application or method operating on the device 500.
  • the memory 504 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable Programmable Read Only Memory (EPROM), Programmable Read Only Memory (PROM), Read Only Memory (ROM), magnetic memory, flash memory, magnetic or optical disk.
  • SRAM static random access memory
  • EEPROM electrically erasable programmable read-only memory
  • EPROM erasable Programmable Read Only Memory
  • PROM Programmable Read Only Memory
  • ROM Read Only Memory
  • magnetic memory flash memory
  • flash memory magnetic or optical disk.
  • the power supply component 506 provides power to various components of the device 500.
  • the power supply component 506 may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power for the device 500.
  • the multimedia component 508 includes a screen that provides an output interface between the device 500 and the user.
  • the screen may include a liquid crystal display (LCD) and a touch panel (TP). If the screen includes a touch panel, the screen may be implemented as a touch screen to receive input signals from the user.
  • the touch panel includes one or more touch sensors to sense touch, sliding, and gestures on the touch panel. The touch sensor can not only sense the boundary of the touch or slide action, but also detect the duration and pressure related to the touch or slide operation.
  • the audio component 510 is configured to output and/or input audio signals.
  • the audio component 510 includes a microphone (MIC), and when the device 500 is in an operation mode, such as a call mode, a recording mode, and a voice recognition mode, the microphone is configured to receive external audio signals.
  • the received audio signal may be further stored in the memory 504 or transmitted via the communication component 516.
  • the audio component 510 further includes a speaker for outputting audio signals.
  • the I/O interface 512 provides an interface between the processing component 502 and a peripheral interface module.
  • the peripheral interface module may be a keyboard, a click wheel, a button, and the like. These buttons may include but are not limited to: home button, volume button, start button, and lock button.
  • the sensor component 514 includes one or more sensors for providing the device 500 with various aspects of status assessment.
  • the sensor component 514 can detect the open/close state of the device 500 and the relative positioning of components.
  • the component is the display and the keypad of the device 500.
  • the sensor component 514 can also detect: the device 500 or a component of the device 500 The location changes, the presence or absence of contact between the user and the device 500, the orientation or acceleration/deceleration of the device 500, and the temperature change of the device 500.
  • the sensor assembly 514 may include a proximity sensor configured to detect the presence of nearby objects when there is no physical contact.
  • the sensor component 514 may also include a light sensor, such as a CMOS or CCD image sensor, configured for use in imaging applications.
  • the sensor component 514 may also include an acceleration sensor, a gyroscope sensor, a magnetic sensor, a pressure sensor, or a temperature sensor.
  • the communication component 516 is configured to facilitate wired or wireless communication between the device 500 and other devices.
  • the device 500 can access a wireless network based on a communication standard, such as WiFi, 2G, or 5G, or a combination thereof.
  • the communication component 516 receives a broadcast signal or broadcast related information from an external broadcast management system via a broadcast channel.
  • the communication component 516 further includes a near field communication (NFC) module to facilitate short-range communication.
  • the NFC module can be implemented based on radio frequency identification (RFID) technology, infrared data association (IrDA) technology, ultra-wideband (UWB) technology, Bluetooth (BT) technology and other technologies.
  • RFID radio frequency identification
  • IrDA infrared data association
  • UWB ultra-wideband
  • Bluetooth Bluetooth
  • the device 500 may be implemented by one or more application specific integrated circuits (ASIC), digital signal processors (DSP), digital signal processing devices (DSPD), programmable logic devices (PLD), field programmable A gate array (FPGA), controller, microcontroller, microprocessor, or other electronic components are implemented to implement the above methods.
  • ASIC application specific integrated circuits
  • DSP digital signal processors
  • DSPD digital signal processing devices
  • PLD programmable logic devices
  • FPGA field programmable A gate array
  • controller microcontroller, microprocessor, or other electronic components are implemented to implement the above methods.
  • non-transitory computer-readable storage medium including instructions, such as the memory 504 including instructions, which may be executed by the processor 520 of the device 500 to complete the foregoing method.
  • the non-transitory computer-readable storage medium may be ROM, CD-ROM, magnetic tape, floppy disk, optical data storage device, etc.
  • the device 500 can execute a natural language processing method, including: determining the used BiLSTM-CRF model to perform word slot recognition on the dialogue data input by the user Then output the word slot recognition result; determine the feedback information based on the word slot recognition result and the user's feedback on the word slot recognition result; perform enhancement learning on the BiLSTM-CRF model according to the feedback information.
  • the BiLSTM-CRF model is used as the basic framework. After the BiLSTM-CRF model outputs the word slot recognition result to the outside, the dialogue robot can obtain corresponding feedback information according to the word slot recognition result, and compare the BiLSTM with the feedback information. -The CRF model performs enhanced learning to realize dynamic self-learning of the model to reduce the process of manual labeling and improve the efficiency and accuracy of word slot recognition.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Artificial Intelligence (AREA)
  • Computational Linguistics (AREA)
  • Mathematical Physics (AREA)
  • Data Mining & Analysis (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Databases & Information Systems (AREA)
  • Software Systems (AREA)
  • Evolutionary Computation (AREA)
  • Computing Systems (AREA)
  • Biophysics (AREA)
  • Molecular Biology (AREA)
  • Biomedical Technology (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Audiology, Speech & Language Pathology (AREA)
  • Human Computer Interaction (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Medical Informatics (AREA)
  • Probability & Statistics with Applications (AREA)
  • Machine Translation (AREA)

Abstract

本公开实施例是一种自然语言处理的方法、装置及设备,应用于人机对话系统中的对话机器人,其中所述方法包括:确定使用的双向长短期记忆网络算法及条件随机场算法BiLSTM-CRF模型对用户输入的对话数据进行词槽识别后输出的词槽识别结果;基于所述词槽识别结果以及用户对所述词槽识别结果的反馈确定反馈信息;根据所述反馈信息对所述BiLSTM-CRF模型进行增强学习。

Description

自然语言处理的方法、装置及设备
相关申请的交叉引用
本申请基于申请号为201910687763.0、申请日为2019年07月29日的中国专利申请提出,并要求该中国专利申请的优先权,该中国专利申请的全部内容在此引入本申请作为参考。
技术领域
本公开实施例涉及人机对话技术领域,尤其涉及自然语言处理的方法、装置及设备。
背景技术
自然语言处理是一门融合语言学、计算机科学、数学的科学,研究实现人与计算机之间用自然语言进行有效通信的理论和方法。在自然语言处理中,序列标注模型是常用的模型,被广泛应用于文本处理等相关领域。
解决序列标注问题目前流行的方法包括隐马尔科夫模型(HMM)、条件随机场(CRFs)以及循环神经网络(RNNs)。但上述模型都存在模型能力有限、无法进行自学习等问题。
发明内容
为克服相关技术中存在的问题,本公开实施例提供了一种自然语言处理的方法、装置及设备。
根据本公开实施例的第一方面,提供一种自然语言处理的方法,所述方法应用于人机对话系统中的对话机器人,所述方法包括:
确定使用的双向长短期记忆网络算法及条件随机场算法BiLSTM-CRF模型对用户输入的对话数据进行词槽识别后输出的词槽识别结果;
基于所述词槽识别结果以及用户对所述词槽识别结果的反馈确定反馈信息;
根据所述反馈信息对所述BiLSTM-CRF模型进行增强学习。
可选地,所述人机对话系统还包括中央控制模块;
在确定使用的所述BiLSTM-CRF模型对用户输入的对话数据进行词槽识别后输出的词槽识别结果之后,所述方法还包括:
所述对话机器人将所述BiLSTM-CRF模型输出的词槽识别结果输出至所述中央控制模块;
获取所述中央控制模块从接收到的针对所述对话数据的词槽识别结果集合中决策出的目标词槽识别结果;
其中,所述词槽识别结果集合包括所述BiLSTM-CRF模型输出的词槽识别结果以及其他对话机器人输出的词槽识别结果;所述目标词槽识别结果用于作为所述人机对话系统对用户的回复结果输出。
可选地,所述基于所述词槽识别结果以及用户对所述词槽识别结果的反馈确定反馈信息包括:
响应于所述目标词槽识别结果与所述BiLSTM-CRF模型输出的词槽识别结果不一致,则将所述不一致的反馈信息确定为负反馈信息;
响应于所述目标词槽识别结果与所述BiLSTM-CRF模型输出的词槽识别结果一致,则根据用户对所述回复结果的反馈操作确定反馈信息。
可选地,所述根据用户对所述回复结果的反馈操作确定反馈信息,包括:
响应于用户的正向反馈率大于或等于预设阈值,将该反馈信息确定为正反馈信息;
响应于所述正向反馈率小于预设阈值,将该反馈信息确定为负反馈信息;
其中,所述正向反馈率为根据一段时间内用户对所述回复结果的反馈 操作确定的。
可选地,所述根据所述反馈信息进行模型增强学习,包括:
将所述反馈信息反馈到所述BiLSTM-CRF模型中的CRF层,以由所述CRF层根据所述反馈信息进行模型增强训练。
根据本公开实施例的第二方面,提供一种自然语言处理的装置,所述装置应用于人机对话系统中的对话机器人,所述装置包括:
词槽识别结果确定模块,被配置为确定使用的双向长短期记忆网络算法及条件随机场算法BiLSTM-CRF模型对用户输入的对话数据进行词槽识别后输出的词槽识别结果;
反馈信息确定模块,被配置为基于所述词槽识别结果以及用户对所述词槽识别结果的反馈确定反馈信息;
模型增强学习模块,被配置为根据所述反馈信息对所述BiLSTM-CRF模型进行增强学习。
可选地,所述人机对话系统还包括中央控制模块;
所述装置还包括:
词槽识别结果输出模块,被配置为将所述BiLSTM-CRF模型输出的词槽识别结果输出至所述中央控制模块;
目标词槽识别结果确定模块,被配置为获取所述中央控制模块从接收到的针对所述对话数据的词槽识别结果集合中决策出的目标词槽识别结果;
其中,所述词槽识别结果集合包括所述BiLSTM-CRF模型输出的词槽识别结果以及其他对话机器人输出的词槽识别结果;所述目标词槽识别结果用于作为所述人机对话系统对用户的回复结果输出。
可选地,所述反馈信息确定模块包括:
第一反馈信息确定子模块,被配置为响应于所述目标词槽识别结果与所述BiLSTM-CRF模型输出的词槽识别结果不一致,则将所述不一致的反馈信息确定为负反馈信息;
第二反馈信息确定子模块,被配置为响应于所述目标词槽识别结果与所述BiLSTM-CRF模型输出的词槽识别结果一致,则根据用户对所述回复结果的反馈操作确定反馈信息。
可选地,所述第二反馈信息确定子模块具体被配置为:
响应于用户的正向反馈率大于或等于预设阈值,将该反馈信息确定为正反馈信息;
响应于所述正向反馈率小于预设阈值,将该反馈信息确定为负反馈信息;
其中,所述正向反馈率为根据一段时间内用户对所述回复结果的反馈操作确定的。
可选地,所述模型增强学习模块具体被配置为:
将所述反馈信息反馈到所述BiLSTM-CRF模型中的CRF层,以由所述CRF层根据所述反馈信息进行模型增强训练。
根据本公开实施例的第三方面,提供一种人机对话设备,所述人机对话设备中包括对话机器人,所述人机对话设备包括:
处理器;
配置为存储处理器可执行指令的存储器;
其中,所述处理器被配置为:
确定使用的双向长短期记忆网络算法及条件随机场算法BiLSTM-CRF模型对用户输入的对话数据进行词槽识别后输出的词槽识别结果;
基于所述词槽识别结果以及用户对所述词槽识别结果的反馈确定反馈信息;
根据所述反馈信息对所述BiLSTM-CRF模型进行增强学习。
根据本公开实施例的第四方面,提供一种计算机可读存储介质,其上存储有计算机程序,所述程序被处理器执行时实现上述方法的步骤。
本公开的实施例提供的技术方案可以包括以下有益效果:
在本公开实施例中,以BiLSTM-CRF模型作为基本框架,当BiLSTM-CRF模型对外输出词槽识别结果以后,对话机器人可以根据该词槽识别结果获取对应的反馈信息,并根据该反馈信息对BiLSTM-CRF模型进行增强学习,从而实现模型的动态自学习,以减少人工标注的过程,提升词槽识别的效率和准确率。
应当理解的是,以上的一般描述和后文的细节描述仅是示例性和解释性的,并不能限制本公开实施例。
附图说明
此处的附图被并入说明书中并构成本说明书的一部分,示出了符合本公开的实施例,并与说明书一起用于解释本公开实施例的原理。
图1是本公开根据一示例性实施例示出的一种自然语言处理的方法实施例的步骤流程图;
图2是本公开根据一示例性实施例示出的又一种自然语言处理的方法实施例的步骤流程图;
图3是本公开根据一示例性实施例示出的BiLSTM-CRF模型示意图;
图4是本公开根据一示例性实施例示出的一种自然语言处理的装置实施例的框图;
图5是本公开根据一示例性实施例示出的一种人机交互设备的框图。
具体实施方式
这里将详细地对示例性实施例进行说明,其示例表示在附图中。下面的描述涉及附图时,除非另有表示,不同附图中的相同数字表示相同或相似的要素。以下示例性实施例中所描述的实施方式并不代表与本公开实施例相一致的所有实施方式。相反,它们仅是与如所附权利要求书中所详述的、本公开实施例的一些方面相一致的装置和方法的例子。
在本公开实施例使用的术语是仅仅出于描述特定实施例的目的,而非旨在限制本公开实施例。在本公开实施例和所附权利要求书中所使用的单数形式的“一种”、“所述”和“该”也旨在包括多数形式,除非上下文清楚地表示其他含义。还应当理解,本文中使用的术语“和/或”是指并包含一个或多个相关联的列出项目的任何或所有可能组合。
应当理解,尽管在本公开实施例可能采用术语第一、第二、第三等来描述各种信息,但这些信息不应限于这些术语。这些术语仅用来将同一类型的信息彼此区分开。例如,在不脱离本公开实施例范围的情况下,第一信息也可以被称为第二信息,类似地,第二信息也可以被称为第一信息。取决于语境,如在此所使用的词语“如果”可以被解释成为“在……时”或“当……时”或“响应于确定”。
参考图1是本公开根据一示例性实施例示出的一种自然语言处理的方法实施例的步骤流程图,本公开实施例的方法可以应用于人机对话系统中的对话机器人,具体可以包括如下步骤:
步骤101,确定使用的BiLSTM-CRF模型对用户输入的对话数据进行词槽识别后输出的词槽识别结果。
在该步骤中,当本对话机器人接收到用户输入的对话数据时,可以采用BiLSTM(Bi-directional Long Short-Term Memory,双向长短期记忆网络算法)-CRF(Conditional Random Field algorithm,条件随机场算法)模型对该对话数据进行词槽识别,并获取BiLSTM-CRF模型输出的词槽识别结果。其中,词槽识别可以理解为序列标注任务,解决序列标注问题。
步骤102,基于所述词槽识别结果以及用户对所述词槽识别结果的反馈确定反馈信息。
在该步骤中,当对话机器人获得词槽识别结果以后,可以进一步确定该词槽识别结果的反馈信息,示例性地,该反馈信息可以包括用户对该词槽识别结果的反馈。
步骤103,根据所述反馈信息对所述BiLSTM-CRF模型进行增强学习。
该步骤中,对话机器人可以根据获得的该词槽识别结果的反馈信息,对BiLSTM-CRF模型进行增强学习,从而实现模型的动态自学习,以减少人工标注的过程,提升词槽识别的效率和准确率。
参考图2是本公开根据一示例性实施例示出的另一种自然语言处理的方法实施例的步骤流程图,本公开实施例的方法可以应用于人机对话系统中的对话机器人,具体可以包括如下步骤:
步骤201,确定使用的BiLSTM-CRF模型对用户输入的对话数据进行词槽识别后输出的词槽识别结果。
词槽识别(Slot tagging)是NLU(Natural Language Understanding,自然语言理解)模块的重要组成部分,而NLU模块是人机对话系统中最核心的部分,整个NLU模块的功能主要包括如下两个:对用户意图(Intent)的理解,和对用户表达的语句中核心槽位(Slot,即词槽)的解析。意图(Intent)是一个分类器,确定用户表达的语句的类型,进而由确定的类型对应的程序(即Bot(语音机器人))进行专门的解析。比如用户说:“给我放一首快乐的歌吧”,这个时候可以判断用户的意图分类是音乐,因此可以召唤出音乐机器人(Bot)给用户推荐一首歌播放,用户听着觉得不对的时候,说:“换一首”,还是这个音乐机器人继续为用户服务,直到用户表达别的问题,意图已经不是音乐的时候,再切换成别的机器人为用户服务。
人机对话系统中还包括一个与NLU模块通信的中央控制模块,当NLU模块针对一个对话语句解析出超过一种意图时,可以由中央控制模块将用户的对话语句发送给声明可以处理用户意图的Bot(即ChatBot,对话机器人),并由这些Bot返回自己处理完毕的结果。
而Bot需要理解对话语句中的内容,为简便起见,可以只选择最核心重要的部分进行理解,并忽略其他非核心的内容,而那些最核心重要的部分称之为槽位(Slot),即词槽。
本公开实施例中的对话机器人Bot可以采用BiLSTM-CRF模型对用户输入的对话数据进行词槽识别。其中,词槽识别可以理解为序列标注任务,解决序列标注问题。
以下以中文句子为例,采用BiLSTM-CRF模型进行词槽识别过程如下:
将一个含有n个字的句子(字的序列)记作
x=(x 1,x 2,…,x n)
其中,x i表示句子第i个字在字典中的id,进而可以得到每个字的独热编码(one-hot)向量,维数是字典大小。
如图3的BiLSTM-CRF模型示意图所示:
模型的第一层是查找层(look-up layer),利用预训练或随机初始化的嵌入(embedding)矩阵将句子中的每个字由one-hot向量映射为低维稠密的字向量,在输入下一层之前,设置dropout(随机失活,是对具有深度结构的神经网络进行优化的方法,在学习过程中通过将隐含层的部分权重或输出随机归零,降低节点间的相互依赖性,从而实现神经网络的正则化,防止神经网络的过拟合)以缓解过拟合。
模型的第二层是双向LSTM(Long Short-Term Memory,长短期记忆网络)层,包括正向长短期记忆网络(forward LSTM)以及反向长短期记忆网络(backward LSTM)。将一个句子的各个字的char embedding(字符嵌入)序列(x 1,x 2,…,x n)作为双向LSTM各个时间步的输入,再将正向LSTM输出的
Figure PCTCN2019110894-appb-000001
隐状态序列与反向LSTM的
Figure PCTCN2019110894-appb-000002
在各个位置输出的隐状态进行按位置拼接,得到完整的隐状态序列。
在设置dropout后,接入一个线性层,将隐状态向量从m维映射到k维,k是标注集的标签数,从而由输出层(LSTM’s output)输出自动提取的句子特征,记作矩阵P=(p 1,p 2,…,p n)。
模型的第三层是CRF层,进行句子级的序列标注,CRF层的参数是一 个(k+2)*(k+2)的矩阵A,A ij表示的是从第i个标签到第j个标签的转移得分,进而在为一个位置进行标注的时候可以利用此前已经标注过的标签,之所以要加2是因为要为句子首部添加一个起始状态以及为句子尾部添加一个终止状态。
从图3可以看出,针对“中国很大”的句子,BiLSTM-CRF模型最终输出的词槽识别结果为:
Figure PCTCN2019110894-appb-000003
其中,在BIO标注集中,B-PER、I-PER分别表示人名首字、人名非首字;B-LOC、I-LOC分别表示地名首字、地名非首字;B-ORG、I-ORG分别表示组织机构名首字、组织机构名非首字;O表示该字不属于命名实体的一部分。当然,在特定领域中,还会相应地自定义领域内的各种实体类型,例如电影类型(video)、天气类型(weather)等类型。
步骤202,所述对话机器人将所述BiLSTM-CRF模型输出的词槽识别结果输出至所述中央控制模块。
步骤203,获取所述中央控制模块从接收到的针对所述对话数据的词槽识别结果集合中决策出的目标词槽识别结果。
在该实施例中,当对话机器人获得BiLSTM-CRF模型输出的词槽识别结果以后,可以将该词槽识别结果输出给中央控制模块。对于中央控制模块而言,其可以接收到来自不同对话机器人Bot发来的针对同一对话数据的词槽识别结果,组成词槽识别结果集合,则词槽识别结果集合可以包括BiLSTM-CRF模型输出的词槽识别结果以及其他对话机器人输出的词槽识别结果。
在实际中,当NLU模块针对一个对话语句解析出超过一种意图时,可以由中央控制模块进行决策,从多种意图中决策出一种意图作为用户意图,并确定出与用户意图对应的至少一个Bot分别进行解析处理,得到处理后 的词槽识别结果集合。随后,中央控制模块可以从词槽识别结果集合中决策出最匹配用户意图的目标词槽识别结果,该目标词槽识别结果可以用于作为人机对话系统对用户的回复结果输出。
步骤204,响应于所述目标词槽识别结果与所述BiLSTM-CRF模型输出的词槽识别结果不一致,则将所述不一致的反馈信息确定为负反馈信息。
本实施例可以结合目标词槽结果与BiLSTM-CRF模型输出的词槽识别结果的比较结果,来确定BiLSTM-CRF模型输出的词槽识别结果的反馈信息。
作为一种示例,该反馈信息可以包括正反馈信息以及负反馈信息。例如,正反馈信息可以用数值1表示,负反馈信息可以用数值-1表示。
在该步骤中,若BiLSTM-CRF模型输出的词槽识别结果与目标词槽识别结果不一致,则可以将反馈信息确定为负反馈信息。
在一种实现方式中,本对话机器人中可以设置反馈计算器(Reward Calculator)来记录当前词槽识别结果的反馈信息。例如,若上述反馈信息确定为负反馈信息,则Reward Calculator的记录为reward=-1。
步骤205,响应于所述目标词槽识别结果与所述BiLSTM-CRF模型输出的词槽识别结果一致,则根据用户对所述回复结果的反馈操作确定反馈信息。
在该步骤中,如果BiLSTM-CRF模型输出的词槽识别结果与目标词槽识别结果一致,则可以结合用户对回复结果的反馈操作来确定反馈信息。
也就是说,即使BiLSTM-CRF模型输出的词槽识别结果与目标词槽识别结果一致,也可以结合用户对回复结果的反馈操作来确定反馈信息,因为用户是对回复结果是否合理的真正评判端。
在本公开实施例的一种可能的实施方式中,步骤205可以包括如下子步骤:
子步骤S11,响应于用户的正向反馈率大于或等于预设阈值,将该反馈 信息确定为正反馈信息。
子步骤S12,响应于所述正向反馈率小于预设阈值,将该反馈信息确定为负反馈信息。
其中,正向反馈率为根据一段时间内用户对回复结果的反馈操作确定的。
在实际中,在一个时间段内,可能有多个用户向人机对话系统发出相同或类似的对话数据,则针对该对话数据,若BiLSTM-CRF模型输出的词槽识别结果作为回复结果,可以统计该时间段内多个用户对该回复结果的反馈操作(如点击或点赞等操作),如果正向反馈率大于或等于预设阈值,则表示用户的反馈是积极反馈,此时可以将该词槽识别结果的反馈信息确定为正反馈信息,即reward=1;否则,如果正向反馈率小于预设阈值,则表示用户的反馈是消极反馈,此时可以将该词槽识别结果的反馈信息确定为负反馈信息,即reward=-1。
当确定反馈信息以后,还可以将该反馈信息存储在缓存区中,以供后续使用。
步骤206,将所述反馈信息反馈到所述BiLSTM-CRF模型中的CRF层,以由所述CRF层根据所述反馈信息进行模型增强训练。
在该步骤中,Reward Calculator确定BiLSTM-CRF模型输出的词槽识别结果的反馈信息以后,可以将记录的反馈信息反馈到BiLSTM-CRF模型的CRF层中,CRF层将各个词槽识别结果以及对应的反馈信息作为训练数据进行模型的增强学习训练,通过增强学习完成整个学习的闭环,从而使得训练得到的BiLSTM-CRF模型能够获得更加准确的词槽识别结果。
在一种例子中,BiLSTM-CRF模型进行增强学习的四要素可以包括:
Action:词槽识别结果Y
State:待识别序列X
Policy:p(y|x),即在序列X的条件下,生成结果Y的概率
Reward:反馈信息。
在本实施例中,当获得BiLSTM-CRF模型输出的词槽识别结果以后,可以结合中央控制模块与用户对该词槽识别结果的反馈,确定反馈信息,免去用户标注的麻烦,将该反馈信息返回至BiLSTM-CRF模型的CRF层进行增强学习,可以提高BiLSTM-CRF模型的准确率。
为了使本领域技术人员能够更好地理解本公开实施例,以下列举具体的例子对本公开实施例进行示例性说明:
例如,用户输入的对话数据为“明日的天气”,由于“明日的天气”是一个电影名称,因此BiLSTM-CRF模型对“明日的天气”进行词槽识别后,输出的词槽识别结果为:明日的天气/video。中央控制模块接收到多个Bot发送的词槽识别结果后,最终决策的该对话数据的用户意图是天气,因此最终得到的目标词槽识别结果是“明日/date天气/weather”。Reward Calculator获得中央控制模块决策的目标词槽识别结果以后,将其与BiLSTM-CRF模型输出的词槽识别结果进行匹配,判定两者不一致(一个是video的结果,一个是weather的结果),因此,将该BiLSTM-CRF模型输出的词槽识别结果的反馈信息设置为reward=-1。
反之,如果中央控制模块最终决策的该对话数据的意图是video,因此最终得到的目标词槽识别结果是“明日的天气/video”。Reward Calculator获得中央控制模块决策的目标词槽识别结果以后,将其与BiLSTM-CRF模型输出的词槽识别结果进行匹配,判定两者一致(都是video的结果),此时,Reward Calculator会统计一段时间内用户对该目标词槽识别结果对应的回复结果(例如,中央控制模块向用户回复明日的天气的电影资源)的点击率,如果点击率比较低(低于预设阈值),此时可以认为该回复结果没有满足用户,则可以将该词槽识别结果的反馈信息设置为reward=-1,反之,如果点击率比较高(高于预设阈值),此时可以认为该回复结果满足用户,则可以将该词槽识别结果的反馈信息设置为reward=1。
Reward Calculator将反馈信息反馈到BiLSTM-CRF模型中,由模型进行动态自学习,假设reward=-1,则下次当模型再次接收到“明日的天气”的对话数据时,则判定其不属于video的标注结果,而是属于weather的标注结果,输出的识别结果为“明日/date天气/weather”,从而达到了学习的目的。
以上实施方式中的各种技术特征可以任意进行组合,只要特征之间的组合不存在冲突或矛盾,但是限于篇幅,未进行一一描述,因此上述实施方式中的各种技术特征的任意进行组合也属于本说明书公开的范围。
与前述自然语言处理的方法实施例相对应,本公开实施例还提供了自然语言处理的装置的实施例。
如图4所示,图4是本公开根据一示例性实施例示出的一种自然语言处理的装置实施例的框图,本公开实施例的装置应用于人机对话系统中的对话机器人,该装置具体可以包括如下模块:
词槽识别结果确定模块401,被配置为确定使用的BiLSTM-CRF模型对用户输入的对话数据进行词槽识别后输出的词槽识别结果;
反馈信息确定模块402,被配置为基于所述词槽识别结果以及用户对所述词槽识别结果的反馈确定反馈信息;
模型增强学习模块403,被配置为根据所述反馈信息对所述BiLSTM-CRF模型进行增强学习。
由上述实施例可见,本公开实施例通过词槽识别结果确定模块401获取BiLSTM-CRF模型输出的槽识别结果,并通过反馈信息确定模块402基于该词槽识别结果以及用户对该词槽识别结果的反馈确定反馈信息,以减少人工的标注量。然后通过模型增强学习模块403根据该反馈信息对BiLSTM-CRF模型进行增强学习,实现模型的动态自学习,从而可以提升模型的词槽识别的准确率。
在本公开实施例的一种可选实施例中,人机对话系统还包括中央控制 模块;所述装置还包括如下模块:
词槽识别结果输出模块,被配置为将所述BiLSTM-CRF模型输出的词槽识别结果输出至所述中央控制模块;
目标词槽识别结果确定模块,被配置为获取所述中央控制模块从接收到的针对所述对话数据的词槽识别结果集合中决策出的目标词槽识别结果;
其中,所述词槽识别结果集合包括所述BiLSTM-CRF模型输出的词槽识别结果以及其他对话机器人输出的词槽识别结果;所述目标词槽识别结果用于作为所述人机对话系统对用户的回复结果输出。
在本公开实施例的另一种可选实施例中,所述反馈信息确定模块402可以包括如下子模块:
第一反馈信息确定子模块,被配置为响应于所述目标词槽识别结果与所述BiLSTM-CRF模型输出的词槽识别结果不一致,则将所述不一致的反馈信息确定为负反馈信息;
第二反馈信息确定子模块,被配置为响应于所述目标词槽识别结果与所述BiLSTM-CRF模型输出的词槽识别结果一致,则根据用户对所述回复结果的反馈操作确定反馈信息。
由上述实施例可见,反馈信息可以包括中央处理模块的反馈信息以及用户的反馈信息,丰富了反馈信息的确定维度,以提高标注的准确性。
在本公开实施例的一种可选实施例中,所述第二反馈信息确定子模块具体被配置为:
响应于用户的正向反馈率大于或等于预设阈值,将该反馈信息确定为正反馈信息;
响应于所述正向反馈率小于预设阈值,将该反馈信息确定为负反馈信息;
其中,所述正向反馈率为根据一段时间内用户对所述回复结果的反馈操作确定的。
由上述实施例可知,在进行反馈信息的标注时,可以结合中央控制模块以及用户的反馈信息进行分析,从而可以提高标注的准确度。
在本公开实施例的一种可选实施例中,所述模型增强学习模块403具体被配置为:
将所述反馈信息反馈到所述BiLSTM-CRF模型中的CRF层,以由所述CRF层根据所述反馈信息进行模型增强训练。
由上述实施例可知,通过模型增强学习模块403根据反馈信息进行模型增强学习,可以实现模型的动态自学习,提升模型的词槽识别的准确率。
上述装置中各个模块的功能和作用的实现过程具体详情见上述系统实施例中的具体描述,在此不再赘述。
对于装置实施例而言,由于其基本对应于系统实施例,所以相关之处参见系统实施例的部分说明即可。以上所描述的装置实施例仅仅是示意性的,其中所述作为分离部件说明的单元可以是或者也可以不是物理上分开的,作为单元显示的部件可以是或者也可以不是物理单元,即可以位于一个地方,或者也可以分布到多个网络单元上。可以根据实际的需要选择其中的部分或者全部模块来实现本公开实施例方案的目的。本领域普通技术人员在不付出创造性劳动的情况下,即可以理解并实施。
如图5所示,图5是本公开根据一示例性实施例示出的一种人机对话设备500的框图。
参照图5,设备500可以包括以下一个或多个组件:处理组件502,存储器504,电源组件506,多媒体组件508,音频组件510,输入/输出(I/O)接口512,传感器组件514,以及通信组件516。
处理组件502通常控制设备500的整体操作,处理组件502可以包括一个或多个处理器520来执行指令,以完成上述的方法的全部或部分步骤。此外,处理组件502可以包括一个或多个模块,便于处理组件502和其他组件之间的交互。例如,处理组件502可以包括多媒体模块,以方便多媒 体组件508和处理组件502之间的交互。
存储器504被配置为存储各种类型的数据以支持在设备500的操作。这些数据的示例包括用于在设备500上操作的任何应用程序或方法的指令。存储器504可以由任何类型的易失性或非易失性存储设备或者它们的组合实现,如静态随机存取存储器(SRAM)、电可擦除可编程只读存储器(EEPROM)、可擦除可编程只读存储器(EPROM)、可编程只读存储器(PROM)、只读存储器(ROM)、磁存储器、快闪存储器、磁盘或光盘。
电源组件506为设备500的各种组件提供电力。电源组件506可以包括:电源管理系统,一个或多个电源,及其他与为设备500生成、管理和分配电力相关联的组件。
多媒体组件508包括在所述设备500和用户之间的提供一个输出接口的屏幕。在一些实施例中,屏幕可以包括液晶显示器(LCD)和触摸面板(TP)。如果屏幕包括触摸面板,屏幕可以被实现为触摸屏,以接收来自用户的输入信号。触摸面板包括一个或多个触摸传感器以感测触摸、滑动和触摸面板上的手势。所述触摸传感器不仅可以感测触摸或滑动动作的边界,而且还检测与所述触摸或滑动操作相关的持续时间和压力。音频组件510被配置为输出和/或输入音频信号。例如,音频组件510包括一个麦克风(MIC),当设备500处于操作模式,如呼叫模式、记录模式和语音识别模式时,麦克风被配置为接收外部音频信号。所接收的音频信号可以被进一步存储在存储器504或经由通信组件516发送。在一些实施例中,音频组件510还包括一个扬声器,用于输出音频信号。
I/O接口512为处理组件502和外围接口模块之间提供接口,上述外围接口模块可以是键盘、点击轮、按钮等。这些按钮可包括但不限于:主页按钮、音量按钮、启动按钮和锁定按钮。
传感器组件514包括一个或多个传感器,用于为设备500提供各个方面的状态评估。例如,传感器组件514可以检测到设备500的打开/关闭状 态、组件的相对定位,例如所述组件为设备500的显示器和小键盘,传感器组件514还可以检测:设备500或设备500中一个组件的位置改变,用户与设备500接触的存在或不存在,设备500方位或加速/减速和设备500的温度变化。传感器组件514可以包括接近传感器,被配置用来在没有任何的物理接触时检测附近物体的存在。传感器组件514还可以包括光传感器,如CMOS或CCD图像传感器,被配置为在成像应用中使用。在一些实施例中,该传感器组件514还可以包括加速度传感器、陀螺仪传感器、磁传感器、压力传感器或温度传感器。
通信组件516被配置为便于设备500和其他设备之间有线或无线方式的通信。设备500可以接入基于通信标准的无线网络,如WiFi、2G或5G、或它们的组合。在一个示例性实施例中,通信组件516经由广播信道接收来自外部广播管理系统的广播信号或广播相关信息。在一个示例性实施例中,所述通信组件516还包括近场通信(NFC)模块,以促进短程通信。例如,在NFC模块可基于射频识别(RFID)技术、红外数据协会(IrDA)技术、超宽带(UWB)技术、蓝牙(BT)技术和其他技术来实现。
在示例性实施例中,设备500可以被一个或多个应用专用集成电路(ASIC)、数字信号处理器(DSP)、数字信号处理设备(DSPD)、可编程逻辑器件(PLD)、现场可编程门阵列(FPGA)、控制器、微控制器、微处理器或其他电子元件实现,用于执行上述方法。
在示例性实施例中,还提供了一种包括指令的非临时性计算机可读存储介质,例如包括指令的存储器504,上述指令可由设备500的处理器520执行以完成上述方法。例如,所述非临时性计算机可读存储介质可以是ROM、CD-ROM、磁带、软盘和光数据存储设备等。
其中,当所述存储介质中的指令由所述处理器执行时,使得设备500能够执行一种自然语言处理的方法,包括:确定使用的BiLSTM-CRF模型对用户输入的对话数据进行词槽识别后输出的词槽识别结果;基于所述词 槽识别结果以及用户对所述词槽识别结果的反馈确定反馈信息;根据所述反馈信息对所述BiLSTM-CRF模型进行增强学习。
本领域技术人员在考虑说明书及实践这里公开的发明后,将容易想到本公开实施例的其它实施方案。本公开实施例旨在涵盖本公开实施例的任何变型、用途或者适应性变化,这些变型、用途或者适应性变化遵循本公开实施例的一般性原理并包括本公开实施例未公开的本技术领域中的公知常识或惯用技术手段。说明书和实施例仅被视为示例性的,本公开实施例的真正范围和精神由下面的权利要求指出。
应当理解的是,本公开实施例并不局限于上面已经描述并在附图中示出的精确结构,并且可以在不脱离其范围进行各种修改和改变。本公开实施例的范围仅由所附的权利要求来限制。
以上所述仅为本公开实施例的较佳实施例而已,并不用以限制本公开实施例,凡在本公开实施例的精神和原则之内,所做的任何修改、等同替换、改进等,均应包含在本公开实施例保护的范围之内。
工业实用性
本公开实施例中,以BiLSTM-CRF模型作为基本框架,当BiLSTM-CRF模型对外输出词槽识别结果以后,对话机器人可以根据该词槽识别结果获取对应的反馈信息,并根据该反馈信息对BiLSTM-CRF模型进行增强学习,从而实现模型的动态自学习,以减少人工标注的过程,提升词槽识别的效率和准确率。

Claims (12)

  1. 一种自然语言处理的方法,所述方法应用于人机对话系统中的对话机器人,所述方法包括:
    确定使用的双向长短期记忆网络算法及条件随机场算法BiLSTM-CRF模型对用户输入的对话数据进行词槽识别后输出的词槽识别结果;
    基于所述词槽识别结果以及用户对所述词槽识别结果的反馈确定反馈信息;
    根据所述反馈信息对所述BiLSTM-CRF模型进行增强学习。
  2. 根据权利要求1所述的方法,其中,所述人机对话系统还包括中央控制模块;
    在确定使用的所述BiLSTM-CRF模型对用户输入的对话数据进行词槽识别后输出的词槽识别结果之后,所述方法还包括:
    所述对话机器人将所述BiLSTM-CRF模型输出的词槽识别结果输出至所述中央控制模块;
    获取所述中央控制模块从接收到的针对所述对话数据的词槽识别结果集合中决策出的目标词槽识别结果;
    其中,所述词槽识别结果集合包括所述BiLSTM-CRF模型输出的词槽识别结果以及其他对话机器人输出的词槽识别结果;所述目标词槽识别结果用于作为所述人机对话系统对用户的回复结果输出。
  3. 根据权利要求2所述的方法,其中,所述基于所述词槽识别结果以及用户对所述词槽识别结果的反馈确定反馈信息包括:
    响应于所述目标词槽识别结果与所述BiLSTM-CRF模型输出的词槽识别结果不一致,则将所述不一致的反馈信息确定为负反馈信息;
    响应于所述目标词槽识别结果与所述BiLSTM-CRF模型输出的词槽 识别结果一致,则根据用户对所述回复结果的反馈操作确定反馈信息。
  4. 根据权利要求3所述的方法,其中,所述根据用户对所述回复结果的反馈操作确定反馈信息,包括:
    响应于用户的正向反馈率大于或等于预设阈值,将该反馈信息确定为正反馈信息;
    响应于所述正向反馈率小于预设阈值,将该反馈信息确定为负反馈信息;
    其中,所述正向反馈率为根据一段时间内用户对所述回复结果的反馈操作确定的。
  5. 根据权利要求1-4任一项所述的方法,其中,所述根据所述反馈信息进行模型增强学习,包括:
    将所述反馈信息反馈到所述BiLSTM-CRF模型中的CRF层,以由所述CRF层根据所述反馈信息进行模型增强训练。
  6. 一种自然语言处理的装置,所述装置应用于人机对话系统中的对话机器人,所述装置包括:
    词槽识别结果确定模块,被配置为确定使用的双向长短期记忆网络算法及条件随机场算法BiLSTM-CRF模型对用户输入的对话数据进行词槽识别后输出的词槽识别结果;
    反馈信息确定模块,被配置为基于所述词槽识别结果以及用户对所述词槽识别结果的反馈确定反馈信息;
    模型增强学习模块,被配置为根据所述反馈信息对所述BiLSTM-CRF模型进行增强学习。
  7. 根据权利要求6所述的装置,其中,所述人机对话系统还包括中央控制模块;
    所述装置还包括:
    词槽识别结果输出模块,被配置为将所述BiLSTM-CRF模型输出的 词槽识别结果输出至所述中央控制模块;
    目标词槽识别结果确定模块,被配置为获取所述中央控制模块从接收到的针对所述对话数据的词槽识别结果集合中决策出的目标词槽识别结果;
    其中,所述词槽识别结果集合包括所述BiLSTM-CRF模型输出的词槽识别结果以及其他对话机器人输出的词槽识别结果;所述目标词槽识别结果用于作为所述人机对话系统对用户的回复结果输出。
  8. 根据权利要求7所述的装置,其中,所述反馈信息确定模块包括:
    第一反馈信息确定子模块,被配置为响应于所述目标词槽识别结果与所述BiLSTM-CRF模型输出的词槽识别结果不一致,则将所述不一致的反馈信息确定为负反馈信息;
    第二反馈信息确定子模块,被配置为响应于所述目标词槽识别结果与所述BiLSTM-CRF模型输出的词槽识别结果一致,则根据用户对所述回复结果的反馈操作确定反馈信息。
  9. 根据权利要求8所述的装置,其中,所述第二反馈信息确定子模块具体被配置为:
    响应于用户的正向反馈率大于或等于预设阈值,将该反馈信息确定为正反馈信息;
    响应于所述正向反馈率小于预设阈值,将该反馈信息确定为负反馈信息;
    其中,所述正向反馈率为根据一段时间内用户对所述回复结果的反馈操作确定的。
  10. 根据权利要求6-9任一项所述的装置,其中,所述模型增强学习模块具体被配置为:
    将所述反馈信息反馈到所述BiLSTM-CRF模型中的CRF层,以由所述CRF层根据所述反馈信息进行模型增强训练。
  11. 一种人机对话设备,所述人机对话设备中包括对话机器人,所述人机对话设备包括:
    处理器;
    配置为存储处理器可执行指令的存储器;
    其中,所述处理器被配置为:
    确定使用的双向长短期记忆网络算法及条件随机场算法BiLSTM-CRF模型对用户输入的对话数据进行词槽识别后输出的词槽识别结果;
    基于所述词槽识别结果以及用户对所述词槽识别结果的反馈确定反馈信息;
    根据所述反馈信息对所述BiLSTM-CRF模型进行增强学习。
  12. 一种计算机可读存储介质,其上存储有计算机程序,所述程序被处理器执行时实现权利要求1-5任一项所述方法的步骤。
PCT/CN2019/110894 2019-07-29 2019-10-12 自然语言处理的方法、装置及设备 WO2021017173A1 (zh)

Priority Applications (3)

Application Number Priority Date Filing Date Title
RU2019141376A RU2726739C1 (ru) 2019-07-29 2019-10-12 Способ, аппарат и устройство для обработки естественного языка
JP2019564011A JP7101706B2 (ja) 2019-07-29 2019-10-12 自然言語処理方法、装置及び設備
KR1020197035003A KR102330061B1 (ko) 2019-07-29 2019-10-12 자연 언어 처리 방법, 장치, 기기 및 컴퓨터 판독 가능 저장매체

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201910687763.0 2019-07-29
CN201910687763.0A CN110413756B (zh) 2019-07-29 2019-07-29 自然语言处理的方法、装置及设备

Publications (1)

Publication Number Publication Date
WO2021017173A1 true WO2021017173A1 (zh) 2021-02-04

Family

ID=68363742

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/110894 WO2021017173A1 (zh) 2019-07-29 2019-10-12 自然语言处理的方法、装置及设备

Country Status (7)

Country Link
US (1) US11501078B2 (zh)
EP (1) EP3772012A1 (zh)
JP (1) JP7101706B2 (zh)
KR (1) KR102330061B1 (zh)
CN (1) CN110413756B (zh)
RU (1) RU2726739C1 (zh)
WO (1) WO2021017173A1 (zh)

Families Citing this family (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110909541A (zh) * 2019-11-08 2020-03-24 杭州依图医疗技术有限公司 指令生成方法、系统、设备和介质
CN111198937B (zh) * 2019-12-02 2023-06-30 泰康保险集团股份有限公司 对话生成及装置、计算机可读存储介质、电子设备
CN111597808B (zh) * 2020-04-24 2023-07-25 北京百度网讯科技有限公司 仪表盘绘制处理方法、装置、电子设备和存储介质
CN111681647B (zh) * 2020-06-10 2023-09-05 北京百度网讯科技有限公司 用于识别词槽的方法、装置、设备以及存储介质
CN112507126B (zh) * 2020-12-07 2022-11-15 厦门渊亭信息科技有限公司 一种基于循环神经网络的实体链接装置和方法
CN113515822B (zh) * 2021-01-28 2022-11-29 长春工业大学 一种基于归零神经网络的张拉整体结构找形方法
CN113268995B (zh) * 2021-07-19 2021-11-19 北京邮电大学 中文学术关键词抽取方法、装置和存储介质
US11520815B1 (en) * 2021-07-30 2022-12-06 Dsilo, Inc. Database query generation using natural language text
CN115099224A (zh) * 2022-07-08 2022-09-23 江苏理工学院 融合BiLSTM+CRF与规则匹配的中文PDF内容抽取方法和装置
CN117173491B (zh) * 2023-10-24 2024-02-23 广州思德医疗科技有限公司 医学图像的标注方法、装置、电子设备及存储介质

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20170372199A1 (en) * 2016-06-23 2017-12-28 Microsoft Technology Licensing, Llc Multi-domain joint semantic frame parsing
CN108108449A (zh) * 2017-12-27 2018-06-01 哈尔滨福满科技有限责任公司 一种面向医疗领域的基于多源异构数据问答系统及该系统的实现方法
CN109063035A (zh) * 2018-07-16 2018-12-21 哈尔滨工业大学 一种面向出行领域的人机多轮对话方法
WO2019070684A1 (en) * 2017-10-03 2019-04-11 Google Llc AUTOMATED PROGRAMMABLE ASSISTANT BY USER
WO2019084810A1 (zh) * 2017-10-31 2019-05-09 腾讯科技(深圳)有限公司 一种信息处理方法及终端、计算机存储介质
US20190156210A1 (en) * 2017-11-17 2019-05-23 Facebook, Inc. Machine-Learning Models Based on Non-local Neural Networks

Family Cites Families (23)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20070106496A1 (en) * 2005-11-09 2007-05-10 Microsoft Corporation Adaptive task framework
GB2440316A (en) 2006-07-25 2008-01-30 Optinose As Nasal inhaler with scrubber
US9978365B2 (en) * 2008-10-31 2018-05-22 Nokia Technologies Oy Method and system for providing a voice interface
EP2839391A4 (en) 2012-04-20 2016-01-27 Maluuba Inc CONVERSATION AGENT
RU2665239C2 (ru) * 2014-01-15 2018-08-28 Общество с ограниченной ответственностью "Аби Продакшн" Автоматическое извлечение именованных сущностей из текста
RU2595489C2 (ru) * 2014-06-18 2016-08-27 Самсунг Электроникс Ко., Лтд. Выделение временных выражений для текстов на естественном языке
MX2018002346A (es) 2015-08-28 2018-04-11 Agbiome Inc Composiciones y metodos para controlar enfermedades de plantas.
WO2017083504A1 (en) 2015-11-12 2017-05-18 Semantic Machines, Inc. Interaction assistant
CN107305575B (zh) * 2016-04-25 2021-01-26 北京京东尚科信息技术有限公司 人机智能问答系统的断句识别方法和装置
CN106445147B (zh) * 2016-09-28 2019-05-10 北京百度网讯科技有限公司 基于人工智能的对话系统的行为管理方法及装置
CN106569998A (zh) * 2016-10-27 2017-04-19 浙江大学 一种基于Bi‑LSTM、CNN和CRF的文本命名实体识别方法
GB2559617B (en) 2017-02-13 2020-06-24 Toshiba Kk A dialogue system, a dialogue method and a method of adapting a dialogue system
GB2559618B (en) * 2017-02-13 2020-07-08 Toshiba Kk A dialogue system, a dialogue method and a method of adapting a dialogue system
CN106919702B (zh) * 2017-02-14 2020-02-11 北京时间股份有限公司 基于文档的关键词推送方法及装置
CN106933809A (zh) * 2017-03-27 2017-07-07 三角兽(北京)科技有限公司 信息处理装置及信息处理方法
JP6851894B2 (ja) * 2017-04-24 2021-03-31 株式会社東芝 対話システム、対話方法及び対話プログラム
US11933699B2 (en) 2017-07-06 2024-03-19 Waters Technologies Corporation Devices and methods for preparing filtered solutions for lateral flow testing
JP6805112B2 (ja) * 2017-11-08 2020-12-23 株式会社東芝 対話システム、対話方法および対話プログラム
CN108363690A (zh) * 2018-02-08 2018-08-03 北京十三科技有限公司 基于神经网络的对话语义意图预测方法及学习训练方法
CN109003605B (zh) * 2018-07-02 2020-04-21 北京百度网讯科技有限公司 智能语音交互处理方法、装置、设备及存储介质
CN109697679A (zh) * 2018-12-27 2019-04-30 厦门智融合科技有限公司 知识产权服务导引方法及系统
US20200293874A1 (en) * 2019-03-12 2020-09-17 Microsoft Technology Licensing, Llc Matching based intent understanding with transfer learning
CN109918680B (zh) * 2019-03-28 2023-04-07 腾讯科技(上海)有限公司 实体识别方法、装置及计算机设备

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20170372199A1 (en) * 2016-06-23 2017-12-28 Microsoft Technology Licensing, Llc Multi-domain joint semantic frame parsing
WO2019070684A1 (en) * 2017-10-03 2019-04-11 Google Llc AUTOMATED PROGRAMMABLE ASSISTANT BY USER
WO2019084810A1 (zh) * 2017-10-31 2019-05-09 腾讯科技(深圳)有限公司 一种信息处理方法及终端、计算机存储介质
US20190156210A1 (en) * 2017-11-17 2019-05-23 Facebook, Inc. Machine-Learning Models Based on Non-local Neural Networks
CN108108449A (zh) * 2017-12-27 2018-06-01 哈尔滨福满科技有限责任公司 一种面向医疗领域的基于多源异构数据问答系统及该系统的实现方法
CN109063035A (zh) * 2018-07-16 2018-12-21 哈尔滨工业大学 一种面向出行领域的人机多轮对话方法

Also Published As

Publication number Publication date
CN110413756A (zh) 2019-11-05
EP3772012A1 (en) 2021-02-03
JP2021535411A (ja) 2021-12-16
RU2726739C1 (ru) 2020-07-15
CN110413756B (zh) 2022-02-15
US11501078B2 (en) 2022-11-15
KR20210016262A (ko) 2021-02-15
JP7101706B2 (ja) 2022-07-15
KR102330061B1 (ko) 2021-11-26
US20210034966A1 (en) 2021-02-04

Similar Documents

Publication Publication Date Title
WO2021017173A1 (zh) 自然语言处理的方法、装置及设备
US20200334420A1 (en) Contextual language generation by leveraging language understanding
US20210405868A1 (en) Neural Network for Keyboard Input Decoding
CN107102746B (zh) 候选词生成方法、装置以及用于候选词生成的装置
WO2019233219A1 (zh) 对话状态确定方法及装置、对话系统、计算机设备、存储介质
CN110797019B (zh) 多命令单一话语输入方法
CN107436691B (zh) 一种输入法进行纠错的方法、客户端、服务器及装置
WO2020073532A1 (zh) 客服机器人对话状态识别方法及装置、电子设备、计算机可读存储介质
WO2019212729A1 (en) Generating response based on user's profile and reasoning on contexts
US11978453B2 (en) Natural language processing routing
CN110674246A (zh) 问答模型训练方法、自动问答方法及装置
CN107767205A (zh) 评价信息的展示系统、方法、客户端及处理方法、服务器
CN111144093A (zh) 一种智能文本处理方法、装置、电子设备及存储介质
KR20210032875A (ko) 음성 정보 처리 방법, 장치, 프로그램 및 저장 매체
CN108803890A (zh) 一种输入方法、输入装置和用于输入的装置
CN114154459A (zh) 语音识别文本处理方法、装置、电子设备及存储介质
CN112036174B (zh) 一种标点标注方法及装置
US11640823B1 (en) Natural language processing routing
US11705110B2 (en) Electronic device and controlling the electronic device
CN113590831B (zh) 自然语言处理方法、设备及存储介质
CN109976548B (zh) 一种输入方法及输入装置
CN113591495A (zh) 语音翻译方法、装置及存储介质
KR20200021409A (ko) 전자 장치 및 그 제어 방법
TW201916003A (zh) 基於機率規則之對話理解方法
US20220261554A1 (en) Electronic device and controlling method of electronic device

Legal Events

Date Code Title Description
ENP Entry into the national phase

Ref document number: 2019564011

Country of ref document: JP

Kind code of ref document: A

121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 19939190

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 19939190

Country of ref document: EP

Kind code of ref document: A1