A semantic recognition method and system based on dual channel feature matching

Abstract

The present invention provides a semantic recognition method and system based on dual channel feature matching, including: obtaining question information sent by a user and the sending channel feature of the question information; obtaining the user's intention according to the question information; obtaining the corresponding standard question according to the intention; judging whether there is a semantic channel feature in the question information, if so, sending the user the answer information matching the semantic channel feature under the standard question, otherwise, sending the user the answer information matching the sending channel feature under the standard question. The present invention improves the accuracy of the customer service system's questions and answers through the combined judgment of the source channel feature and the user's semantic channel feature, and enables customers to obtain a better quality experience service.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a semantic recognition method and system based on dual-channel feature matching. Background Art

[0002] Generally speaking, an intelligent interactive system based on natural semantic processing interacts with users in a serial manner through the following steps: first, the adaptation module receives the questions sent by the user and performs basic word segmentation and matching processing; then it enters the natural language processing (NLP) module to perform part of speech, word vector, and entity recognition through hidden Markov models, deep learning models, etc. to obtain user intent; then it completes the matching and scoring of Frequently Asked Questions (FAQ) according to the search and deep learning combined framework, and returns the corresponding knowledge answers.

[0003] In order to interact with users in a targeted manner, the current system sends channel features to the intelligent interaction system at the adaptation layer based on the different source channels connected to the intelligent interaction system. In the FAQ answer maintenance, corresponding labels for different source channel features are set and provided to the system to match the returned answers based on the sent channel features, so that more accurate answers can be obtained.

[0004] For the existing intelligent interaction system based on natural semantic processing, when processing customer Q&A requests from different source channels, the channel features sent up are routed to different FAQ answer branches to present customer content. This can obtain targeted and accurate answers and improve customer experience, but it will also cause another business pain point that the channel is too targeted and cannot achieve full coverage of service effects, making it difficult for users to obtain high-quality full-volume services. Summary of the invention

[0005] 1. Technical issues to be resolved

[0006] In view of the above problems, the present invention provides a semantic recognition method and system based on dual channel feature matching, which is used to at least partially solve the technical problems of traditional methods such as excessive channel targeting and inability to provide full coverage services.

[0007] (II) Technical solution

[0008] On the one hand, the present invention provides a semantic recognition method based on dual channel feature matching, including: obtaining question information sent by a user and the sending channel characteristics of the question information; obtaining the user's intention based on the question information; obtaining the corresponding standard question based on the intention; judging whether there is a semantic channel feature in the question information, and if so, sending the user answer information matching the semantic channel feature under the standard question, otherwise, sending the user answer information matching the sending channel feature under the standard question.

[0009] Furthermore, obtaining the user intention according to the question information includes: processing the question information through natural language to obtain semantic features of the question information, and obtaining an intention candidate set according to the semantic features, and selecting the user intention from the intention candidate set.

[0010] Furthermore, selecting the user intent from the intent candidate set includes: selecting the intent with the highest matching degree from the intent candidate set as the user intent.

[0011] Furthermore, obtaining corresponding standard questions according to the intention includes: searching based on a pre-built FAQ database to determine corresponding standard questions, and the standard questions correspond to answers from at least one channel.

[0012] Furthermore, the pre-built FAQ database is provided with a "semantic channel feature" dimension, which is parallel to the "sending channel feature" dimension.

[0013] Furthermore, the dimension of “semantic channel characteristics” and the dimension of “sending channel characteristics” fall into the FAQ database in the form of tags.

[0014] Furthermore, judging whether there is a semantic channel feature in the question information includes: extracting question entities from the question information through a syntactic template, and judging whether there is a semantic channel feature.

[0015] Furthermore, the form of obtaining the question information sent by the user includes at least one of text, audio, and video files.

[0016] Furthermore, obtaining the question information sent by the user includes: performing text processing on the question information, and the text processing method includes text segmentation.

[0017] On the other hand, the present invention provides a semantic recognition system based on dual channel feature matching, including: a receiving module, used to receive question information sent by a user and the sending channel characteristics of the question information; a natural language processing module, used to obtain the user's intention based on the question information; a generation module, used to obtain the corresponding standard question based on the intention; an output module, used to determine whether there is a semantic channel feature in the question information, and if so, send the user answer information matching the semantic channel feature under the standard question, otherwise, send the user answer information matching the sending channel feature under the standard question.

[0018] Furthermore, it also includes: a text processing module, which is used to perform text processing on the question information, and the text processing method includes text segmentation.

[0019] Another aspect of the present invention provides an electronic device, comprising: a processor; a memory storing a computer executable program, which, when executed by the processor, enables the processor to execute the aforementioned semantic recognition method based on dual-channel feature matching.

[0020] Another aspect of the present invention provides a computer-readable medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the aforementioned semantic recognition method based on dual-channel feature matching.

[0021] (III) Beneficial effects

[0022] The disclosed embodiments provide a semantic recognition method and system based on dual channel feature matching, which returns the corresponding answer after comprehensive judgment based on the semantic channel features of questions submitted by users and the features submitted by the source channel, thereby improving the accuracy of questions and answers in the customer service system and enabling customers to obtain a better quality experience service. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 The application scenario of the semantic recognition method and system based on dual-channel feature matching according to the embodiment of the present disclosure is schematically shown;

[0024] Figure 2 A flowchart of a semantic recognition method based on dual-channel feature matching according to an embodiment of the present disclosure is schematically shown;

[0025] Figure 3 A block diagram of a semantic recognition system based on dual-channel feature matching according to an embodiment of the present disclosure is schematically shown;

[0026] Figure 4 A flowchart schematically shows a specific method for semantic recognition based on dual-channel feature matching according to an embodiment of the present disclosure;

[0027] Figure 5The block diagram of a computer system according to an embodiment of the present disclosure is schematically shown. DETAILED DESCRIPTION

[0028] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with specific embodiments and with reference to the accompanying drawings.

[0029] The terms used herein are only for describing specific embodiments and are not intended to limit the present disclosure. The terms "comprise", "include", etc. used herein indicate the existence of features, steps, operations and / or components, but do not exclude the existence or addition of one or more other features, steps, operations or components.

[0030] All terms (including technical and scientific terms) used herein have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein should be interpreted as having a meaning consistent with the context of this specification, and should not be interpreted in an idealized or overly rigid manner.

[0031] Some block diagrams and / or flow charts are shown in the accompanying drawings. It should be understood that some blocks or combinations thereof in the block diagrams and / or flow charts may be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that these instructions, when executed by the processor, may create a device for implementing the functions / operations described in these block diagrams and / or flow charts.

[0032] Therefore, the technology of the present disclosure can be implemented in the form of hardware and / or software (including firmware, microcode, etc.). In addition, the technology of the present disclosure can take the form of a computer program product on a computer-readable medium storing instructions, which can be used by an instruction execution system or in combination with an instruction execution system. In the context of the present disclosure, a computer-readable medium can be any medium that can contain, store, transmit, propagate, or transmit instructions. For example, a computer-readable medium can include, but is not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, device, or propagation medium. Specific examples of computer-readable media include: magnetic storage devices, such as magnetic tapes or hard disks (HDDs); optical storage devices, such as compact disks (CD-ROMs); memories, such as random access memories (RAMs) or flash memory; and / or wired / wireless communication links.

[0033] The embodiment of the present disclosure provides a semantic recognition method based on dual channel feature matching, see Figure 1, including: obtaining the question information sent by the user and the sending channel characteristics of the question information; obtaining the user's intention based on the question information; obtaining the corresponding standard question based on the intention; judging whether there is a semantic channel characteristic in the question information, if so, sending the answer information matching the semantic channel characteristic under the standard question to the user, otherwise, sending the answer information matching the sending channel characteristic under the standard question to the user.

[0034] Figure 1 The application scenario 100 of the semantic recognition method and system based on dual channel feature matching according to an embodiment of the present disclosure is schematically shown. It should be noted that: Figure 1 What is shown are merely examples of application scenarios to which the embodiments of the present disclosure can be applied, in order to help those skilled in the art understand the technical content of the present disclosure, but it does not mean that the embodiments of the present disclosure cannot be used in other devices, systems, environments or scenarios.

[0035] like Figure 1 As shown, according to the embodiments of the present disclosure, the application scenarios of the semantic recognition method and system based on dual-channel feature matching can be applicable to customer service systems, including customer service systems of electronic devices such as PCs, mobile phones, and tablets. The present disclosure does not make specific limitations, and any electronic device that can open the customer service system is within the protection scope of the present disclosure.

[0036] For the existing intelligent interactive system based on natural semantic processing, when processing customer Q&A requests from different source channels, the channel features are sent to different FAQ answer branches for customer content presentation. The pain point business scenario of this disclosure is described as follows: Please refer to Figure 1 , the FAQ knowledge is [How to apply for a credit card], and the answer for users of the PC channel is: please log in to the ICBC portal homepage; the answer for mobile channel users is: please log in to the ICBC APP. When the user asks [How to apply for a credit card on a mobile phone] through the PC channel, after the intention is recognized and the channel feature is matched, the intelligent interactive system answers [Please log in to the ICBC portal homepage], which does not meet the customer's expectations.

[0037] Figure 2 The flowchart of the semantic recognition method based on dual-channel feature matching according to an embodiment of the present disclosure is schematically shown.

[0038] like Figure 2 As shown, the method includes operations S1 to S4. Among them:

[0039] In S1, the questions submitted by users enter the intelligent interactive system, which obtains the content of the question information sent by the user and the channel through which it is sent, including PC channel, mobile phone channel, tablet channel, etc.

[0040] In S2, the content of the question information is processed, including part of speech, word vector, and entity recognition through hidden Markov models and deep learning models to obtain user intent;

[0041] In S3, standard questions matching the user's intention are retrieved from the pre-built FAQ database;

[0042] In S4, it is determined whether the question information sent by the user contains the content of the semantic channel. If so, the answer matching the semantic channel characteristics under the standard question is returned; if not, the answer matching the sent channel characteristics under the standard question is returned.

[0043] Therefore, when the user asks the question [How to apply for a credit card via mobile phone] through the PC channel, after the intention is recognized and the semantic channel feature is matched, the intelligent interactive system answers [Please log in to ICBC APP], meeting the customer's expectations. Therefore, the semantic recognition method disclosed in the present invention can more accurately understand the questions that customers actually want to ask, rather than simply returning a standard answer to the customer based on the hard channel identifier. Traditional methods often ignore the restrictive content in the question information, making the channel too targeted and unable to achieve a full coverage service effect. The method disclosed in the present invention uses a method of judging by a combination of sending channel features and semantic channel features, and gives priority to semantic channel features, so that customers can obtain a more comprehensive and more accurate question-and-answer experience.

[0044] Based on the above embodiment, obtaining user intent according to question information includes: processing the question information through natural language to obtain semantic features of the question information, obtaining an intent candidate set according to the semantic features, and selecting user intent from the intent candidate set.

[0045] Natural language processing enables computers to understand the meaning of natural language texts and realize dialogue between people and computers. It mainly includes corpus preprocessing steps; feature engineering steps, which represent the words and phrases after word segmentation into types that can be calculated by computers; model training steps; indicator evaluation steps, which perform necessary evaluations on the model so that the model has good generalization capabilities for the corpus. Entering NLP, part of speech, word vector, and entity recognition are performed through hidden Markov models, deep learning models, etc. to obtain the semantic features of question information, and based on the semantic features, the intent candidate set is obtained, where the intent candidate set includes standard questions under several intent categories.

[0046] Based on the above embodiment, selecting the user intent from the intent candidate set includes: selecting the intent with the highest matching degree from the intent candidate set as the user intent.

[0047] After obtaining the candidate set of intents, each intent in the candidate set needs to be scored to obtain the degree of match between each intent and the question information sent by the user. The intent with the highest score is the intent with the highest match, which is the intent that best matches the customer's question information. The intent with the highest match is selected as the user intent.

[0048] On the basis of the above embodiment, obtaining the corresponding standard question according to the intention includes: searching based on a pre-built FAQ database to determine the corresponding standard question, and the standard question corresponds to an answer from at least one channel.

[0049] In the FAQ database, multiple standard questions and standard answers are combined to form multiple "standard question-standard answer" pairs. Here, each standard question corresponds to multiple standard answers for sending channels. For example, the question is [How to apply for a credit card]. For users of the PC channel, the answer is configured as follows: please log in to the ICBC portal homepage; for mobile channel users, the answer is configured as follows: please log in to the ICBC APP. In the FAQ answer maintenance, corresponding feature labels of different source channels are set for the system to return answers based on the channel feature matching sent up, so that more accurate answers can be obtained. Search through the FAQ database to determine the standard questions that match the user's question intent, obtain the answers to the standard questions that match the user's question intent and the sending channel, and output the answers.

[0050] On the basis of the above-mentioned embodiment, the pre-constructed FAQ database is provided with a "semantic channel feature" dimension, which is parallel to the "sending channel feature" dimension.

[0051] The present invention newly designs a "feature value" dimension in the FAQ answer maintenance of the intelligent interactive system. This dimension is parallel to the "access channel" dimension and is implemented in the form of a syntactic template.

[0052] On the basis of the above-mentioned embodiment, the dimension of “semantic channel characteristics” and the dimension of “sending channel characteristics” are included in the FAQ database in the form of tags.

[0053] Both dimensions are included in the FAQ database in a marked form, and are used for answer selection after the intent matching is completed. Before the intelligent interactive system is used for customer interaction, the business maintains the corresponding "feature value" dimension based on the expected semantic channel features of the hotspot triggered answer. This maintenance does not affect the NLP model itself, and does not require retraining, and can take effect immediately.

[0054] Based on the above embodiment, judging whether there is a semantic channel feature in the question information includes: extracting question entities from the question information through a syntactic template, and judging whether there is a semantic channel feature.

[0055] The syntactic template decomposes complex problems by extracting problem entities, matching templates, and generating problem dependency graphs. Determine whether there are semantic channel features. For example, when a user asks: How to apply for a credit card with a mobile phone, the mobile phone is the channel feature contained in the user's question information. Compared with the user's sending channel feature, this channel feature is more in line with the user's intention. Therefore, the semantic channel feature is given priority.

[0056] Based on the above embodiment, the form of obtaining the question information sent by the user includes at least one of text, audio, and video files.

[0057] The question information sent by the user may be in the form of a text, a voice or a video. Of course, other sending methods are also feasible as long as the receiving end can recognize the information.

[0058] On the basis of the above embodiment, obtaining the question information sent by the user includes: performing text processing on the question information, and the text processing method includes text segmentation.

[0059] Before obtaining the user's intention based on the question information, the question information is also preprocessed; if it is text, it is directly segmented; if it is audio, speech recognition is performed, and the result of speech recognition is converted into text data before word segmentation; if it is a video, it is parsed into text data and images, and the image is recognized and converted into text data before word segmentation. The word segmentation processing of text information also includes marking and splitting it according to the input text information; parameter labeling, matching the marked and split text information according to the predetermined rules, and extracting the key information; the output of key information, according to the reply rules, the key information is generated into the corresponding reply template and output.

[0060] Another embodiment of the present disclosure provides a semantic recognition system 300 based on dual channel feature matching, see Figure 3 , including: a receiving module, used to receive question information sent by a user and the sending channel characteristics of the question information; a natural language processing module, used to obtain the user's intention based on the question information; a generating module, used to obtain the corresponding standard question based on the intention; an output module, used to determine whether there is a semantic channel feature in the question information, and if so, send the user the answer information matching the semantic channel feature under the standard question, otherwise, send the user the answer information matching the sending channel feature under the standard question.

[0061] Figure 3The structure diagram of the related technical information processing method is schematically shown. The receiving module is used to obtain the content of the question information sent by the user and the sending channel, which includes PC channel, mobile phone channel, tablet channel, etc.; the NLP module is used to obtain the user's intention; the generation module is used to obtain the standard question matching the user's intention through retrieval; the output module is used to determine whether the question information sent by the user contains the content of the semantic channel. If so, the answer matching the semantic channel feature under the standard question is returned; if not, the answer matching the channel feature sent under the standard question is returned.

[0062] On the basis of the above embodiment, it further includes: a text processing module, which is used to perform text processing on the question information, and the text processing method includes text segmentation.

[0063] The word segmentation processing of text information includes marking and splitting the input text information; parameter marking, matching the marked and split text information according to predetermined rules, and extracting key information; output of key information, generating the key information into the corresponding reply template according to the reply rules, and outputting it.

[0064] Figure 4 This is a specific embodiment of the semantic recognition method disclosed in the present invention. After the user submits a question into the intelligent interactive system, the question information sent by the user is obtained, which is S1; the adaptation module performs basic word segmentation and matching processing; then the system enters the NLP module to perform part of speech, word vector, and entity recognition through hidden Markov models, deep learning models, etc. to obtain user intent, which is S2; after scoring the intent, the matching standard question information is obtained, and multiple channel answers are hung under the standard question, which is S3. First, the syntactic template in the answer is used to determine whether there is channel information in the question submitted by the customer. If so, the corresponding answer is returned. Otherwise, the channel information is determined through the hard channel identifier sent from the application side and the answer is returned, which is S4.

[0065] Figure 5 A block diagram of an electronic device according to another embodiment of the present disclosure is schematically shown.

[0066] like Figure 5 As shown, the electronic device 500 includes a processor 510 and a computer-readable storage medium 520. The electronic device 500 can execute the method according to the embodiment of the present disclosure.

[0067] Specifically, the processor 510 may include, for example, a general-purpose microprocessor, an instruction set processor and / or a related chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 510 may also include an onboard memory for cache purposes. The processor 510 may be a single processing unit or multiple processing units for executing different actions of the method flow according to an embodiment of the present disclosure.

[0068] Computer readable storage medium 520, for example, can be any medium capable of containing, storing, conveying, propagating or transmitting instructions. For example, readable storage media can include, but are not limited to, electronic, magnetic, optical, electromagnetic, infrared or semiconductor systems, devices, devices or propagation media. Specific examples of readable storage media include: magnetic storage devices, such as magnetic tapes or hard disks (HDDs); optical storage devices, such as compact disks (CD-ROMs); memory, such as random access memory (RAM) or flash memory; and / or wired / wireless communication links.

[0069] The computer-readable storage medium 520 may include a computer program 521 , which may include code / computer-executable instructions, which, when executed by the processor 510 , enable the processor 510 to perform the method flow according to the embodiment of the present disclosure and any variation thereof.

[0070] The computer program 521 may be configured to have, for example, a computer program code including a computer program module. For example, in an exemplary embodiment, the code in the computer program 521 may include one or more program modules, for example, 521A, module 521B, ... It should be noted that the division method and number of modules are not fixed, and those skilled in the art may use appropriate program modules or program module combinations according to actual conditions, and when these program module combinations are executed by the processor 510, the processor 510 may execute the method flow according to the embodiment of the present disclosure and any variation thereof.

[0071] The present disclosure also provides a computer-readable storage medium, which may be included in the device / system / system described in the above embodiments, or may exist independently without being assembled into the device / system / system. The above computer-readable storage medium carries one or more programs, and when the above one or more programs are executed, the method according to the embodiment of the present disclosure is implemented.

[0072] Although the present disclosure has been shown and described with reference to specific exemplary embodiments of the present disclosure, it should be understood by those skilled in the art that various changes in form and details may be made to the present disclosure without departing from the spirit and scope of the present disclosure as defined by the appended claims and their equivalents. Therefore, the scope of the present disclosure should not be limited to the above-mentioned embodiments, but should be determined not only by the appended claims, but also by the equivalents of the appended claims.

[0073] The specific embodiments described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A semantic recognition method based on dual channel feature matching, comprising: Obtain the question information sent by the user and the characteristics of the sending channel of the question information; Acquire the user intention according to the question information; Obtaining corresponding standard questions according to the intention; Determine whether there is a semantic channel feature in the question information. If so, send the answer information matching the semantic channel feature under the standard question to the user. Otherwise, send the answer information matching the sending channel feature under the standard question to the user, with the semantic channel feature as priority; wherein the channel includes one of a PC channel, a mobile phone channel, and a tablet channel.

2. The semantic recognition method based on dual channel feature matching according to claim 1 is characterized in that: The obtaining of the user intention according to the question information comprises: The question information is processed by natural language to obtain semantic features of the question information, and an intent candidate set is obtained based on the semantic features, and the user intent is selected from the intent candidate set.

3. The semantic recognition method based on dual channel feature matching according to claim 2 is characterized in that: The selecting the user intent from the intent candidate set comprises: The intent with the highest matching degree is selected from the intent candidate set as the user intent.

4. The semantic recognition method based on dual channel feature matching according to claim 1 is characterized in that: The obtaining of the corresponding standard questions according to the intention includes: A search is performed based on a pre-built FAQ database to determine a corresponding standard question, where the standard question corresponds to an answer from at least one channel.

5. The semantic recognition method based on dual channel feature matching according to claim 4 is characterized in that: The pre-built FAQ database is provided with a "semantic channel feature" dimension, which is parallel to the "sending channel feature" dimension.

6. The semantic recognition method based on dual channel feature matching according to claim 5 is characterized in that: The "semantic channel characteristics" dimension and the "sending channel characteristics" dimension are included in the FAQ database in a marked form.

7. The semantic recognition method based on dual channel feature matching according to claim 1 is characterized in that: The determining whether there is a semantic channel feature in the question information includes: The question entity is extracted from the question information through the syntactic template, and it is determined whether there is a semantic channel feature.

8. The semantic recognition method based on dual channel feature matching according to claim 1 is characterized in that: The form of obtaining the question information sent by the user includes at least one of text, audio, and video files.

9. The semantic recognition method based on dual channel feature matching according to claim 8 is characterized in that: The obtaining of question information sent by the user includes: The question information is subjected to text processing, wherein the text processing includes text segmentation.

10. A semantic recognition system based on dual channel feature matching, characterized in that: include: A receiving module, used for receiving question information sent by a user and characteristics of a sending channel of the question information; A natural language processing module, used to obtain the user's intention based on the question information; A generation module obtains corresponding standard questions according to the intent; The output module is used to determine whether there is a semantic channel feature in the question information. If so, the answer information matching the semantic channel feature under the standard question is sent to the user; otherwise, the answer information matching the sending channel feature under the standard question is sent to the user, with the semantic channel feature being given priority; wherein the channel includes one of a PC channel, a mobile phone channel, and a tablet channel.

11. The semantic recognition system based on dual channel feature matching according to claim 10, characterized in that: Also includes: The text processing module is used to perform text processing on the question information, wherein the text processing method includes text segmentation.

12. An electronic device comprising: processor; A memory storing a computer executable program, which, when executed by the processor, enables the processor to execute the semantic recognition method based on dual-channel feature matching as described in any one of claims 1 to 9.

13. A computer readable medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the semantic recognition method based on dual-channel feature matching as described in any one of claims 1 to 9 is implemented.

Images