Information identification method and device

Abstract

The present application discloses an information processing method and device. After receiving the information to be processed, first, the information to be processed is identified with intent to obtain the target intent of the information to be processed; secondly, according to the target intent, the information to be processed is extracted for the first time to obtain the first slot set included in the information to be processed; and, based on the target intent and the first slot set, the information to be processed is extracted for the second time to obtain the second slot set included in the information to be processed; finally, according to the first slot set and the second slot set, the information identification result of the information to be processed is determined. In this way, the second slot information extraction process is added, and the relationship, subject information and structural information of each slot are fully utilized to effectively recall the slots missed in the first slot information extraction process, and effectively increase the accuracy and efficiency of slot information extraction. Thereby, the communication rounds in the information interaction process are effectively reduced, and the user experience is significantly improved.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to an information recognition method and device. Background Art

[0002] With the development of AI technology, chatbot (dialogue system) has been widely used, and higher requirements have been put forward for the system performance of chatbot. The recognition rate and communication rounds of chatbot slots are important performance indicators for evaluating chatbots. Only by obtaining more information in fewer communication rounds can the user experience be effectively improved. The complete NLU (Nature Language Understanding) of the dialogue system includes domain intent recognition and slot extraction. Domain intent recognition is used to tell the machine the content of the user's query, such as querying the weather, and slot extraction is used to tell the machine the detailed information of the slots that need to be filled for the recognized intent, such as the time to query the weather, the location to query the weather, the progress of the audio and video content, and the date of the audio and video playback. In the process of slot extraction, due to the limitations of data, models and other factors, it is easy to cause the omission of slots, resulting in the lack of slot information that needs to be filled.

[0003] Currently, the main method to solve this problem is to start the next round of dialogue. The machine actively asks the user for slot information that has not been recognized. For example, the user said, "I may go on a business trip to another place in two days, and then I will probably stay in Silicon Valley for a few days. Do you have any recommendations on what to wear?" The chatbot only recognizes that the domain intent is to query the weather. The slots that need to be filled in the weather query include time and location. The location is Beijing, but the time is omitted. At this time, the system will initiate a second round of dialogue and ask the user which day to query. The increase in the number of communication rounds in this solution will delay the processing time of user questions, affect the performance of the chatbot, and seriously reduce the user experience.

[0004] Application Contents

[0005] The present application provides an information processing method and device.

[0006] According to the first aspect of the present application, there is provided an information processing method, the method comprising: receiving information to be processed; performing intent recognition on the information to be processed to obtain a target intent of the information to be processed; performing a first slot information extraction on the information to be processed according to the target intent to obtain a first slot set included in the information to be processed; performing a second slot information extraction on the information to be processed based on the target intent and the first slot set to obtain a second slot set included in the information to be processed; and determining an information recognition result of the information to be processed according to the first slot set and the second slot set.

[0007] According to another embodiment of the present application, based on the target intent information and the first slot set, a second slot information extraction is performed on the information to be processed to obtain a second slot set included in the information to be processed, including: according to the first slot set, feature extraction is performed on the information to be processed to obtain an input feature vector for the second slot feature extraction; slot information extraction is performed on the input feature vector to obtain the second slot set.

[0008] According to another embodiment of the present application, the input feature vector includes word vectors, position vectors and slot label vectors of the words included in the information to be processed; accordingly, the information to be processed is subjected to feature extraction based on the first slot set to obtain an input feature vector for a second slot feature extraction, including: determining the slot label vectors of the words included in the information to be processed based on the first slot label information included in the first slot set; determining the word vectors and position vectors of the words in the information to be processed; wherein the position vector is used to show the position of each word in the information to be processed in the information to be processed.

[0009] According to another embodiment of the present application, slot information is extracted from the input feature vector to obtain a second slot set included in the information to be processed, including: splicing the input feature vector to obtain a splicing vector; performing dimensionality reduction processing on the splicing vector to obtain an encoding vector; performing semantic analysis on the information to be processed to obtain a semantic analysis result; determining the semantic vectors of the characters of the information to be processed based on the semantic analysis result; determining the intention vectors of the characters of the information to be processed based on the target intention; and normalizing the encoding vector, the semantic vector and the intention vector to obtain the second slot set included in the information to be processed.

[0010] According to another embodiment of the present application, performing semantic analysis on the information to be processed to obtain a semantic analysis result includes: performing at least one of syntactic construction and dependency analysis on the information to be processed to obtain the semantic analysis result.

[0011] According to another embodiment of the present application, before performing a second slot information extraction on the information to be processed based on the target intent and the first slot set, the method also includes: obtaining a number of training corpora, the training corpora including sentence content, intent information of the sentence content, and slot label information; training a first slot information extraction model based on the sentence content, the intent information, and the slot label information to perform a first slot information extraction on the information to be processed.

[0012] According to another embodiment of the present application, the method also includes: fuzzifying at least part of the slot labels of the slot label information to obtain a slot labeling result; training a second slot information extraction model based on the sentence content, the intention labeling result and the slot labeling result to perform a second slot information extraction on the information to be processed.

[0013] According to another embodiment of the present application, at least part of the slot labels of the slot label information are fuzzy processed to obtain the slot labeling result, including: using a random sampling method to mask at least part of the labels in the slot label information to obtain the slot labeling result.

[0014] According to another embodiment of the present application, determining the information identification result of the information to be processed based on the first slot set and the second slot set includes: merging the first slot set and the slot set to obtain a slot label set and a slot vocabulary set; deduplicating the slot label set and the slot vocabulary set to obtain the information identification result.

[0015] According to the second aspect of the present application, an information identification device is also disclosed, which includes: a receiving module for receiving information to be processed; an intention identification module for performing intention identification on the information to be processed to obtain the target intention of the information to be processed; a first extraction module for performing a first slot information extraction on the information to be processed according to the target intention to obtain a first slot set included in the information to be processed; a second extraction module for performing a second slot information extraction on the information to be processed based on the target intention and the first slot set to obtain a second slot set included in the information to be processed; an information processing module for determining the information identification result of the information to be processed based on the first slot set and the second slot set.

[0016] The information processing method and device of the embodiment of the present application, after receiving the information to be processed, firstly, perform intent recognition on the information to be processed to obtain the target intent of the information to be processed; secondly, perform a first slot information extraction on the information to be processed according to the target intent to obtain the first slot set included in the information to be processed; and, based on the target intent and the first slot set, perform a second slot information extraction on the information to be processed to obtain the second slot set included in the information to be processed; finally, determine the information recognition result of the information to be processed according to the first slot set and the second slot set. In this way, the second slot information extraction process is added, and the relationship, subject information and structural information of each slot are fully utilized to effectively recall the slots missed in the first slot information extraction process, and effectively increase the accuracy and efficiency of slot information extraction. Thereby, the communication rounds in the information interaction process are effectively reduced, and the user experience is significantly improved.

[0017] It should be understood that the teachings of the present application are not required to achieve all of the beneficial effects described above, but specific technical solutions can achieve specific technical effects, and other embodiments of the present application can also achieve beneficial effects not mentioned above. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] By reading the detailed description below with reference to the accompanying drawings, the above and other purposes, features and advantages of the exemplary embodiments of the present application will become readily understood. In the accompanying drawings, several embodiments of the present application are shown in an exemplary and non-limiting manner, wherein:

[0019] In the drawings, the same or corresponding reference numerals represent the same or corresponding parts.

[0020] Figure 1 A schematic diagram of the implementation process of the information identification method according to an embodiment of the present application is shown;

[0021] Figure 2 A schematic diagram of the implementation process of an information identification method according to another embodiment of the present application is shown;

[0022] Figure 3 A schematic diagram of the model structure of a second slot information extraction model of an information identification method according to an embodiment of the present application is shown;

[0023] Figure 4 A schematic diagram of the composition structure of the information identification device according to an embodiment of the present application is shown. DETAILED DESCRIPTION

[0024] The principles and spirit of the present application will be described below with reference to several exemplary embodiments. It should be understood that these embodiments are provided only to enable those skilled in the art to better understand and implement the present application, and are not intended to limit the scope of the present application in any way. On the contrary, these embodiments are provided to make the present application more thorough and complete, and to fully convey the scope of the present application to those skilled in the art.

[0025] First, the application scenario of the present application is briefly described. The information recognition method in the embodiment of the present application is mainly used in the process of information interaction, for example: in the process of interaction between smart speakers, smart customer service and users, information sent by users is received and the received information is responded to. The information sent by the user can be text information or audio information, and the information is recognized to obtain the recognition result, so as to respond to the user based on the recognition result. Of course, the above is only a simple example of the application scenario of the technical solution of the present application, and the technical solution of the present application can also be applied to other applicable scenarios.

[0026] The technical solution of the present application is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0027] Figure 1 A schematic diagram of the implementation flow of an information identification method according to an embodiment of the present application is shown.

[0028] refer to Figure 1 The information processing method of the embodiment of the present application at least includes the following operation flow: operation 101, receiving information to be processed; operation 102, performing intent recognition on the information to be processed to obtain the target intent of the information to be processed; operation 103, performing a first slot information extraction on the information to be processed according to the target intent, and obtaining a first slot set included in the information to be processed; operation 104, performing a second slot information extraction on the information to be processed based on the target intent and the first slot set, and obtaining a second slot set included in the information to be processed; operation 105, determining the information recognition result of the information to be processed according to the first slot set and the second slot set.

[0029] In operation 101, information to be processed is received.

[0030] In this embodiment of the present application, the information to be processed may be an audio signal or text content. If the information to be processed is an audio signal, the audio signal needs to be converted into text content first.

[0031] In operation 102, the information to be processed is subjected to intent recognition to obtain the target intent of the information to be processed.

[0032] In this embodiment of the present application, the intent recognition method commonly used in the NLP (Nature Language Process) process can be used to realize the intent recognition of the information to be processed. For example, the text content of the information to be processed is "I may go on a business trip to another place in two days, um... and then, I estimate that I will stay in Shenzhen for a few days. Do you have any recommendations on what to wear?", and the intent recognition of the information to be processed is obtained, and the target intent of the information to be processed is "query the weather".

[0033] In operation 103, a first slot information extraction is performed on the information to be processed according to the target intent to obtain a first slot set included in the information to be processed.

[0034] In this embodiment of the present application, the first slot information extraction is performed on the information to be processed, and the first slot set obtained may include the slot label of the slot that needs to be filled for the target intent, and may also include the slot vocabulary corresponding to the slot label included in the information to be processed. For example: the text content of the information to be processed is "I may go on a business trip to another place in two days, um... and then, I estimate that I will stay in Shenzhen for a few days. Do you have any recommendations on what to wear?" The first slot set may include the slot label "City" that needs to be filled for the target intent "Query the weather" of the information to be processed, and may also include the slot vocabulary "Shenzhen" corresponding to the slot label "City".

[0035] In this embodiment of the present application, the identification of slot information is essentially an entity recognition process in NLP, and a general entity recognition method can be used to realize the identification of slot information in the information to be processed. For example, a large amount of corpus can be obtained, the entities included in the obtained corpus can be annotated, and a slot information extraction model can be trained using a neural network algorithm. Then, the trained model is used to extract the slot information in the information to be processed.

[0036] In operation 104 , based on the target intent and the first slot set, a second slot information extraction is performed on the information to be processed to obtain a second slot set included in the information to be processed.

[0037] In this embodiment of the present application, due to the first slot information extraction process, there is a tendency for slot information to be missed. For example, for the information to be processed "I may go on a business trip to another place in two days, um... and then, I will probably stay in Shenzhen for a few days, what to wear, do you have any recommendations?", in the first slot extraction process, only the slot label "City" and the corresponding slot vocabulary "Shenzhen" are extracted, but the slot label "Date" contained in the information to be processed and the slot vocabulary "two days later" corresponding to the slot label are missed.

[0038] In this embodiment of the present application, the same model structure as that used in the first slot extraction process can be used to train a recall layer model for extracting slot information again when some slot information is missing. Then, based on the result of the first slot extraction, the recall layer model is used to perform a second slot extraction.

[0039] In operation 105 , an information identification result of the information to be processed is determined according to the first slot set and the second slot set.

[0040] In this embodiment of the present application, the following operating steps are used to determine the information identification result of the information to be processed based on the first slot set and the second slot set: the first slot set and the slot set are merged to obtain a slot label set and a slot vocabulary set, and then the slot label set and the slot vocabulary set are deduplicated to obtain the information identification result.

[0041] For example, the information to be processed is "I may have to go on a business trip to another place in two days. Um... and then, I will probably stay in Shenzhen for a few days. Do you have any recommendations on what to wear?" The first slot set includes {[City, Shenzhen]}, and the second slot set includes {[Date, two days later], [City, Shenzhen]}, then the information recognition result is {[Date, two days later], [City, Shenzhen]}.

[0042] Figure 2 A schematic diagram of the implementation flow of an information identification method according to another embodiment of the present application is shown.

[0043] refer to Figure 2 Another embodiment of the information identification method of the present application embodiment at least includes the following operation process:

[0044] Operation 201: receiving information to be processed.

[0045] In this embodiment of the present application, the operation steps of this embodiment of the present application are explained by taking the information to be processed as "I may have to go on a business trip to another place in two days, and then I will probably stay in place A for a few days. Do you have any recommendations on what to wear?"

[0046] Operation 202 : performing intent recognition on the information to be processed to obtain the target intent of the information to be processed.

[0047] In this embodiment of the present application, refer to the above Figure 1 In operation 102, it can be obtained that the target intent of the information to be processed is "query weather".

[0048] Operation 203 : extracting slot information of the information to be processed for the first time according to the target intent, and obtaining a first slot set included in the information to be processed.

[0049] In this embodiment of the present application, in the process of performing the first slot information extraction on the information to be processed and obtaining the first slot set included in the information to be processed, there may be omissions of slot information. Therefore, the slot information is extracted for a second time in the present application. In addition, before performing the second slot information extraction on the information to be processed based on the target intent and the first slot set, a number of training corpora are obtained, and the training corpora include sentence content, intent information of the sentence content, and slot label information; based on the sentence content, intent information, and slot label information, the first slot information extraction model is trained to perform the first slot information extraction on the information to be processed.

[0050] In another embodiment of the present application, at least part of the slot labels of the slot label information are fuzzy processed to obtain slot labeling results; based on the sentence content, intention labeling results and slot labeling results, a second slot information extraction model is trained to perform a second slot information extraction on the information to be processed. Figure 3 A schematic diagram of the model structure of the second slot information extraction model of the information identification method of this embodiment of the present application is shown.

[0051] In this embodiment of the present application, operations 204 to 209 are implemented using the second slot information extraction model. Figure 3 FIG. 2 is a schematic diagram showing the structure of the second slot information extraction model of the information identification method of this embodiment of the present application. Therefore, the specific operations of operations 204 to 209 will be Figure 3 The model structure is described in detail in . We will not go into details here.

[0052] Operation 204 : performing feature extraction on the information to be processed according to the first slot set to obtain an input feature vector for a second slot feature extraction.

[0053] In this embodiment of the present application, the input feature vector includes the word vector, position vector and slot label vector of the word included in the information to be processed. Accordingly, the following operation can be used to extract the feature of the information to be processed according to the first slot set to obtain the input feature vector of the second slot feature extraction: according to the first slot label information included in the first slot set, the slot label vector of the word included in the information to be processed is determined, and the word vector and position vector of the word of the information to be processed are determined. Among them, the position vector is used to show the position of each word in the information to be processed in the information to be processed.

[0054] Operation 205 , concatenate the input feature vectors to obtain a concatenated vector; perform dimensionality reduction processing on the concatenated vector to obtain a coding vector.

[0055] Operation 206 , performing semantic analysis on the information to be processed to obtain a semantic analysis result.

[0056] In this embodiment of the present application, semantic analysis of the information to be processed includes: performing at least one of syntactic construction and dependency analysis on the information to be processed. For example, "I am going to Hangzhou on a business trip the day after tomorrow. What kind of clothes are suitable there?" can be determined by syntactic construction that "I" is the subject and "go" is the predicate.

[0057] Operation 207 : determining the semantic vector of the word in the information to be processed according to the semantic analysis result.

[0058] In operation 208, the intention vector of the word of the information to be processed is determined according to the target intention.

[0059] Operation 209 , normalize the encoding vector, the semantic vector, and the intention vector to obtain a second slot set included in the information to be processed.

[0060] Operation 210: determining information identification results of the information to be processed according to the first slot set and the second slot set.

[0061] The specific implementation process of operations 201, 202, 203 and 210 is the same as Figure 1 The specific implementation processes of operations 101, 102, 103 and 104 in the illustrated embodiment are similar and will not be described in detail here.

[0062] Figure 3 A schematic diagram of the model structure of the second slot information extraction model of the information identification method of this embodiment of the present application is shown.

[0063] refer to Figure 3 , briefly explain the model training and application process of the second slot information extraction model. Figure 3 As shown, during the model training process, several training corpora are first obtained. The training corpora include sentence content, intent information of sentence content and slot label information. For example: (1) Sentence content: I want to listen to song B by singer A; intent information of sentence content: play music; slot label information: slot vocabulary "singer A" corresponds to slot label "Singer", and slot vocabulary "song B" corresponds to slot label "Song". (2) Sentence content: I am going to Hangzhou on a business trip the day after tomorrow; intent information of sentence content: check the weather; slot label information: slot vocabulary "the day after tomorrow" corresponds to slot label "Date", slot vocabulary "Hangzhou" corresponds to slot label "City", and slot vocabulary "clothes" corresponds to slot label "DressingIndex".

[0064] In this embodiment of the present application, at least part of the slot labels of the slot label information are fuzzified to obtain the slot labeling result. Here, a random sampling method can be adopted to mask at least part of the labels in the slot label information to obtain the slot labeling result. For example: after fuzzifying at least part of the slot labels of the slot label information, the slot label "Song" corresponding to the slot vocabulary "Song" in sentence (1) is masked, and the slot label "City" corresponding to the slot vocabulary "Hangzhou" in sentence (2) is masked. After the fuzzification process, the training corpus is feature extracted, and the slot label vectors of each word in sentence (1) are obtained as "E0, E0, E0, E0" respectively. Singer 、E Singer 、E Singer , E0, E0, E0, E0", the position vectors are "E0, E1, E2, E3, E4, E5, E6, E7, E8, E9", and the word vectors are "E 我 、E 想 、E 听 、E 歌 、E 手 、E A 、E 的 、E 歌 、E 曲 、E B ". Similarly, the slot label vector, position vector and word vector of each word in sentence (2) can be obtained. In this way, after feature extraction is performed on the acquired training corpus, the input feature vectors of all training corpora are obtained, and weighted splicing is performed on the input feature vectors of each training corpus to obtain a spliced ​​vector. Here, the input feature vector is spliced. In the process of weighted splicing of the slot label vector, position vector and word vector of the training corpus, the initial weight value can be set on average, for example: the word vector, position vector and slot label vector each account for 1 / 3. As the number of training corpora increases, the weight value of the weighted splicing tends to be stable and serves as the final model parameter.

[0065] Furthermore, the concatenated vector is input into the Transformer encoder (decoder) for dimensionality reduction processing to obtain the dimensionality reduction processing result. The dependency relationship of the training corpus is introduced to construct the syntactic relationship of the training corpus. Here, the dependency relationship and the construction of the syntactic relationship can be realized by the semantic analysis method in general NLP. The dependency relationship of the training corpus, the constructed syntactic relationship of the training corpus and the dimensionality reduction processing result are normalized. Here, the normalization processing can use the Softmax function or other applicable normalization processing functions to finally obtain the probability value of the word in the training corpus being a slot vocabulary, and determine that the probability value is greater than the set threshold as a slot vocabulary. After multiple training processes, until the model recognizes that the slot labels and slot vocabulary masked during the fuzzy processing of the training corpus are greater than the set recognition rate. It means that the accuracy of the model meets expectations.

[0066] Taking the information to be processed as "I may have to go on a business trip to another place in two days, and then I will probably stay in place A for a few days. Do you have any recommendations on what to wear?", the following explains the process of performing a second slot information extraction on the information to be processed using the trained second slot information extraction model.

[0067] In this embodiment of the present application, in the first slot extraction result, only the slot label of "A place" is identified as "City". Therefore, according to the first slot extraction result, feature extraction is performed on the information to be processed, and the slot label vector of the information to be processed is determined to be "E0, E0, E0, E0, E0, E0 … 、E City 、E City 、E … ”, the position vectors are “E0, E1, E2, E3, E4, E … 、E 22 、E 23 、E … ”, the word vectors are “E 我 、E 可 、E 能 、E 过 、E 两 、E …… 、E A 、E 地 、E … ", the slot label vector, position vector and word vector are weightedly concatenated and then input into the Transformer encoder.

[0068] Furthermore, a semantic analysis is performed on the information to be processed, specifically including constructing syntactic operations on the information to be processed and determining the dependencies of the information to be processed. The results of the semantic analysis and the output of the Transformer encoder are normalized to obtain the probability that the characters in the information to be processed are slot words, and the corresponding slot labels are determined. For example: the final probability that "two days later" is a slot word is 78%, which is greater than the set threshold of 70%. It is thus determined that "two days later" is a slot word, and "two days later" is time information, and the slot label is "Date". As a result, the slot label "Date" and the corresponding slot vocabulary "two days later" that were missed during the first slot extraction process are successfully recalled.

[0069] The information processing method and device of the embodiment of the present application, after receiving the information to be processed, firstly, perform intent recognition on the information to be processed to obtain the target intent of the information to be processed; secondly, perform a first slot information extraction on the information to be processed based on the target intent to obtain the first slot set included in the information to be processed; and, based on the target intent and the first slot set, perform a second slot information extraction on the information to be processed to obtain the second slot set included in the information to be processed; finally, determine the information recognition result of the information to be processed based on the first slot set and the second slot set. Therefore, the design of the second slot information extraction process can make full use of the relationship between each slot, thematic information and structural information, etc., effectively recall the slots missed in the first slot information extraction process, and increase the accuracy and efficiency of slot information extraction. Thereby, the communication rounds in the information interaction process are effectively reduced, and the user experience is significantly improved.

[0070] Similarly, based on the above information processing method, an embodiment of the present application also provides a computer-readable storage medium, which stores a program. When the program is executed by a processor, the processor performs at least the following operation steps: operation 101, receiving information to be processed; operation 102, performing intent recognition on the information to be processed to obtain the target intent of the information to be processed; operation 103, performing a first slot information extraction on the information to be processed according to the target intent, and obtaining a first slot set included in the information to be processed; operation 104, performing a second slot information extraction on the information to be processed based on the target intent and the first slot set, and obtaining a second slot set included in the information to be processed; operation 105, determining the information recognition result of the information to be processed according to the first slot set and the second slot set.

[0071] Further, based on the above information method, the present application embodiment also provides an information identification device, such as Figure 4The device 40 includes: a receiving module 401, which is used to receive information to be processed; an intention recognition module 402, which is used to perform intention recognition on the information to be processed and obtain the target intention of the information to be processed; a first extraction module 403, which is used to perform a first slot information extraction on the information to be processed according to the target intention and obtain a first slot set included in the information to be processed; a second extraction module 404, which is used to perform a second slot information extraction on the information to be processed based on the target intention and the first slot set and obtain a second slot set included in the information to be processed; an information processing module 405, which is used to determine the information recognition result of the information to be processed according to the first slot set and the second slot set.

[0072] According to another embodiment of the present application, the second extraction module 404 includes: a vector extraction sub-module, used to perform feature extraction on the information to be processed according to the first slot set to obtain an input feature vector for the second slot feature extraction; a slot extraction sub-module, used to perform slot information extraction on the input feature vector to obtain a second slot set.

[0073] According to another embodiment of the present application, the input feature vector includes a word vector, a position vector and a slot label vector of the words included in the information to be processed; accordingly, the vector extraction submodule performs feature extraction on the information to be processed according to the first slot set to obtain an input feature vector for the second slot feature extraction, including: determining the slot label vector of the words included in the information to be processed according to the first slot label information included in the first slot set; determining the word vector and the position vector of the words in the information to be processed; wherein the position vector is used to show the position of each word in the information to be processed in the information to be processed.

[0074] According to another embodiment of the present application, the slot extraction submodule extracts slot information from the input feature vector to obtain a second slot set included in the information to be processed, including: splicing the input feature vector to obtain a splicing vector; performing dimensionality reduction processing on the splicing vector to obtain a coding vector; performing semantic analysis on the information to be processed to obtain a semantic analysis result; determining the semantic vectors of the characters in the information to be processed based on the semantic analysis result; determining the intention vectors of the characters in the information to be processed based on the target intention; normalizing the coding vector, the semantic vector and the intention vector to obtain the second slot set included in the information to be processed.

[0075] According to another embodiment of the present application, semantic analysis is performed on the information to be processed to obtain a semantic analysis result, including: performing at least one of syntactic construction and dependency analysis on the information to be processed to obtain a semantic analysis result.

[0076] According to another embodiment of the present application, the device 40 also includes: an acquisition module, which is used to acquire a number of training corpora before performing a second slot information extraction on the information to be processed based on the target intent and the first slot set, the training corpora including sentence content, intent information of the sentence content, and slot label information; a first training module, which is used to train a first slot information extraction model based on the sentence content, intent information and slot label information to perform a first slot information extraction on the information to be processed.

[0077] According to another embodiment of the present application, the device 40 also includes: a fuzzy processing module, which is used to fuzzy at least part of the slot labels of the slot label information to obtain a slot labeling result; a second training module, which is used to train a second slot information extraction model based on the sentence content, intention labeling results and slot labeling results, so as to perform a second slot information extraction on the information to be processed.

[0078] According to another embodiment of the present application, the fuzzy processing module includes: a masking submodule, which is used to mask at least part of the labels in the slot label information by random sampling to obtain a slot marking result.

[0079] According to another embodiment of the present application, the information processing module 405 includes: a slot merging sub-module, used to merge the first slot set and the slot set to obtain a slot label set and a slot vocabulary set; a slot deduplication sub-module, used to deduplicate the slot label set and the slot vocabulary set to obtain an information recognition result.

[0080] It should be noted that: the above description of the information processing device embodiment is different from the above description of the information processing device embodiment. Figures 1 to 3 The description of the method embodiment shown is similar, with the same Figures 1 to 3 For technical details not disclosed in the information processing device embodiments of this application, please refer to the aforementioned Figures 1 to 3 The description of the method embodiment shown in the figure can be understood, and in order to save space, it will not be repeated.

[0081] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or device including the element.

[0082] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are only schematic. For example, the division of units is only a logical function division. There may be other division methods in actual implementation, such as: multiple units or components can be combined, or can be integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the components shown or discussed can be through some interfaces, and the indirect coupling or communication connection of devices or units can be electrical, mechanical or other forms.

[0083] The units described above as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units; they may be located in one place or distributed on multiple network units; some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0084] In addition, all functional units in the embodiments of the present application may be integrated into one processing unit, or each unit may be a separate unit, or two or more units may be integrated into one unit; the above-mentioned integrated units may be implemented in the form of hardware or in the form of hardware plus software functional units.

[0085] A person skilled in the art can understand that all or part of the steps of implementing the above method embodiment can be completed by hardware related to program instructions, and the aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps of the above method embodiment; and the aforementioned storage medium includes: mobile storage devices, read-only memories (ROM), magnetic disks or optical disks, etc., various media that can store program codes.

[0086] Alternatively, if the above-mentioned integrated unit of the present application is implemented in the form of a software function module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiment of the present application can essentially or in other words, the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a storage medium and includes a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the methods of each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as mobile storage devices, ROMs, magnetic disks, or optical disks.

[0087] The above are only specific implementations of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. An information processing method, the method comprising: Receive information to be processed; Performing intent recognition on the information to be processed to obtain the target intent of the information to be processed; According to the target intent, extracting the first slot information of the information to be processed to obtain a first slot set included in the information to be processed; Based on the target intent and the first slot set, using a second slot information extraction model, performing a second slot information extraction on the information to be processed to obtain a second slot set included in the information to be processed, wherein the second slot information extraction model is used to extract slot information again when part of the slot information is missing; An information identification result of the information to be processed is determined according to the first slot set and the second slot set.

2. According to the method of claim 1, based on the target intent and the first slot set, performing a second slot information extraction on the information to be processed to obtain a second slot set included in the information to be processed, comprising: According to the first slot set, feature extraction is performed on the information to be processed to obtain an input feature vector for a second slot feature extraction; Slot information is extracted from the input feature vector to obtain the second slot set.

3. The method according to claim 2, wherein the input feature vector comprises a word vector, a position vector and a slot label vector of the word included in the information to be processed; Correspondingly, the feature extraction of the information to be processed is performed according to the first slot set to obtain an input feature vector for the second slot feature extraction, including: Determine, according to the first slot label information included in the first slot set, a slot label vector of the word included in the information to be processed; Determine a word vector and a position vector of a word of the information to be processed; The position vector is used to show the position of each word in the information to be processed.

4. The method according to claim 3, extracting slot information from the input feature vector to obtain a second slot set included in the information to be processed, comprising: Concatenating the slot label vector, the position vector and the word vector to obtain a concatenated vector; Performing dimensionality reduction processing on the concatenated vector to obtain a coding vector; Performing semantic analysis on the information to be processed to obtain a semantic analysis result; According to the semantic analysis result, the semantic vector of the word of the information to be processed is determined Determining the intention vector of the word of the information to be processed according to the target intention; The encoding vector, the semantic vector and the intention vector are normalized to obtain a second slot set included in the information to be processed.

5. According to the method of claim 4, the step of performing semantic analysis on the information to be processed to obtain a semantic analysis result comprises: Perform at least one of syntactic construction and dependency analysis on the information to be processed to obtain the semantic analysis result.

6. The method according to claim 1, before extracting slot information for the second time from the information to be processed using a second slot information extraction model based on the target intent and the first slot set, the method further comprises: Acquire a plurality of training corpora, wherein the training corpora include sentence content, intent information of the sentence content, and slot label information; According to the sentence content, the intent information and the slot label information, a first slot information extraction model is trained to perform a first slot information extraction on the information to be processed.

7. The method according to claim 6, further comprising: Performing fuzzy processing on at least part of the slot labels of the slot label information to obtain a slot labeling result; The second slot information extraction model is trained according to the sentence content, the intent information and the slot labeling result to perform a second slot information extraction on the information to be processed.

8. According to the method of claim 7, fuzzifying at least part of the slot labels of the slot label information to obtain a slot labeling result comprises: At least part of the labels in the slot label information are masked by random sampling to obtain the slot labeling result.

9. The method according to any one of claims 1 to 8, wherein determining the information identification result of the information to be processed according to the first slot set and the second slot set comprises: Merging the first slot set and the second slot set to obtain a slot label set and a slot vocabulary set; The slot label set and the slot vocabulary set are deduplicated to obtain the information recognition result.

10. An information identification device, comprising: A receiving module, used for receiving information to be processed; An intention recognition module, used to perform intention recognition on the information to be processed to obtain the target intention of the information to be processed; A first extraction module, configured to extract slot information of the information to be processed for the first time according to the target intent, to obtain a first slot set included in the information to be processed; A second extraction module is used to perform a second slot information extraction on the information to be processed based on the target intent and the first slot set using a second slot information extraction model to obtain a second slot set included in the information to be processed, wherein the second slot information extraction model is used to extract slot information again when some slot information is missing; An information processing module is used to determine the information identification result of the information to be processed according to the first slot set and the second slot set.

Images