Identification method, identification system, electronic device and storage medium

By preprocessing and part-of-speech annotation of the input information, combining sliding window matching and pinyin matching, and using organizational structure tree sorting, the problem of low accuracy of person name recognition in human-computer dialogue is solved, and the recognition accuracy and recall rate are improved.

CN113919329BActive Publication Date: 2024-11-22YONYOU NETWORK TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202111130756.4
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2021-09-26
Publication Date
2024-11-22
Estimated Expiration
2041-09-26

AI Technical Summary

Technical Problem

In human-computer dialogue, the traditional human-name matching method has low recognition accuracy and recall rates due to the error of voice to text, especially in the recognition of human-names, which have many false alarms and missed reports.

Method used

By preprocessing the input information, converting it into multiple meta words, and matching it with the sliding window with part-of-speech annotation, only meta words with predetermined part-of-speech avoid homophones, and adding pinyin matching to improve the Recall rate, use organizational structure tree sorting to recommend the best matching results.

Benefits of technology

It significantly improves the accuracy and recall rate of person name recognition, reduces false alarms and missed reports, and enhances the accuracy of contact recognition in human-computer dialogue.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN113919329B_ABST
    Figure CN113919329B_ABST
Patent Text Reader

Abstract

The present invention proposes a recognition method, a recognition system, an electronic device and a storage medium. The recognition method includes: in response to a user's request operation, preprocessing the input information to obtain multiple meta-words; determining a meta-word with a predetermined part of speech among the multiple meta-words; matching the meta-word with a preset information table to obtain a recognition result. The recognition method provided by the present invention performs sliding window matching based on the meta-word and the information in the preset information table, which can avoid the sliding window from creating some so-called new words for matching, thereby further improving the recognition accuracy. For each meta-word, combined with its part of speech, only the meta-word with a predetermined part of speech will be matched with the information in the preset information table, so that many false alarms based on homophones can be filtered out, thereby preventing an excessively high false alarm rate.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of human-computer dialogue, and in particular to an identification method, an identification system, an electronic device and a storage medium. Background Art

[0002] With the continuous advancement of speech recognition technology and natural language processing technology, various human-computer dialogue VPAs (Virtual Personal Assistants) are widely used in smart devices to enhance human-computer interaction functions. However, in the application of human-computer dialogue robots, due to the particularity of human names, there will be many errors in the speech-to-text conversion process, especially human names, which are basically impossible to convert correctly. Therefore, the traditional name matching method is not effective in this case, with low accuracy and recall rates. Summary of the invention

[0003] The present invention aims to solve at least one of the technical problems existing in the prior art.

[0004] To this end, a first aspect of the present invention provides an identification method.

[0005] A second aspect of the present invention provides an identification system.

[0006] A third aspect of the present invention provides an identification system.

[0007] A fourth aspect of the present invention provides an electronic device.

[0008] A fifth aspect of the present invention provides a storage medium.

[0009] In view of this, according to the first aspect of the present invention, a recognition method is proposed, which is used for an electronic device. The recognition method includes: in response to a user's request operation, preprocessing the input information to obtain multiple meta-words; determining meta-words with predetermined parts of speech among the multiple meta-words; matching the meta-words with a preset information table to obtain a recognition result.

[0010] The recognition method provided by the present invention responds to the user's request operation, pre-processes the input information, and obtains multiple meta-words. A sliding window is matched based on the meta-words and the information in the preset information table. For each meta-word, in combination with its part of speech, only the meta-word with the predetermined part of speech will be matched with the information in the preset information table, so that many false alarms caused by homophony can be filtered out, thereby preventing an excessively high false alarm rate.

[0011] For example, for input information such as "We have a regular meeting every Wednesday afternoon", after preprocessing, the meta-word set is {every Wednesday afternoon, we, have, to, hold, regular meeting}. If the part of speech is not considered, then the word "regular meeting" will match the contact "A" with the same pronunciation in the address book, which is obviously not what we want. Therefore, we only consider those meta-words with the part of speech marked as the name PER and match them with the contacts in the address book to prevent too high a false positive rate.

[0012] It should be noted here that the matching method directly based on the sliding window is prone to many false positives. The sliding window may cover words from different words, forming meaningless words, and then it can match the names of people in the address book, thereby forming false positives. For example, in the input information "there is a meeting at 3 pm", "to be held" is not a meaningful unit, but based on this word, it can match "B" with the same pronunciation as the name in the address book, thereby causing false positives. The present invention first pre-processes the input information, and the pre-processed input information is converted into multiple meta-words, each of which is a meaningful minimum semantic unit. On this basis, the matching method of the sliding window is adopted, which can avoid the sliding window from creating some so-called new words for matching, thereby further improving the accuracy of recognition. Among them, the input information can be voice information or text information. The preset information table is determined according to the recognition requirements. When it is necessary to identify the contact, the preset information table can be a list of names. Further, the list of names can be a list of Chinese characters and / or a list of pinyin. Of course, the preset information table is not limited to a list of names.

[0013] The identification method is used in electronic devices, and specifically can be used in various scenarios where contact identification is required in mobile and PC dialogue robots.

[0014] The above identification method according to the present invention may also have the following technical features:

[0015] In the above technical solution, the step of matching the meta-words with the preset information table specifically includes: when the meta-words are Chinese character keywords, matching the meta-words with the first preset information table; and / or converting the meta-words into pinyin keywords, matching the pinyin keywords with the second preset information table.

[0016] In this technical solution, since the input information may be voice information, text conversion is required during recognition. The converted text may only have the same pronunciation, and the Chinese characters may be very different. Matching only Chinese characters will result in a large number of missed reports. In order to improve the recall rate, pinyin matching can be added on the basis of Chinese character matching. Specifically, when the meta-word is a Chinese character keyword, it is matched with the first preset information table. Further, the meta-word in Chinese character form is converted into a pinyin keyword, and the pinyin keyword is matched with the second preset information table. This overcomes errors in the speech-to-text conversion process, especially missed reports caused by name conversion errors.

[0017] In any of the above technical solutions, the step of matching the meta-words with the preset information table to obtain the recognition result specifically includes: matching the meta-words with the preset information table to obtain at least two matching results; sorting the at least two matching results according to the organizational structure tree of the user's organization, and determining the recognition result according to the sorting order.

[0018] In this technical solution, for the same meta-word, due to the use of the pinyin matching method, two or more homophonic matching results may be obtained, or by using the Chinese matching method, at least two candidate words with the same name may be obtained. At this time, the two or more matching results are ranked and recommended according to the organizational tree of the organization where the user is located, so that the user can select the best possible matching result.

[0019] In any of the above technical solutions, at least two matching results are sorted according to the organizational structure tree, and the step of determining the recognition result according to the sorting order specifically includes: determining the tree path between any one of the at least two matching results and the user on the organizational structure tree, and determining the distance between any one of the matching results and the user based on the tree path, to obtain at least two distances corresponding to the at least two matching results; and sorting the at least two distances to determine the recognition result.

[0020] In this technical solution, by determining the tree path between any of the above matching results and the user on the organizational tree, and determining the distance between any of the matching results and the user according to the tree path, the distances corresponding to the above matching results are obtained. By sorting these distances, the above matching results can be sorted, and then the above matching results can be sorted and recommended.

[0021] In actual applications, the information of the dialog user, such as the department, etc., can be obtained, and the organizational structure tree in the application scenario can also be obtained. Therefore, the tree path between the user and any matching result in the organizational structure tree can be determined based on the department information.

[0022] In addition, it is generally believed that the contacts involved in the conversation are more likely to be colleagues in the same department or colleagues in a superior-subordinate relationship. This is reflected in the organizational structure tree in that the candidate contacts are closer based on the tree path in the organizational structure tree. Therefore, the closest matching result can be used as the recognition result.

[0023] In any of the above technical solutions, the step of matching the meta-word with the preset information table specifically includes: when the meta-word fails to match the preset information table, segmenting the meta-word and setting a sliding window; matching with the preset information table according to the sliding window.

[0024] In this technical solution, when a meta-word with a predetermined part of speech is matched with a preset information table and no matching result is found, the meta-word is split separately and then matched using a sliding window method. This can avoid inaccurate definition of word segmentation boundaries due to speech-to-text errors or other reasons, thereby further improving recognition accuracy.

[0025] In any of the above technical solutions, the step of preprocessing the input information to obtain multiple meta-words specifically includes: when the input information is voice information, converting the voice information into text information; segmenting the text information and tagging the part-of-speech of the segmentation results to obtain multiple meta-words.

[0026] In this technical solution, when the input information is voice information, the voice information is first converted into text information. Then the text information is segmented to obtain basic word units (meaningful minimum semantic units), and then the words are tagged with parts of speech. Specifically, the part of speech of each word can be given in combination with a pre-trained language model, and finally a meta-word set is obtained, which includes multiple meta-words. Subsequent recognition based on these meta-words can greatly improve the accuracy of recognition.

[0027] For example, "P comes from A Co., Ltd." will be segmented normally to get meaningful words such as P, from, and A Co., Ltd. At the same time, we combine the pre-trained language model to give the part-of-speech tag for each word. P is marked as PER (name), from is marked as V (verb), and A Co., Ltd. is marked as ORG (organization). Based on the useful information of these segmentation and part-of-speech tags, subsequent contact recognition can greatly improve the recognition accuracy.

[0028] In any of the above technical solutions, the predetermined parts of speech include: names of persons, places, institutions, and departments.

[0029] In this technical solution, the predetermined parts of speech include names of people, places, institutions, and departments, but are not limited thereto.

[0030] According to the second aspect of the present invention, a recognition system is proposed, comprising: a preprocessing unit, used to respond to user request instructions, preprocess input information, and obtain multiple meta-words; a determination unit, used to determine meta-words with predetermined parts of speech among the multiple meta-words; and a recognition unit, used to match the meta-words with a preset information table to obtain a recognition result.

[0031] The recognition system provided by the present invention comprises a preprocessing unit, a determination unit and a recognition unit. In response to a user's request operation, the preprocessing unit preprocesses the input information to obtain a plurality of meta-words. The recognition unit matches the sliding window based on the meta-words and the information in the preset information table. For each meta-word, in combination with its part of speech, only the meta-words with the predetermined part of speech will be matched with the information in the preset information table, so that many false alarms caused by homophony can be filtered out, thereby preventing an excessively high false alarm rate.

[0032] For example, for input information such as "We have a regular meeting every Wednesday afternoon", after preprocessing, the meta-word set is {every Wednesday afternoon, we, have, to, hold, regular meeting}. If the part of speech is not considered, then the word "regular meeting" will match the contact "A" with the same pronunciation in the address book, which is obviously not what we want. Therefore, we only consider those meta-words with the part of speech marked as the name PER and match them with the contacts in the address book to prevent too high a false positive rate.

[0033] It should be noted here that the matching method directly based on the sliding window is prone to many false positives. The sliding window may cover words from different words, forming meaningless words, and then it can match the names of people in the address book, thereby forming false positives. For example, in the input information "there is a meeting at 3 pm", "to be held" is not a meaningful unit, but based on this word, it can match "B" with the same pronunciation as the name in the address book, thereby causing false positives. The present invention first pre-processes the input information, and the pre-processed input information is converted into multiple meta-words, each of which is a meaningful minimum semantic unit. On this basis, the matching method of the sliding window is adopted, which can avoid the sliding window from creating some so-called new words for matching, thereby further improving the accuracy of recognition. Among them, the input information can be voice information or text information. The preset information table is determined according to the recognition requirements. When it is necessary to identify the contact, the preset information table can be a list of names. Further, the list of names can be a list of Chinese characters and / or a list of pinyin. Of course, the preset information table is not limited to a list of names.

[0034] The recognition system is used in electronic devices, and specifically, can be used in various scenarios where contact recognition is required in mobile and PC dialogue robots.

[0035] According to a third aspect of the present invention, an identification system is proposed, comprising: a memory storing programs or instructions; and a processor, which implements the steps of the identification method of any of the above technical solutions when executing the programs or instructions.

[0036] The recognition system provided by the present invention includes a memory and a processor. The memory stores a program or instruction, and when the processor executes the program or instruction, the steps of the recognition method of any of the above technical solutions can be implemented. Therefore, the recognition system has all the beneficial effects of the recognition method, which will not be discussed one by one here.

[0037] According to a fourth aspect of the present invention, an electronic device is provided, comprising the identification system according to the above technical solution.

[0038] The electronic device provided by the present invention includes the identification system of any of the above technical solutions. Therefore, the electronic device has all the beneficial effects of the identification system, which will not be discussed one by one here.

[0039] According to a fifth aspect of the present invention, a storage medium is provided, on which a program is stored, and when the program is executed by a processor, the steps of the identification method of any of the above technical solutions are implemented.

[0040] The storage medium provided by the present invention can implement the steps of the identification method of any of the above technical solutions when the program stored therein is executed. Therefore, all the beneficial effects of the above identification method are achieved and will not be discussed one by one here.

[0041] Additional aspects and advantages of the present invention will become apparent from the following description or may be learned by practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] The above and / or additional aspects and advantages of the present invention will become apparent and easily understood from the description of the embodiments in conjunction with the following drawings, in which:

[0043] Figure 1 It is one of the flowcharts of the identification method of one embodiment of the present invention;

[0044] Figure 2 This is a second flow chart of an identification method according to an embodiment of the present invention;

[0045] Figure 3 This is a third flow chart of an identification method according to an embodiment of the present invention;

[0046] Figure 4 This is a fourth flow chart of an identification method according to an embodiment of the present invention;

[0047] Figure 5 This is a fifth flow chart of an identification method according to an embodiment of the present invention;

[0048] Figure 6 This is a sixth flow chart of an identification method according to an embodiment of the present invention;

[0049] Figure 7 is a schematic block diagram of an identification system according to an embodiment of the present invention;

[0050] Figure 8 is a schematic block diagram of an identification system according to an embodiment of the present invention;

[0051] Fig. 9 is a schematic diagram of the recognition process of a recognition method according to a specific embodiment of the present invention;

[0052] Fig.10 It is a schematic diagram of an organizational structure tree of a specific embodiment of the present invention. DETAILED DESCRIPTION

[0053] In order to more clearly understand the above-mentioned purpose, features and advantages of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that the embodiments of the present invention and the features in the embodiments can be combined with each other without conflict.

[0054] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Therefore, the protection scope of the present invention is not limited to the specific embodiments disclosed below.

[0055] Refer to the following Figures 1 to 10 The identification method, identification system, electronic device and storage medium according to some embodiments of the present invention are described.

[0056] Embodiment 1:

[0057] Figure 1 FIG. 1 is a flow chart of an identification method according to an embodiment of the present invention. The identification method includes:

[0058] Step 102, in response to the user's request operation, pre-processing the input information to obtain a plurality of meta-words;

[0059] Step 104, determining a meta-word with a predetermined part of speech among the plurality of meta-words;

[0060] Step 106, matching the meta-words with the preset information table to obtain a recognition result.

[0061] The recognition method provided by the embodiment of the present invention responds to the user's request operation, pre-processes the input information, and obtains multiple meta-words. A sliding window is matched based on the meta-words and the information in the preset information table. For each meta-word, combined with its part of speech, only the meta-words with the predetermined part of speech will be matched with the information in the preset information table, so that many false alarms caused by homophones can be filtered out, thereby preventing an excessively high false alarm rate.

[0062] For example, for input information such as "We have a regular meeting every Wednesday afternoon", after preprocessing, the meta-word set is {every Wednesday afternoon, we, have, to, hold, regular meeting}. If the part of speech is not considered, then the word "regular meeting" will match the contact "A" with the same pronunciation in the address book, which is obviously not what we want. Therefore, we only consider those meta-words with the part of speech marked as the name PER and match them with the contacts in the address book to prevent too high a false positive rate.

[0063] It should be noted here that the matching method directly based on the sliding window is prone to many false positives. The sliding window may cover words from different words, forming meaningless words, and then it can match the names of people in the address book, thereby forming false positives. For example, in the input information "there is a meeting at 3 pm", "to be held" is not a meaningful unit, but based on this word, it can match "B" with the same pronunciation as the name in the address book, thereby causing false positives. The present invention first pre-processes the input information, and the pre-processed input information is converted into multiple meta-words, each of which is a meaningful minimum semantic unit. On this basis, the matching method of the sliding window is adopted, which can avoid the sliding window from creating some so-called new words for matching, thereby further improving the accuracy of recognition. Among them, the input information can be voice information or text information. The preset information table is determined according to the recognition requirements. When it is necessary to identify the contact, the preset information table can be a list of names. Further, the list of names can be a list of Chinese characters and / or a list of pinyin. Of course, the preset information table is not limited to a list of names.

[0064] The identification method is used in electronic devices, and specifically can be used in various scenarios where contact identification is required in mobile and PC dialogue robots.

[0065] Embodiment 2:

[0066] Figure 2 FIG2 is a flow chart of an identification method according to an embodiment of the present invention. The identification method includes:

[0067] Step 202, in response to the user's request operation, pre-process the input information to obtain a plurality of meta-words;

[0068] Step 204, determining a meta-word with a predetermined part of speech among the plurality of meta-words;

[0069] Step 206, when the meta-word is a Chinese character keyword, the meta-word is matched with the first preset information table; and / or the meta-word is converted into a pinyin keyword, and the pinyin keyword is matched with the second preset information table.

[0070] In this embodiment, since the input information may be voice information, text conversion is required during recognition. The converted text may only have the same pronunciation, and the Chinese characters may be very different. Matching only Chinese characters will result in a large number of missed reports. In order to improve the recall rate, pinyin matching can be added on the basis of Chinese character matching. Specifically, when the meta-word is a Chinese character keyword, it is matched with the first preset information table. Further, the meta-word in Chinese character form is converted into a pinyin keyword, and the pinyin keyword is matched with the second preset information table. This overcomes the errors in the speech-to-text conversion process, especially the missed reports caused by name conversion errors.

[0071] Embodiment three:

[0072] Figure 3 FIG3 is a flow chart of an identification method according to an embodiment of the present invention. The identification method includes:

[0073] Step 302, in response to the user's request operation, pre-processing the input information to obtain a plurality of meta-words;

[0074] Step 304, determining a meta-word with a predetermined part of speech among the plurality of meta-words;

[0075] Step 306, matching the meta-words with the preset information table to obtain at least two matching results;

[0076] Step 308: sort the at least two matching results according to the organizational structure tree of the organization to which the user belongs, and determine the recognition result according to the sorting order.

[0077] In this embodiment, for the same meta-word, due to the use of the pinyin matching method, two or more matching results with the same pronunciation may be obtained, or by using the Chinese matching method, at least two candidate words with the same name may be obtained. At this time, the two or more matching results are ranked and recommended according to the organizational tree of the organization where the user is located, so that the user can select the best possible matching result.

[0078] Embodiment 4:

[0079] Figure 4 FIG4 is a flow chart of an identification method according to an embodiment of the present invention. The identification method includes:

[0080] Step 402, in response to the user's request operation, pre-process the input information to obtain a plurality of meta-words;

[0081] Step 404, determining a meta-word with a predetermined part of speech among the plurality of meta-words;

[0082] Step 406, matching the meta-words with the preset information table to obtain at least two matching results;

[0083] Step 408, determining a tree path between any one of the at least two matching results and the user on the organization structure tree, and determining a distance between any one of the matching results and the user based on the tree path, to obtain at least two distances corresponding to the at least two matching results;

[0084] Step 410: sort at least two distances to determine a recognition result.

[0085] In this embodiment, by determining the tree path between any of the matching results and the user on the organizational tree, and determining the distance between any of the matching results and the user according to the tree path, the distances corresponding to the matching results are obtained. By sorting these distances, the matching results can be sorted, and then the matching results can be sorted and recommended.

[0086] In actual applications, the information of the dialog user, such as the department, etc., can be obtained, and the organizational structure tree in the application scenario can also be obtained. Therefore, the tree path between the user and any matching result in the organizational structure tree can be determined based on the department information.

[0087] In addition, it is generally believed that the contacts involved in the conversation are more likely to be colleagues in the same department or colleagues in a superior-subordinate relationship. This is reflected in the organizational structure tree in that the candidate contacts are closer based on the tree path in the organizational structure tree. Therefore, the closest matching result can be used as the recognition result.

[0088] Embodiment five:

[0089] Figure 5 FIG5 is a flowchart of an identification method according to an embodiment of the present invention. The identification method includes:

[0090] Step 502, in response to the user's request operation, pre-process the input information to obtain a plurality of meta-words;

[0091] Step 504, determining a meta-word with a predetermined part of speech among the plurality of meta-words;

[0092] Step 506, matching the meta-words with the preset information table;

[0093] Step 508, when no candidate word is matched, the meta-word is segmented and a sliding window is set; and matching is performed according to the sliding window and a preset information table;

[0094] Step 510, whether a match result is found; if so, proceed to step 512; if not, end;

[0095] Step 512, whether the number of matching results is greater than 2; if yes, proceed to step 516, if not, proceed to step 514;

[0096] Step 514, recommending the matching result as a recognition result to the user;

[0097] Step 516, determining a tree path between any one of the at least two matching results and the user on the organization structure tree, and determining a distance between any one of the matching results and the user based on the tree path, to obtain at least two distances corresponding to the at least two matching results;

[0098] Step 518: sort the at least two distances to determine the recognition result.

[0099] In this embodiment, when a meta-word with a predetermined part of speech is matched with a preset information table and no matching result is found, the meta-word is split separately and then matched using a sliding window method. This can avoid inaccurate definition of word segmentation boundaries due to speech-to-text errors or other reasons, thereby further improving recognition accuracy.

[0100] Embodiment six:

[0101] Figure 6 FIG6 is a flowchart of an identification method according to an embodiment of the present invention. The identification method includes:

[0102] Step 602, receiving input information input by the user;

[0103] Step 604: when the input information is voice information, convert the voice information into text information; segment the text information, and perform part-of-speech tagging on the segmentation results to obtain multiple meta-words;

[0104] Step 606, determining a meta-word with a predetermined part of speech among the multiple meta-words;

[0105] Step 608, matching the meta-words with the preset information table;

[0106] Step 610, when no candidate word is matched, the meta-word is segmented and a sliding window is set; and matching is performed according to the sliding window and a preset information table;

[0107] Step 612, whether a match result is found; if so, proceed to step 614; if not, end;

[0108] Step 614, whether the number of matching results is greater than 2; if yes, proceed to step 618, if no, proceed to step 616;

[0109] Step 616, recommending the matching result as a recognition result to the user;

[0110] Step 618, determining a tree path between any one of the at least two matching results and the user on the organization structure tree, and determining a distance between any one of the matching results and the user based on the tree path, to obtain at least two distances corresponding to the at least two matching results;

[0111] Step 620: sort at least two distances to determine a recognition result.

[0112] In this embodiment, when the input information is voice information, the voice information is first converted into text information. Then the text information is segmented to obtain basic word units (meaningful minimum semantic units), and then the words are tagged with parts of speech. Specifically, the part of speech of each word can be given in combination with a pre-trained language model, and finally a meta-word set is obtained, which includes multiple meta-words. Subsequent recognition based on these meta-words can greatly improve the accuracy of recognition.

[0113] For example, "P comes from A Co., Ltd." will be segmented normally to get meaningful words such as P, from, and A Co., Ltd. At the same time, we combine the pre-trained language model to give the part-of-speech tag for each word. P is marked as PER (name), from is marked as V (verb), and A Co., Ltd. is marked as ORG (organization). Based on the useful information of these segmentation and part-of-speech tags, subsequent contact recognition can greatly improve the recognition accuracy.

[0114] In any of the above embodiments, the predetermined parts of speech include: names of persons, places, institutions, and departments.

[0115] In this embodiment, the predetermined parts of speech include, but are not limited to, names of persons, places, institutions, and departments.

[0116] Embodiment seven:

[0117] Figure 7 is a schematic block diagram of an identification system 700 according to an embodiment of the present invention. The identification system 700 includes:

[0118] A preprocessing unit 702, configured to respond to a user request instruction and preprocess the input information to obtain a plurality of meta-words;

[0119] A determining unit 704, configured to determine a meta-word having a predetermined part of speech among the plurality of meta-words;

[0120] The recognition unit 706 is used to match the meta-words with the preset information table to obtain a recognition result.

[0121] The recognition system 700 provided by the embodiment of the present invention includes a preprocessing unit 702, a determination unit 704, and a recognition unit 706. In response to the user's request operation, the preprocessing unit 702 preprocesses the input information to obtain multiple meta-words. The recognition unit 706 performs sliding window matching based on the meta-words and the information in the preset information table. For each meta-word, combined with its part of speech, only the meta-words with the predetermined part of speech will be matched with the information in the preset information table, so that many false alarms caused by homophones can be filtered out, thereby preventing an excessively high false alarm rate.

[0122] For example, for input information such as "We have a regular meeting every Wednesday afternoon", after preprocessing, the meta-word set is {every Wednesday afternoon, we, have, to, hold, regular meeting}. If the part of speech is not considered, then the word "regular meeting" will match the contact "A" with the same pronunciation in the address book, which is obviously not what we want. Therefore, we only consider those meta-words with the part of speech marked as the name PER and match them with the contacts in the address book to prevent too high a false positive rate.

[0123] It should be noted here that the matching method directly based on the sliding window is prone to many false positives. The sliding window may cover words from different words, forming meaningless words, and then it can match the names of people in the address book, thereby forming false positives. For example, in the input information "there is a meeting at 3 pm", "to be held" is not a meaningful unit, but based on this word, it can match "B" with the same pronunciation as the name in the address book, thereby causing false positives. The present invention first pre-processes the input information, and the pre-processed input information is converted into multiple meta-words, each of which is a meaningful minimum semantic unit. On this basis, the matching method of the sliding window is adopted, which can avoid the sliding window from creating some so-called new words for matching, thereby further improving the accuracy of recognition. Among them, the input information can be voice information or text information. The preset information table is determined according to the recognition requirements. When it is necessary to identify the contact, the preset information table can be a list of names. Further, the list of names can be a list of Chinese characters and / or a list of pinyin. Of course, the preset information table is not limited to a list of names.

[0124] The recognition system 700 is used in electronic devices, and specifically, can be used in various scenarios where contact recognition is required in mobile and PC dialogue robots.

[0125] Embodiment eight:

[0126] Figure 8 8 is a schematic block diagram of an identification system 800 according to an embodiment of the present invention. The identification system 800 includes:

[0127] Memory 802, storing programs or instructions;

[0128] Processor 804, when executing a program or an instruction, implements the steps of the identification method in any of the above embodiments.

[0129] The recognition system 800 provided in the embodiment of the present invention includes a memory 802 and a processor 804. The memory 802 stores a program or instruction, and when the processor 804 executes the program or instruction, the steps of the recognition method of any of the above embodiments can be implemented, so the recognition system 800 has all the beneficial effects of the recognition method, which will not be discussed one by one here.

[0130] Embodiment nine:

[0131] An electronic device is proposed, comprising the identification system of any one of the above embodiments.

[0132] The electronic device provided by the embodiment of the present invention includes the identification system of any one of the above embodiments. Therefore, the electronic device has all the beneficial effects of the identification system, which will not be discussed one by one here.

[0133] Embodiment ten:

[0134] A storage medium is proposed, on which a program is stored. When the program is executed by a processor, the steps of the identification method in any of the above embodiments are implemented.

[0135] The storage medium provided by the embodiment of the present invention can implement the steps of the identification method of any of the above embodiments when the program stored therein is executed. Therefore, all the beneficial effects of the above identification method are achieved and will not be discussed one by one here.

[0136] With the continuous advancement of speech recognition technology and natural language processing technology, various human-computer dialogue VPAs are widely used in smart devices to enhance human-computer interaction functions. However, in human-computer dialogue robot applications, the effect of accurately identifying contacts from user input is not satisfactory.

[0137] To this end, a specific embodiment of the present invention provides a recognition method, the recognition process of the recognition method is as follows: Fig. 9 As shown. Among them, the core recognition algorithm includes:

[0138] 1. First, perform word segmentation and entity recognition on the input text to obtain the sentence word structure.

[0139] The purpose of word segmentation for Chinese sentences is to obtain the smallest meaningful semantic unit. For example, the sentence "P comes from A Co., Ltd." will be segmented normally to obtain meaningful words such as P, from, and A Co., Ltd. At the same time, we combine the pre-trained language model to give the part-of-speech tag for each word. P is marked as PER (name), from is marked as V (verb), and A Co., Ltd. is marked as ORG (organization).

[0140] Based on the useful information of these word segmentations and part-of-speech tags, subsequent contact recognition can greatly improve the recognition accuracy.

[0141] 2. After word segmentation of the input text, we get basic word units, which we call meta-words.

[0142] Taking "P from A Co., Ltd." as an example, the meta-word set obtained after word segmentation is {P, from, A Co., Ltd}. For each meta-word, we will combine its part of speech, and only the meta-words marked as PER will be matched with the contacts in the address book using pinyin and Chinese characters, which can filter out many false positives caused by pinyin.

[0143] For example, for an input like "We have a regular meeting every Wednesday afternoon", after word segmentation, the meta-word set is {every Wednesday afternoon, we, have, hold, regular meeting}. If the part of speech is not considered, the word "regular meeting" will match the contact "A" in the address book with the same pronunciation, which is obviously not what we want. Therefore, we only consider those meta-words with the part of speech tagged as PER and match them with the contacts in the address book to prevent too high a false positive rate.

[0144] There is also a situation where the meta-word marked as PER does not match any contact in the address book. For example, the sentence "Give C a message" has a segmentation result of {pair('give','p'), pair('C send','PER'), pair('message','n')}. For the meta-word C send marked as PER, it does not match in the address book. In fact, we think that there should be a personal name at the position of "C send", but the boundary may not be accurately determined. In fact, the real segmentation result should be {give, C, send a message}. At this time, we will split the meta-word separately, that is, call the original sliding window method to match the word, and finally match the contact "C" in the address book to improve the recall rate.

[0145] In addition, the colloquial and short input makes the semantics vague, especially the errors caused by voice-to-text conversion, which will affect the word segmentation performance and lead to inaccurate word segmentation boundaries. In order to further improve the recall rate, the results after word segmentation are matched with the meta-words as the basic sliding units for sliding window matching. That is, by combining the meta-words in adjacent positions to form new words, and then matching them with the contacts in the address book, identifying the contacts and improving the recall rate.

[0146] 3. For the same word, due to the use of the pinyin matching method, multiple candidates may appear. At this time, it is necessary to sort and recommend multiple candidates so that users can choose the most likely contact.

[0147] In a specific human-computer dialogue scenario, the user's information can be obtained, such as the department they are in, and the enterprise organization structure tree in the application scenario can also be obtained. Since the contacts involved in the conversation are more likely to be colleagues in the same department or colleagues in a superior-subordinate relationship, the candidate contacts in the organization structure tree are closer based on the tree path.

[0148] like Fig.10 As shown, assuming that the inquirer Q is in the third-level department 1, in a conversation text, the same position word matches three candidates A, B, and C, whose departments are first-level department 1, fourth-level department, and first-level department 2 respectively.

[0149] The distance between any two points on the tree node is defined as the shortest number of edges from one node to another. Therefore, the distances between the queryer Q and A, B, and C are 2, 6, and 4, respectively. Therefore, the final recommended sorting order is A, C, and B, which means that Q is most looking for A, followed by C, and finally B.

[0150] 4. In actual conversation scenarios, especially in voice conversations, users' spoken words are generally short and have vague semantics. At this time, the performance of the word segmentation and part-of-speech tagging algorithms based on the language model will be greatly affected. If you still force matching based on the analysis results, it is easy to cause a lot of missed reports. A sentence is short when the number of characters in the sentence is less than a given threshold (in actual applications, the given threshold is 4, but not limited to this). For short sentences, contact matching is performed based on the sliding window method to improve the recall rate.

[0151] For example, the voice originally wanted to express "find D", D is a contact in the address book, but the voice-to-text conversion may appear "Zhao E", and the word segmentation will treat Zhao E as a word and mark it as a noun. At this time, the contact D cannot be recognized. The word segmentation error is because the text is too short and the semantics are vague. Therefore, we do not use the word segmentation result, but directly use a character-based sliding window to match it, so as to recall the contact "D".

[0152] The recognition method provided in this embodiment uses a pre-trained language model to first perform word segmentation and entity recognition on the text to be recognized, and then matches the contacts of pinyin and Chinese characters based on the word segmentation results. In combination with some other specific optimization strategies, contact recognition is performed, which greatly improves the recognition accuracy and recall rate, and is an effective contact recognition method.

[0153] The identification method, identification system, electronic device and storage medium provided by the embodiments of the present invention have the following beneficial technical effects:

[0154] 1. For the user's voice input, add a pinyin matching method to improve the voice input recall rate;

[0155] 2. Based on word segmentation and entity recognition technology, make full use of the word structure relationship of the input text to improve the recognition accuracy and recall rate;

[0156] 3. A sliding window matching method based on meta-words can overcome the influence of speech recognition to text conversion errors and improve the recognition recall rate;

[0157] 4. Design a sorting algorithm for multiple contacts identified with the same word to improve the accuracy of recommendations to users.

[0158] In the description of this specification, the terms "first" and "second" are used for descriptive purposes only and cannot be understood as indicating or implying relative importance unless otherwise clearly specified and limited; the terms "connection", "installation", "fixation" and the like should be understood in a broad sense, for example, "connection" can be a fixed connection, a detachable connection, or an integral connection; it can be directly connected or indirectly connected through an intermediate medium. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0159] In the description of this specification, the description of the terms "one embodiment", "some embodiments", "specific embodiments", etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0160] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A recognition method, characterized in that: The identification method is used for electronic equipment, and the identification method includes: In response to a user's request operation, preprocessing the input information to obtain a plurality of meta-words; Determining a meta-word having a predetermined part of speech among the plurality of meta-words; The meta-words are matched with the preset information table to obtain a recognition result; the step of matching the meta-words with the preset information table specifically includes: When the meta-word is a Chinese character keyword, matching the meta-word with a first preset information table; and / or Converting the meta-words into pinyin keywords, and matching the pinyin keywords with a second preset information table; The step of matching the meta-words with the preset information table to obtain the recognition result specifically includes: Matching the meta-word with the preset information table to obtain at least two matching results; Sorting the at least two matching results according to the organizational structure tree of the organization to which the user belongs, and determining the recognition result according to the sorting order, includes: Determine a tree path between any one of the at least two matching results and the user on the organization structure tree, and determine a distance between any one of the at least two matching results and the user according to the tree path, to obtain at least two distances corresponding to the at least two matching results; Sorting the at least two distances to determine the recognition result; When the meta-word fails to match the preset information table, segmenting the meta-word and setting a sliding window; Matching is performed according to the sliding window and the preset information table.

2. The identification method according to claim 1, characterized in that: The step of preprocessing the input information to obtain a plurality of meta-words specifically includes: When the input information is voice information, converting the voice information into text information; The text information is segmented, and part-of-speech marking is performed on the segmentation results to obtain the multiple meta-words.

3. The identification method according to claim 2, characterized in that: The predetermined parts of speech include: names of persons, places, institutions, and departments.

4. A recognition system, characterized in that: include: A preprocessing unit, used to respond to a user request instruction, preprocess the input information, and obtain a plurality of meta-words; A determining unit, configured to determine a meta-word having a predetermined part of speech among the plurality of meta-words; The recognition unit is used to match the meta-word with the preset information table to obtain a recognition result, wherein the step of matching the meta-word with the preset information table specifically includes: When the meta-word is a Chinese character keyword, matching the meta-word with a first preset information table; and / or Converting the meta-words into pinyin keywords, and matching the pinyin keywords with a second preset information table; Matching the meta-word with the preset information table to obtain at least two matching results; Sorting the at least two matching results according to the organizational structure tree of the organization to which the user belongs, and determining the recognition result according to the sorting order, includes: Determine a tree path between any one of the at least two matching results and the user on the organization structure tree, and determine a distance between any one of the at least two matching results and the user according to the tree path, to obtain at least two distances corresponding to the at least two matching results; Sorting the at least two distances to determine the recognition result; When the meta-word fails to match the preset information table, segmenting the meta-word and setting a sliding window; Matching is performed according to the sliding window and the preset information table.

5. A recognition system, characterized in that: include: Memory, storing programs or instructions; A processor, wherein when executing the program or instruction, the processor implements the steps of the identification method according to any one of claims 1 to 3.

6. An electronic device, characterized in that: Comprising an identification system as claimed in claim 4 or 5.

7. A storage medium having a program stored thereon, characterized in that: When the program is executed by a processor, the identification method according to any one of claims 1 to 3 is implemented.