A method, system and medium for selecting answers to neuropsychological scales based on syntax trees

Abstract

The present invention discloses a method, system and medium for screening answers to a neuropsychological scale based on a syntax tree, and the method steps include: 1) obtaining the audio of a tester and transcribing the audio into a psychological scale text; 2) performing dependency syntactic analysis on a word sequence marked with word boundaries to obtain text features; 3) performing central word extraction on the text features to obtain the central word of the psychological scale text; 4) analyzing the central word of the psychological scale text to obtain a semantic role recognition result; 5) determining whether the semantic role recognition result contains the semantic role of the clinical summary in the MMSE scale, and if so, obtaining the answer to the psychological scale. The system includes an audio acquisition and conversion module, a word segmentation module, a part-of-speech tagging module, an entity recognition module, a dependency syntactic analysis module, a central word extraction and analysis module, and an answer output module; the present invention can accurately extract the true intention of the tester and improve the accuracy of the application of artificial intelligence technology in the neuropsychological scale.

Description

Technical Field

[0001] The present invention relates to the application field of artificial intelligence technology in neuropsychological scales for Alzheimer's disease, and specifically to a method, system and medium for screening answers to neuropsychological scales based on a syntax tree. Background Art

[0002] Alzheimer's disease (AD) is an age-related syndrome, and its incidence rate increases rapidly with age, becoming the fourth largest "killer" of the elderly after heart disease, cerebrovascular disease, and tumors. However, for Chongqing and underdeveloped areas around the world, its awareness rate, consultation rate, diagnosis rate, and treatment rate are all lower than the world average. Therefore, artificial intelligence technology that can accurately understand the Chongqing dialect will help promote early screening for dementia, especially early screening for Alzheimer's disease. Clinical studies have shown that quantifiable signs of cognitive decline associated with AD and mild cognitive impairment (MCI) are detectable in patients' language. Machine learning (ML) models have been shown to successfully detect AD using speech and language variables, such as analyzing sentence content from text recognized by artificial intelligence technology to infer the health status of patients. Since the transcript should be accurate enough to correctly represent syntactic and linguistic features, current methods usually rely on transcription by trained transcribers. But this will take a lot of time and effort.

[0003] In the MMSE scale test, the subjects are required to answer some daily questions in a timely manner, such as "What day is today?", "What day of the week is today?". Ideally, the tester can directly say the answer, but for the elderly, their recall and memory abilities have declined, and the first answer may not be what they really want to answer, and a second confirmation is needed. For example, the tester may answer, "The 18th, oh, no, the 19th", "Wednesday or Thursday, it should be Thursday". If only simple text matching is used, it will be difficult to filter out the tester's true intention (answer) from the text in the above examples. NLP technology can help us solve this type of problem. We can summarize this type of problem as effective entity extraction. Referring to the entity extraction task in NLP, it belongs to supervised learning, which requires a lot of manual annotation to complete the training. This will bring obstacles to the application of artificial intelligence technology in the field of elderly health management.

[0004] Therefore, it is necessary to extract the key information of the MMSE scale. There are many studies on key information extraction. After deep learning has been applied to many NLP scenarios and achieved good results, algorithms based on neural network models have gradually replaced key information extraction algorithms based on statistical algorithms. From the above analysis, it can be seen that in neuropsychological scale screening, the answer screening of complex logical transcription texts can be summarized as entity extraction and semantic role judgment, among which semantic role judgment refers to judging the semantic role of the extracted entity from the unstructured text on the basis of entity recognition, thereby judging the semantic information of the entity in the text.

[0005] From the implementation point of view, there are two categories: pipeline model and joint model. The main process of entity extraction and semantic role judgment based on pipeline is: extract entity from sentences with labeled target entity pairs and semantic roles, and finally output the semantic role of the existing entity as the prediction result. These models are mainly based on the network structure of LSTM, Transformer, BERT pre-training and its improved model. In summary, it can be considered that the pipeline model first extracts entity and then analyzes semantic roles. Under sufficient expectations, some pipeline models have good results, but they also have disadvantages, such as error propagation: the error of the entity recognition module will affect the performance of the subsequent semantic role analysis; ignore the relationship between the two subtasks; redundant information: when the semantic analysis of the identified entities is performed, some independent entities will bring redundant information, which increases the error rate. Compared with the pipeline method, the joint model can utilize the close information between entities and semantics, extract entities and annotate the semantics of entities at the same time, which solves the problems of the pipeline method well. Due to the different objects modeled in the joint learning method, the joint learning method can be divided into parameter sharing method and sequence labeling method. The parameter sharing method models entities and semantics separately, while the sequence labeling method directly models entity-semantics.

[0006] In summary, the answer screening algorithm in the neuropsychological scale can be further improved and summarized as follows:

[0007] 1) Pipeline-based entity extraction and semantic analysis may lead to errors, which will further expand the recognition errors.

[0008] 2) The extraction algorithm based on the joint model has high data requirements, which will increase the cost of applying AI technology in neuropsychological scales.

[0009] 3) The complex answer logic in the neuropsychological scale can be enumerated, and different sentence patterns have unique corresponding semantic roles, so unsupervised learning methods can be used to complete sentence analysis and answer screening. Summary of the invention

[0010] The purpose of the present invention is to provide a method for screening answers to a neuropsychological scale based on a syntax tree, comprising the following steps:

[0011] 1) Obtain the tester's audio and transcribe it into psychological scale text;

[0012] 2) Use the word segmentation module to segment the psychological scale text and obtain the word sequence;

[0013] 3) Using the part-of-speech tagging module to tag each word in the word sequence with a tag that identifies the word boundary; using the entity recognition module to tag the named entity in the word sequence with a tag that identifies the entity word boundary;

[0014] 4) using a dependency syntactic analysis algorithm based on a neural network to perform dependency syntactic analysis on a word sequence marked with word boundaries to obtain text features; the text features include text dependency syntactic relationship features and semantic role features;

[0015] 5) Extract the central word of the text features according to the pre-stored fuzzy word dictionary to obtain the central word of the psychological scale text;

[0016] 6) Analyze the central words of the psychological scale text to obtain semantic role recognition results;

[0017] 7) Determine whether the semantic role recognition result contains the semantic role of the clinical summary in the MMSE scale. If so, combine the text dependency syntactic relationship features and the semantic role features to obtain the psychological scale answer. Otherwise, return to step 1).

[0018] Furthermore, the step of performing word segmentation processing on the psychological scale text by using the word segmentation module includes: marking each character in the psychological scale text with a mark identifying a word boundary by using the word segmentation module.

[0019] Furthermore, the text-dependent syntactic relationship features include subject-predicate relationship, verb-object relationship, intermezzo-object relationship, preposed object, auxiliary modifier, attributive-predicate relationship, adverbial-predicate structure, verb-complement structure, parallel relationship, preposition-object relationship, left-attachment relationship, right-attachment relationship, independent structure, and core relationship.

[0020] Furthermore, the semantic role features include agent, patient, adverbial, beneficiary, condition, parallelism, degree, direction, conversation marker, scope, frequency, place, and manner.

[0021] A system for applying a neuropsychological scale answer screening method based on a syntax tree, comprising an audio acquisition and conversion module, a word segmentation module, a part-of-speech tagging module, an entity recognition module, a dependency syntax analysis module, a central word extraction and analysis module, and an answer output module;

[0022] The audio acquisition and conversion module acquires the tester's audio, transcribes the audio into a psychological scale text, and transmits it to the word segmentation module;

[0023] The word segmentation module performs word segmentation processing on the psychological scale text to obtain a word sequence, and transmits the word sequence to the part-of-speech tagging module and the entity recognition module respectively;

[0024] The part-of-speech tagging module tags each word in the word sequence with a marker that identifies a word boundary;

[0025] The entity recognition module marks a marker identifying the boundary of an entity word on a named entity in a word sequence;

[0026] The dependency syntactic analysis module uses a dependency syntactic analysis algorithm based on a neural network to perform dependency syntactic analysis on a word sequence marked with word boundaries to obtain text features; the text features include text dependency syntactic relationship features and semantic role features;

[0027] The central word extraction and analysis module extracts the central word from the text features according to the pre-stored fuzzy word dictionary to obtain the central word of the psychological scale text;

[0028] The central word extraction and analysis module analyzes the central word of the psychological scale text to obtain a semantic role recognition result;

[0029] The answer output module determines whether the semantic role recognition result contains the semantic role of the clinical summary in the MMSE scale. If so, the text dependency syntactic relationship feature and the semantic role feature are combined to obtain the answer to the psychological scale.

[0030] Furthermore, it also includes a display module for displaying answers to the psychological scale.

[0031] A computer storage medium stores a computer program; when the computer program is called, the steps of the above method are executed.

[0032] The technical effect of the present invention is unquestionable. The present invention solves the problem of how computers understand human speech during intelligent interaction.

[0033] The present invention utilizes the rich, efficient and accurate natural language processing technologies provided by the LTP platform, including Chinese word segmentation, part-of-speech tagging, named entity recognition, dependency syntactic analysis, semantic role labeling, etc., to perform word segmentation, labeling and syntactic analysis on the transcribed text of the tester's answer, accurately extract the tester's true intention, and improve the accuracy of the application of artificial intelligence technology in neuropsychological scales.

[0034] The present invention aims at solving the difficulty that the pipeline and joint model require a large amount of labeled data, and proposes an unsupervised learning algorithm based on LTP to complete text multi-entity screening and semantic role judgment.

[0035] The present invention proposes for the first time that in the MMSE scale using artificial intelligence technology, when the transcribed text contains multiple answers, the correct answer is extracted using a natural language processing algorithm.

[0036] The present invention summarizes the sentence structures of clinical testers' answers to MMSE scale items, and provides a linguistic reference for the subsequent diagnosis of Alzheimer's disease. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 This is a flow chart for LTP analysis;

[0038] Figure 2 For dependency syntactic relations;

[0039] Figure 3 for semantic roles;

[0040] Figure 4 Flowchart for multiple answer screening of neuropsychological scales;

[0041] Figure 5 For semantic analysis example I;

[0042] Figure 6 For semantic analysis example II;

[0043] Figure 7 For semantic analysis example III;

[0044] Figure 8 For semantic analysis example IV. DETAILED DESCRIPTION

[0045] The present invention is further described below in conjunction with the embodiments, but it should not be understood that the above subject matter of the present invention is limited to the following embodiments. Without departing from the above technical ideas of the present invention, various substitutions and changes are made according to the common technical knowledge and customary means in the art, which should all be included in the protection scope of the present invention.

[0046] Embodiment 1:

[0047] See also Figures 1 to 8 , a neuropsychological scale answer screening method based on syntax tree, comprising the following steps:

[0048] 1) Obtain the tester's audio and transcribe it into psychological scale text;

[0049] 2) Use the word segmentation module to segment the psychological scale text and obtain the word sequence;

[0050] 3) Using the part-of-speech tagging module to tag each word in the word sequence with a tag that identifies the word boundary; using the entity recognition module to tag the named entity in the word sequence with a tag that identifies the entity word boundary;

[0051] 4) using a dependency syntactic analysis algorithm based on a neural network to perform dependency syntactic analysis on a word sequence marked with word boundaries to obtain text features; the text features include text dependency syntactic relationship features and semantic role features;

[0052] 5) Extract the central word of the text features according to the pre-stored fuzzy word dictionary to obtain the central word of the psychological scale text;

[0053] 6) Analyze the central words of the psychological scale text to obtain semantic role recognition results;

[0054] 7) Determine whether the semantic role recognition result contains the semantic role of the clinical summary in the MMSE scale. If so, combine the text dependency syntactic relationship features and the semantic role features to obtain the psychological scale answer. Otherwise, return to step 1).

[0055] The step of using the word segmentation module to perform word segmentation processing on the psychological scale text includes: using the word segmentation module to mark each character in the psychological scale text with a mark that identifies a word boundary.

[0056] The text-dependent syntactic relationship features include subject-predicate relationship, verb-object relationship, intermezzo-object relationship, preposed object, auxiliary modifier, attributive-predicate relationship, adverbial-predicate structure, verb-complement structure, parallel relationship, preposition-object relationship, left-attachment relationship, right-attachment relationship, independent structure, and core relationship.

[0057] The semantic role features include agent, patient, adverbial, beneficiary, condition, coordination, degree, direction, conversation marker, scope, frequency, place, and manner.

[0058] Terminology explanation:

[0059] NLP: Natural Language Processing;

[0060] LTP: Language Technology Platform Harbin Institute of Technology Language Technology Platform;

[0061] NER: entity extraction;

[0062] MMSE: Mini-mental State Examination MMSE scale;

[0063] AD: Avenue Alzheimer's disease;

[0064] MCI: Mild Cognitive Impairment.

[0065] Embodiment 2:

[0066] A system for applying a neuropsychological scale answer screening method based on a syntax tree, comprising an audio acquisition and conversion module, a word segmentation module, a part-of-speech tagging module, an entity recognition module, a dependency syntax analysis module, a central word extraction and analysis module, and an answer output module;

[0067] The audio acquisition and conversion module acquires the tester's audio, transcribes the audio into a psychological scale text, and transmits it to the word segmentation module;

[0068] The word segmentation module performs word segmentation processing on the psychological scale text to obtain a word sequence, and transmits the word sequence to the part-of-speech tagging module and the entity recognition module respectively;

[0069] The part-of-speech tagging module tags each word in the word sequence with a marker that identifies a word boundary;

[0070] The entity recognition module marks a marker identifying the boundary of an entity word on a named entity in a word sequence;

[0071] The dependency syntactic analysis module uses a dependency syntactic analysis algorithm based on a neural network to perform dependency syntactic analysis on a word sequence marked with word boundaries to obtain text features; the text features include text dependency syntactic relationship features and semantic role features;

[0072] The central word extraction and analysis module extracts the central word from the text features according to the pre-stored fuzzy word dictionary to obtain the central word of the psychological scale text;

[0073] The central word extraction and analysis module analyzes the central word of the psychological scale text to obtain a semantic role recognition result;

[0074] The answer output module determines whether the semantic role recognition result contains the semantic role of the clinical summary in the MMSE scale. If so, the text dependency syntactic relationship feature and the semantic role feature are combined to obtain the answer to the psychological scale.

[0075] The system also includes a display module for displaying answers to the psychological scale.

[0076] Embodiment 3:

[0077] A computer storage medium stores a computer program; when the computer program is called, the steps of the method described in Example 1 are executed.

[0078] Embodiment 4:

[0079] A method for screening answers to a neuropsychological scale based on a syntax tree, including the following contents:

[0080] The text is processed by language analysis tools, in which the word segmentation module uses LTP to label each character in the sentence with a marker to identify the word boundary. The part-of-speech tagging module models the part-of-speech tagging task after word segmentation as a word-based sequence tagging problem. For the word sequence of the input sentence, the model labels each word in the sentence with a marker to identify the word boundary. The entity recognition module, like the word segmentation module, models named entity recognition as a word-based sequence tagging problem. Semantic analysis is performed through the dependency syntactic analysis module. In LTP, the neural network dependency syntactic analysis algorithm is used, combined with the rich global features and clustering features of the text, to obtain features such as semantic roles and dependency syntactic relationships. The parsing results can be referred to. Figure 2 .

[0081] Predefined fuzzy words are stored in a dictionary, "positive": ['is', 'is', 'like'], "negative": ['is not', 'is not'], "doubt": ['or', 'is it']", and by using the results of LTP analysis, the present invention can quickly retrieve the central word of the test-taker's answer text. Relying on the central word, the present invention can quickly complete all semantic role judgments and object extraction of the text. If the extracted result contains the semantic role of the clinical summary in the MMSE scale, the answer extraction is completed. Otherwise, it cannot be judged and the downstream module continues to judge.

[0082] Embodiment 5:

[0083] A method for screening answers to a neuropsychological scale based on a syntax tree comprises the following steps:

[0084] 1) obtaining the tester's audio and transcribing the audio into a psychological scale text;

[0085] 2) Use the word segmentation module to segment the psychological scale text and obtain the word sequence;

[0086] 3) Use the part-of-speech tagging module to tag each word in the word sequence with a tag that identifies the word boundary; use the entity recognition module to tag the named entity in the word sequence with a tag that identifies the entity word boundary.

[0087] 4) using a dependency syntactic analysis algorithm based on a neural network to perform dependency syntactic analysis on a word sequence marked with word boundaries to obtain text features; the text features include text dependency syntactic relationship features and semantic role features;

[0088] 5) Extract the central word of the text features according to the pre-stored fuzzy word dictionary to obtain the central word of the psychological scale text;

[0089] 6) Analyze the central words of the psychological scale text to obtain semantic role recognition results;

[0090] 7) Determine whether the semantic role recognition result contains the semantic role of the clinical summary in the MMSE scale. If so, combine the text dependency syntactic relationship features and the semantic role features to obtain the psychological scale answer. Otherwise, return to step 1).

[0091] Embodiment 6:

[0092] A method for screening answers to a neuropsychological scale based on a syntax tree, the main content of which is shown in Example 5, wherein the step of using a word segmentation module to perform word segmentation processing on the psychological scale text includes: using the word segmentation module to mark each character in the psychological scale text with a marker that identifies a word boundary.

[0093] Embodiment 7:

[0094] A method for screening answers to a neuropsychological scale based on a grammatical tree, the main content of which is shown in Example 5, wherein the text dependency syntactic relationship features include subject-predicate relationship, verb-object relationship, intermezzo-object relationship, preposed object, auxiliary modifier, attributive-predicate relationship, adverbial-predicate structure, verb-complement structure, parallel relationship, prepositional-object relationship, left-attached relationship, right-attached relationship, independent structure, and core relationship.

[0095] Embodiment 8:

[0096] A method for screening answers to a neuropsychological scale based on a grammatical tree, the main content of which is shown in Example 5, wherein the semantic role features include agent, patient, adverbial, beneficiary, condition, parallel, degree, direction, conversation marker, range, frequency, place, and manner.

[0097] Embodiment 9:

[0098] A system for applying the method for selecting answers to a neuropsychological scale based on a syntax tree as described in any one of embodiments 5 to 8, characterized in that it includes an audio acquisition and conversion module, a word segmentation module, a part-of-speech tagging module, an entity recognition module, a dependency syntax analysis module, a central word extraction and analysis module, and an answer output module;

[0099] The audio acquisition and conversion module acquires the tester's audio, transcribes the audio into a psychological scale text, and transmits it to the word segmentation module;

[0100] The word segmentation module performs word segmentation processing on the psychological scale text to obtain a word sequence, and transmits the word sequence to the part-of-speech tagging module and the entity recognition module respectively;

[0101] The part-of-speech tagging module tags each word in the word sequence with a marker that identifies a word boundary;

[0102] The entity recognition module marks a marker identifying the boundary of an entity word on a named entity in a word sequence;

[0103] The dependency syntactic analysis module uses a dependency syntactic analysis algorithm based on a neural network to perform dependency syntactic analysis on a word sequence marked with word boundaries to obtain text features; the text features include text dependency syntactic relationship features and semantic role features;

[0104] The central word extraction and analysis module extracts the central word from the text features according to the pre-stored fuzzy word dictionary to obtain the central word of the psychological scale text;

[0105] The central word extraction and analysis module analyzes the central word of the psychological scale text to obtain a semantic role recognition result;

[0106] The answer output module determines whether the semantic role recognition result contains the semantic role of the clinical summary in the MMSE scale. If so, the text dependency syntactic relationship feature and the semantic role feature are combined to obtain the answer to the psychological scale.

[0107] Embodiment 10:

[0108] The system using the method for screening answers to a neuropsychological scale based on a syntax tree has its main contents as shown in Example 9, wherein the system also includes a display module for displaying answers to the psychological scale.

Claims

1. A method for screening answers to a neuropsychological scale based on a syntax tree, characterized in that: The following steps are involved: 1) Obtain the tester's audio and transcribe it into psychological scale text; 2) Use the word segmentation module to segment the psychological scale text and obtain the word sequence; 3) Using the part-of-speech tagging module to tag each word in the word sequence with a tag that identifies the word boundary; using the entity recognition module to tag the named entity in the word sequence with a tag that identifies the entity word boundary; 4) using a dependency syntactic analysis algorithm based on a neural network to perform dependency syntactic analysis on a word sequence marked with word boundaries to obtain text features; the text features include text dependency syntactic relationship features and semantic role features; 5) Extract the central word of the text features according to the pre-stored fuzzy word dictionary to obtain the central word of the psychological scale text; Fuzzy words include: "positive": ['is', 'just', 'like'], "negative": ['not', 'not like'], "doubt": ['still', 'is it']"; 6) Analyze the central words of the psychological scale text to obtain semantic role recognition results; 7) Determine whether the semantic role recognition result contains the semantic role of the clinical summary in the MMSE scale. If so, combine the text dependency syntactic relationship features and the semantic role features to obtain the psychological scale answer. Otherwise, return to step 1); The steps of using the word segmentation module to perform word segmentation processing on the psychological scale text include: using the word segmentation module to mark each character in the psychological scale text with a mark that identifies a word boundary; The text dependency syntactic relationship features include subject-predicate relationship, verb-object relationship, intermezzo-object relationship, prepositioned object, auxiliary modifier, attributive-predicate relationship, adverbial-predicate structure, verb-complement structure, parallel relationship, preposition-object relationship, left-attachment relationship, right-attachment relationship, independent structure, and core relationship; The semantic role features include agent, patient, adverbial, beneficiary, condition, coordination, degree, direction, conversation marker, scope, frequency, place, and manner.

2. A system using the method for selecting answers to a neuropsychological scale based on a syntax tree according to claim 1, characterized in that: It includes audio acquisition and conversion module, word segmentation module, part-of-speech tagging module, entity recognition module, dependency syntax analysis module, core word extraction and analysis module, and answer output module; The audio acquisition and conversion module acquires the tester's audio, transcribes the audio into a psychological scale text, and transmits it to the word segmentation module; The word segmentation module performs word segmentation processing on the psychological scale text to obtain a word sequence, and transmits the word sequence to the part-of-speech tagging module and the entity recognition module respectively; The part-of-speech tagging module tags each word in the word sequence with a marker that identifies a word boundary; The entity recognition module marks a marker identifying the boundary of an entity word on a named entity in a word sequence; The dependency syntactic analysis module uses a dependency syntactic analysis algorithm based on a neural network to perform dependency syntactic analysis on a word sequence marked with word boundaries to obtain text features; the text features include text dependency syntactic relationship features and semantic role features; The central word extraction and analysis module extracts the central word from the text features according to the pre-stored fuzzy word dictionary to obtain the central word of the psychological scale text; The central word extraction and analysis module analyzes the central word of the psychological scale text to obtain a semantic role recognition result; The answer output module determines whether the semantic role recognition result contains the semantic role of the clinical summary in the MMSE scale, and if so, combines the text dependency syntactic relationship feature and the semantic role feature to obtain the answer to the psychological scale; A display module for displaying answers to the psychological scale is also included.

3. A computer storage medium, characterized in that: A computer program is stored thereon; when the computer program is called, the method described in claim 1 is executed.

Images