CN112069298A - Human-computer interaction method, device and medium based on semantic web and intention recognition - Google Patents
Human-computer interaction method, device and medium based on semantic web and intention recognition Download PDFInfo
- Publication number
- CN112069298A CN112069298A CN202010756664.6A CN202010756664A CN112069298A CN 112069298 A CN112069298 A CN 112069298A CN 202010756664 A CN202010756664 A CN 202010756664A CN 112069298 A CN112069298 A CN 112069298A
- Authority
- CN
- China
- Prior art keywords
- intention
- question
- semantic
- standard
- candidate
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 230000003993 interaction Effects 0.000 title claims abstract description 76
- 238000000034 method Methods 0.000 title claims abstract description 56
- 238000012549 training Methods 0.000 claims abstract description 63
- 238000013145 classification model Methods 0.000 claims abstract description 30
- 230000002452 interceptive effect Effects 0.000 claims abstract description 16
- 230000009471 action Effects 0.000 claims description 44
- 230000011218 segmentation Effects 0.000 claims description 34
- 238000010276 construction Methods 0.000 claims description 22
- 238000010801 machine learning Methods 0.000 claims description 17
- 238000004422 calculation algorithm Methods 0.000 claims description 15
- 238000002372 labelling Methods 0.000 claims description 11
- 238000012545 processing Methods 0.000 claims description 9
- 238000001914 filtration Methods 0.000 claims description 8
- 238000004590 computer program Methods 0.000 claims description 6
- 239000003607 modifier Substances 0.000 claims description 6
- 238000003064 k means clustering Methods 0.000 claims description 5
- 238000012986 modification Methods 0.000 claims description 5
- 230000004048 modification Effects 0.000 claims description 5
- 238000007781 pre-processing Methods 0.000 claims description 4
- 230000000295 complement effect Effects 0.000 claims description 3
- 238000012216 screening Methods 0.000 claims description 3
- 238000003058 natural language processing Methods 0.000 abstract 1
- 230000008569 process Effects 0.000 description 14
- 238000012360 testing method Methods 0.000 description 6
- 239000002131 composite material Substances 0.000 description 4
- 238000010586 diagram Methods 0.000 description 3
- 239000013598 vector Substances 0.000 description 3
- 238000007476 Maximum Likelihood Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- RWSOTUBLDIXVET-UHFFFAOYSA-N Dihydrogen sulfide Chemical compound S RWSOTUBLDIXVET-UHFFFAOYSA-N 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000007635 classification algorithm Methods 0.000 description 1
- 210000000078 claw Anatomy 0.000 description 1
- 150000001875 compounds Chemical class 0.000 description 1
- 238000013500 data storage Methods 0.000 description 1
- 238000003066 decision tree Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 239000012634 fragment Substances 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000010295 mobile communication Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000006116 polymerization reaction Methods 0.000 description 1
- 239000007787 solid Substances 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/33—Querying
- G06F16/332—Query formulation
- G06F16/3329—Natural language query formulation or dialogue systems
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/35—Clustering; Classification
- G06F16/355—Class or cluster creation or modification
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/20—Natural language analysis
- G06F40/205—Parsing
- G06F40/211—Syntactic parsing, e.g. based on context-free grammar [CFG] or unification grammars
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/20—Natural language analysis
- G06F40/237—Lexical tools
- G06F40/242—Dictionaries
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/20—Natural language analysis
- G06F40/279—Recognition of textual entities
- G06F40/289—Phrasal analysis, e.g. finite state techniques or chunking
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/30—Semantic analysis
- G06F40/35—Discourse or dialogue representation
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Computational Linguistics (AREA)
- Artificial Intelligence (AREA)
- Audiology, Speech & Language Pathology (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- Mathematical Physics (AREA)
- Human Computer Interaction (AREA)
- Machine Translation (AREA)
Abstract
The invention discloses a human-computer interaction method based on semantic web and intention recognition, relates to the field of natural language processing, and aims to accurately recognize question intentions and improve question and answer quality of human-computer interaction. The method comprises the following steps: obtaining common problem solutions in the industry as an interactive data source; performing semantic annotation on the standard questions in the common question answers to construct an industry semantic network; acquiring a training corpus; training an intention recognition classification model through the training corpus; receiving a user question, performing intention identification on the user question through the intention identification classification model to obtain an intention candidate set, performing multiple rounds of man-machine interaction through the industry semantic network based on the intention candidate set, determining a standard question matched with the intention of the user question, and outputting an answer. The invention also discloses an electronic device and a computer storage medium.
Description
Technical Field
The invention relates to the field of natural voice processing, in particular to a man-machine interaction method, equipment and medium based on semantic web and intention recognition.
Background
Personnel of a call center or a customer service center generally have high mobility, so that the problems of high training cost of enterprises, low customer service satisfaction degree and the like are caused, and the operation cost is greatly increased. Therefore, the intelligent customer service is more and more emphasized, but the intelligent customer service faces the problems of inaccurate intention identification, fuzzy intention, incapability of positioning, high training corpus maintenance cost and the like in the interaction process.
In order to solve the above problems, in the prior art, there is chinese patent application 201710575327.5, which discloses a question-answering method and apparatus based on a knowledge graph, acquiring a natural query sentence input by a user, and identifying a global unique identifier GUID of an entity in the natural query sentence for the knowledge graph, where the knowledge graph includes attributes and attribute values of the entity and relationships between the entities; analyzing the natural query statement into a syntax tree according to the context-free grammar rule, and obtaining a logic expression corresponding to the natural query statement according to the syntax tree; generating a machine query statement corresponding to the knowledge graph according to the logic expression and the GUID of the entity; and inquiring the question and answer result corresponding to the machine query statement in the knowledge graph according to the machine query statement, and feeding back the question and answer result to the user. So as to obtain accurate question and answer results aiming at the question and answer. However, the patent application needs a knowledge graph constructed by a large number of corpora, and the current general entity recognition model can only recognize names of people, places, mechanisms and the like, and the recognition of professional entities in a specific field lacks training corpora, which results in lack of question and answer applicability in the specific field, and the entities and attributes cannot be extracted due to the existence of a plurality of spoken sentences in the question and answer process, so that the problem cannot be accurately positioned.
Disclosure of Invention
In order to overcome the defects of the prior art, the invention aims to provide a human-computer interaction method based on semantic web and intention recognition, which can realize accurate recognition of problem intentions and improve the question-answering quality of human-computer interaction.
One of the purposes of the invention is realized by adopting the following technical scheme:
a man-machine interaction method based on semantic web and intention recognition comprises the following steps:
obtaining common problem solutions in the industry as an interactive data source;
performing semantic annotation on the standard questions in the common question answers to construct an industry semantic network;
acquiring a training corpus, wherein the training corpus comprises the standard problem, a similar problem of the standard problem and an intention label corresponding to the standard problem;
training a machine learning model through the training corpus to obtain an intention recognition classification model;
receiving a user question, and performing intention identification on the user question through the intention identification classification model to obtain an intention candidate set, wherein the intention candidate set comprises standard questions in a plurality of intention categories;
and performing multiple rounds of man-machine interaction through the industry semantic network based on the intention candidate set, determining a standard question matched with the intention of the user question, inquiring an answer of the standard question matched with the intention of the user question from the interaction data source, and outputting the answer.
Further, performing word segmentation, part-of-speech tagging and syntactic analysis on the standard problem according to a pre-constructed word segmentation dictionary to obtain a dependency syntactic relation among all the word segments in the standard problem;
extracting semantic attributes of each participle in the standard problem according to a semantic labeling rule, wherein the semantic attributes comprise an individual, a idiom, an action attribute, a fixed language and a data attribute;
marking each participle in the standard problem according to the semantic attribute to obtain a semantic marking result of each participle in the standard problem;
according to the semantic annotation result, constructing a semantic network according to the following query sequence: querying individuals in the standard questions, and taking the individuals as nodes of the semantic network; inquiring action attributes corresponding to the individuals, and taking the action attributes as branch nodes of the individuals; inquiring the data attribute and the shape language corresponding to the action attribute, and respectively taking the inquired data attribute and the shape language as branch nodes of the action attribute; and inquiring a fixed language corresponding to the data attribute, and taking the inquired fixed language as a branch node of the data attribute.
Further, the method also comprises the following steps:
action attribute determination rule: if the core word is the first verb in the standard question, marking the core word as an action attribute; if the core word is a non-verb, searching a verb mark closest to the core word as an action attribute; if the core word is not the first verb, the verb which has a direct relation with the core word is searched and marked as an action attribute;
individual determination rules: when the number of the participles of the standard question is less than 3, marking the first unlabeled participle as an individual; when two unlabeled participles are not connected and have the closest distance, labeling the first unlabeled participle as an individual; when two unlabeled participles are connected and have a modification relation, merging the two unlabeled and connected participles and labeling as an individual;
word segmentation and combination rules: when the participles needing to be combined are connected and not marked as the semantic attributes and are not adjectives or adverbs, combining the fixed relation and the state-middle relation of the standard problem with the modified words; merging the participles parallel to the core word with the object; merging parallel objects in the standard questions; merging a core word and an object of the core word when an action attribute in the standard question is not a core word;
data attribute determination rule: marking an object of an action attribute of the standard question as a data attribute; marking the participles which have a main and subordinate relation with the action attribute in the standard problem as data attributes;
determining rules of fixed language subjects: marking the adjectives or adverbs and other name word modifiers for modifying the action attributes in the standard questions as the shape words; and marking the adjectives or adverbs and other name word modifiers for modifying the data attributes in the standard questions as fixed words.
Further, obtaining the corpus includes:
acquiring the standard problem and an intention label corresponding to the standard problem;
constructing a similar problem to the standard problem, comprising the steps of:
performing word segmentation and part-of-speech tagging on the standard problem, and extracting nouns, verbs and individual words of the standard problem;
searching synonyms of the nouns and the verbs in a universal synonym dictionary, and sequentially and circularly replacing the corresponding nouns and the corresponding verbs through the searched synonyms to obtain a plurality of new sentences;
scoring the new sentences through a language model;
and sequentially replacing the individual words in the new sentences with N positions before grading and sorting into synonyms in the individual synonym dictionary to obtain a plurality of similar problems of the standard problem.
Further, training a machine learning model through the training corpus to obtain an intention recognition classification model, comprising the following steps:
performing word segmentation, part of speech tagging and stop word filtering on the training corpus to obtain a preprocessed training corpus;
constructing the characteristics of the preprocessed training corpus, wherein the construction of the characteristics comprises the construction of custom characteristics, the construction of word characteristics, the construction of semantic characteristics and the construction of syntactic characteristics;
training a machine learning model through the training corpus after feature construction, and determining the weight of features fitted with the intention labels in the training corpus in the machine learning model;
and fixing the weight to obtain an intention recognition classification model.
Further, receiving a user question, and performing intention recognition on the user question through the intention recognition classification model to obtain an intention candidate set, comprising the following steps:
performing word segmentation, part of speech tagging and stop word filtering on the user problem to obtain a preprocessed user problem;
constructing the characteristics of the preprocessed user questions;
calculating a confidence value of each intention category of the features under a fixed weight corresponding to the features through the intention recognition model, sequentially sorting and outputting standard problems under the intention categories corresponding to the related confidence values as candidate problems according to the confidence values from large to small, and forming an intention candidate set, wherein the intention candidate set comprises a preset number of candidate problems and a confidence value corresponding to each candidate problem.
Further, based on the intention candidate set, performing multiple rounds of human-computer interaction through the industry semantic network, determining a standard question matched with the intention of the user question, inquiring an answer of the standard question matched with the intention of the user question from the interaction data source, and outputting the answer, wherein the method comprises the following steps:
when the confidence value of the candidate question with the maximum confidence value in the intention candidate set is larger than a first preset threshold value, inquiring the answer corresponding to the candidate question with the maximum confidence value from the interaction data source, and outputting the answer corresponding to the candidate question with the maximum confidence value;
when the confidence value of the candidate question with the maximum confidence value in the intention candidate set is smaller than a second preset threshold value, judging the user question as an unidentified question;
when the confidence value of the candidate question with the maximum confidence value in the intention candidate set is smaller than the first preset threshold value and larger than the second preset threshold value, determining semantic attribute information missing from the user question according to the industry semantic network and the intention candidate set:
if the semantic attribute information missing from the user question corresponds to a unique candidate word based on the industry semantic network and the intention candidate set, completing the semantic attribute information missing from the user question through the unique candidate word, and querying a completed candidate question matched with the user question from the intention candidate set, and if the semantic attribute information missing from the user question corresponds to the unique candidate word, querying and outputting an answer of the completed candidate question matched with the user question from the interactive data source;
if the semantic attribute information missing the user question corresponds to more than one candidate word based on the industry semantic network and the intention candidate set, completing the semantic attribute missing the user question by each candidate word to form a query question; receiving feedback answers of the inquiry questions to complement the semantic attribute information missing from the user questions to form new user questions, inquiring candidate questions matched with the new user questions in the intention candidate set, and inquiring and outputting answers of the candidate questions matched with the new user questions from the interactive data source if the candidate questions are inquired; and if all the missing semantic attribute information is complemented through multi-round interaction, and the matched candidate problem is not inquired from the intention candidate set, and prompt information which does not understand the user problem is returned.
Further, the method also comprises the following steps:
acquiring an interaction log of the human-computer interaction; extracting unidentified problems in the interaction log;
preprocessing the unidentified problem, and constructing the features of the preprocessed unidentified problem;
clustering the unidentified problems through a K-means clustering algorithm to obtain a plurality of problems of intention categories;
screening questions in the intention categories, and processing the questions in the intention categories, wherein the processing comprises:
comparing the plurality of intention categories with the intention labels of the training corpus, and if the comparison is successful, adding the questions with the same intention categories as the intention labels into the training corpus;
otherwise, adding intention categories or discarding the unidentified problems.
It is a further object of the present invention to provide an electronic device for performing one of the above objects, comprising a processor, a storage medium, and a computer program stored in the storage medium, which when executed by the processor implements the above human-computer interaction method based on semantic web and intent recognition.
It is a further object of the present invention to provide a computer readable storage medium storing one of the objects of the invention, on which a computer program is stored, which computer program, when being executed by a processor, realizes the above-mentioned human-computer interaction method based on semantic web and intent recognition.
Compared with the prior art, the invention has the beneficial effects that:
a semantic network is built and an intention recognition model is trained based on standard questions in the industry, multiple rounds of man-machine interaction can be carried out by combining the semantic network and the intention recognition model, attribute information influencing intention recognition is complemented, and the intention of the question is accurately recognized, so that the question-answering quality of man-machine interaction is improved, the customer experience is improved, and meanwhile, the workload of manual customer service is reduced.
Drawings
FIG. 1 is a flow chart of a human-computer interaction method of the present invention based on semantic Web and intent recognition;
FIG. 2 is a schematic diagram of an industry semantic network constructed in example 1;
fig. 3 is a block diagram of the electronic apparatus of embodiment 2.
Detailed Description
The present invention will now be described in more detail with reference to the accompanying drawings, in which the description of the invention is given by way of illustration and not of limitation. The various embodiments may be combined with each other to form other embodiments not shown in the following description.
Example 1
The embodiment provides a man-machine interaction method based on semantic web and intention recognition, which specifically comprises the following steps:
obtaining common problem solutions in the industry as an interactive data source;
performing semantic annotation on the standard questions in the common question answers to construct an industry semantic network;
acquiring a training corpus, wherein the training corpus comprises the standard problem, a similar problem of the standard problem and an intention label corresponding to the standard problem;
training a machine learning model through the training corpus to obtain an intention recognition classification model;
receiving a user question, and performing intention identification on the user question through the intention identification classification model to obtain an intention candidate set, wherein the intention candidate set comprises standard questions in a plurality of intention categories;
and performing multiple rounds of man-machine interaction through the industry semantic network based on the intention candidate set, determining a standard question matched with the intention of the user question, inquiring an answer of the standard question matched with the intention of the user question from the interaction data source, and outputting the answer.
In practical application, the machine learning models can be trained according to training corpuses of different industries to obtain corresponding intention recognition models, wherein the machine learning models can be any machine learning algorithm or deep learning classification algorithm, including but not limited to KNN, decision trees and the like. .
The man-machine interaction method based on the semantic network and the intention identification can be used for man-machine interaction in a specific field, and because the man-machine interaction in the specific field has high professional knowledge requirements on an intelligent customer service robot, the quality of common problem solutions (FAQ data sources) directly influences the quality of the man-machine interaction, so that standard problems in interactive data sources need to be accurately simplified, the embodiment requires that the FAQ standard problems do not contain redundant dialogues, and the embodiment requires that statements are simplified, if spoken language problems are that my credit cards are lost, corresponding standard problems are that the credit cards are missed, and spoken language problems that the credit cards are missed as similar problems of 'loss of credit cards' to be added into a data set. Therefore, before the human-computer interaction method based on the semantic web and the intention recognition is executed, a common problem solution (FAQ) in a specific industry needs to be collected as an interaction data source, and a problem in the common problem solution as the interaction data source is already processed into a standard problem in advance, so that the obtained common problem solution contains the standard problem, a similar problem of the standard problem and an answer corresponding to the standard problem.
According to language habits, the present embodiment specifies that the standard questions are composed of individuals, action attributes, data attributes, idioms, and idioms, each standard question must have the individual and action attributes, and the remaining attributes do not necessarily need to be.
In the embodiment, an industry semantic network is built through standard questions in the industry, an intention recognition classification model is trained, when a user question is received, the real intention of the user question can be predicted according to the intention recognition model, an intention candidate set composed of the standard questions is obtained, on the basis of the intention candidate set, the intention information lacking in the user question can be determined through the industry semantic network, multiple rounds of man-machine interaction can be performed on the basis of the lacking intention information, the standard question capable of expressing the real intention of the user question is determined from the intention candidate set, the answer of the standard question is inquired from an interaction data source, and the answer is output, so that the customer service is completed. The method of the embodiment has the advantages that the question-answering applicability of a specific field is realized, multiple rounds of man-machine interaction are performed on the basis of the semantic network and the intention recognition classification model, and the missing intention information can be supplemented, so that the real intention of the user question is determined, the question feedback is performed in time, the question-answering quality of the man-machine interaction is improved, and the problem that the interaction cannot be accurately guided due to the fact that the intention information cannot be extracted in the spoken question-answering process can be avoided.
Preferably, semantic annotation is performed on the standard questions in the common question answers, and an industry semantic network is constructed, including the following steps:
performing word segmentation, part-of-speech tagging and syntactic analysis on the standard problem according to a pre-constructed word segmentation dictionary to obtain a dependency syntactic relation among all the segmented words in the standard problem;
extracting semantic attributes of each participle in the standard problem according to a semantic labeling rule, wherein the semantic attributes comprise an individual, a idiom, an action attribute, a fixed language and a data attribute;
marking each participle in the standard problem according to the semantic attribute to obtain a semantic marking result of each participle in the standard problem;
according to the semantic annotation result, constructing a semantic network according to the following query sequence: querying individuals in the standard questions, and taking the individuals as nodes of the semantic network; inquiring action attributes corresponding to the individuals, and taking the action attributes as branch nodes of the individuals; inquiring the data attribute and the shape language corresponding to the action attribute, and respectively taking the inquired data attribute and the shape language as branch nodes of the action attribute; and inquiring a fixed language corresponding to the data attribute, and taking the inquired fixed language as a branch node of the data attribute.
In short, if the standard question queries the participle with the semantic attribute, the participle with the relevant semantic attribute is used as the branch of the semantic attribute.
Before the industry semantic network is constructed, a word segmentation dictionary needs to be constructed in advance, the word segmentation dictionary comprises conventional words, and a new word discovery algorithm needs to be adopted to process standard problems, determine composite new words in the standard problems, and add the composite new words into the word segmentation dictionary.
Specifically, the step of determining the compound new word by the new word discovery algorithm is as follows:
step 1: performing word segmentation on the standard problem, and storing an obtained word segmentation result into a dictionary tree;
step 2: counting word frequency of the word segmentation result by using the dictionary tree, and respectively calculating internal cohesion MI and left-right entropy of the word segmentation result by using a mutual information algorithm and an information entropy algorithm to obtain a word segmentation score (MI + min) (left entropy and right entropy);
step 3: and (3) sorting in descending order according to the score, filtering repeated words, finally, sequentially obtaining composite new words from front to back according to the sorting, and adding the obtained composite new words into a word segmentation dictionary.
The Mutual Information algorithm (MI) indicates whether two variables X and Y have a relationship and the strength of the relationship.
The internal information of a word fragment is defined as: taking the logarithm of the product of the probability of the word segment/the probability of the subsequence to obtain mutual information:
if there are multiple subsequences in the word segment, the mutual information of the multiple subsequences is accumulated as the final "polymerization degree", for example, "cinema" ═ electric + cinema "+" movie + hospital ".
Entropy represents a measure of the uncertainty of the random variable. The concrete expression is as follows: generally, let X be a random variable with a finite value (or X be a probability field of a finite number of discrete events), and the probability that X takes on value X is p (X), the entropy of X is defined as:
and the left-right entropy refers to the entropy of the left boundary and the entropy of the right boundary of the multi-word expression. Taking left entropy as an example, the information entropy is calculated for all possible words on the left of a word and word frequency, and then summed. The formula for the left and right entropy is as follows:
left entropy:
right entropy:
wherein, W represents the words of the entropy to be calculated, aW represents the words collocated on the left side of the words W, and Wb represents the words collocated on the right side of the words W.
In this embodiment, based on a pre-constructed word segmentation dictionary, a Viterbi word segmentation algorithm is adopted to segment a standard problem, a CLAWS algorithm or a VOLSUNGA algorithm is adopted to label the part of speech, left-most derivatives (left-most derivatives) in the PCFG and different rules probabilities are adopted to calculate all possible tree structure probabilities, and a tree corresponding to the maximum value is taken as the syntactic analysis result of the sentence.
Parts of speech include adjectives, adverbs, nouns, numerators, names of people, place names, verbs, quantifiers, prepositions, and the like.
The dependency syntactic relations obtained by syntactic analysis include a predicate relation (i-send), a move-guest relation (send-flower), a cross-guest relation (send-his), a preposed object (book-read), a bilingual (please-me), a middle relation (red-apple), a shape-middle relation (very-beautiful), a move-complement relation (do-complete), a parallel relation (mountain-sea), a mediate relation (in-interior), a left additional relation (sum-sea), a right additional relation (children-people), an independent structure (two single sentences are structurally independent from each other), and a core relation (core words of sentences).
Preferably, according to a semantic labeling rule, extracting semantic attributes of each participle in the standard problem, wherein the semantic attributes include an individual, a shape, an action attribute, a fixed language and a data attribute, and the method includes:
action attribute determination rule: if the core word is the first verb in the standard question, marking the core word as an action attribute; if the core word is a non-verb, searching a verb mark closest to the core word as an action attribute; if the core word is not the first verb, the verb which has a direct relation with the core word is searched and marked as an action attribute;
individual determination rules: when the number of the participles of the standard question is less than 3, marking the first unlabeled participle as an individual; when two unlabeled participles are not connected and have the closest distance, labeling the first unlabeled participle as an individual; when two unlabeled participles are connected and have a modification relation, merging the two unlabeled and connected participles and labeling as an individual;
word segmentation and combination rules: when the participles needing to be combined are connected and not marked as the semantic attributes and are not adjectives or adverbs, combining the fixed relation and the state-middle relation of the standard problem with the modified words; merging the participles parallel to the core word with the object; merging parallel objects in the standard questions; merging a core word and an object of the core word when an action attribute in the standard question is not a core word;
data attribute determination rule: marking an object of an action attribute of the standard question as a data attribute; marking the participles which have a main and subordinate relation with the action attribute in the standard problem as data attributes;
determining rules of fixed language subjects: marking the adjectives or adverbs and other name word modifiers for modifying the action attributes in the standard questions as the shape words; and marking the adjectives or adverbs and other name word modifiers for modifying the data attributes in the standard questions as fixed words.
Taking the construction of the semantic network of the bank field as an example, the standard problem and the semantic annotation result of the bank field are obtained through the semantic annotation process, and are shown in the following table:
and according to the semantic network construction process, constructing the semantic network corresponding to the semantic annotation result to obtain the industry semantic network shown in FIG. 2, wherein the industry semantic network can realize multi-part-of-speech, multi-intention recognition and context association. Multiple parts of speech, such as A card upgrading B card. The intent identifies, for example, how much I would like to look up each of my credit card line and points. The problem contains two intents: the credit card credit score inquiry and the credit card limit inquiry return a plurality of answers. Context association, for example, the first problem is: transacting what the A card needs, the second problem is: then the B card woolen is handled.
Preferably, the obtaining of the corpus comprises:
acquiring the standard problem and an intention label corresponding to the standard problem;
constructing a similar problem to the standard problem, comprising the steps of:
performing word segmentation and part-of-speech tagging on the standard problem, and extracting nouns, verbs and individual words of the standard problem;
searching synonyms of the nouns and the verbs in a universal synonym dictionary, and sequentially and circularly replacing the corresponding nouns and the corresponding verbs through the searched synonyms to obtain a plurality of new sentences;
scoring the new sentences through a language model;
and sequentially replacing the individual words in the new sentences with N positions before grading and sorting into the synonyms in the individual word synonym dictionary to obtain a plurality of similar problems of the standard problem.
In other embodiments of the present invention, when obtaining the corpus, the similar problem of the standard problem may not be constructed, and the corpus consisting of the standard problem, the manually maintained similar problem, and the intention label corresponding to the problem may be directly obtained, wherein the spoken language problem is directly maintained as the similar problem of the corresponding standard problem, and the manually maintained similar problem may be obtained.
Because the intention of the user problem needs to be determined in the human-computer interaction process, the intention recognition classification model is trained by combining the machine learning model in the intention recognition process, and the training corpus is expanded by expanding the similarity problem of the standard problem, so that a large number of training samples with the same intention exist in the training corpus, and the generalization capability of the intention recognition classification model obtained by training is improved.
The embodiment obtains the similarity problem of the standard problem by replacing part of the participles in the standard problem with synonyms in the synonym dictionary.
In the embodiment, before the standard problem is segmented, the synonym of the individual word in the standard problem is initialized, the individual synonym is loaded into the segmentation dictionary, and the customized individual synonym dictionary is constructed in advance according to the individual word and the synonym of the individual word.
In order to prevent the synonym permutation and combination from generating too many similar problems, other words outside the individual are extracted firstly, nouns and verbs in the standard problem are extracted firstly, synonyms of the nouns or verbs are inquired in a synonym dictionary, corresponding nouns and verbs in the standard problem are replaced to be combined into a plurality of new sentences, and when a plurality of synonyms exist in the synonym dictionary, corresponding nouns and verbs in the standard problem are replaced in a permutation and combination mode.
The language model selected in this embodiment uses an n-gram language model to score the new sentences in sequence, and the formula for the n-gram language model to calculate the probability of the new sentences is:
P(w)＝P(w1，w2，w3，w4，...wn)＝P(w1)P(w2|w1)P(w3|w1，w2)...P(wn|w1...wn-1)；
where w denotes the word segmentation result, wnRepresenting the nth word segmentation result. P (w)n|w1，…wn-1) Meaning that when the preceding n-1 word is w1,..wn-1In the case that the nth word is wnThe probability of (c).
When the n value of the n-gram is larger, the constraint force on the next word is stronger, but at the same time, the model is more complicated, and the problems are more, so in another embodiment of the present invention, a Bigram2 meta language model can be used, the Bigram2 meta language model is obtained by simplifying an n-gram speech model, and the formula for calculating the sentence probability is:
P(w)＝P(w1)P(w2|w1)...P(wn|wn-1)，
constructing a Bigram2 meta-language model by calculating a Maximum Likelihood estimation (Maximum Likelihood Estimate), wherein the calculation formula is as follows:
P(wn|wn-1)＝count(wn,wn-1)/count(wn-1)，count(wn-1) Denotes wn-1Number of occurrences in the text.
In this embodiment, a new sentence with N bits before the rank-descending ranking is taken, where the value of N can be determined according to actual situations, and generally N is 10.
And finally, sequentially replacing individuals in the N-bit new sentences before grading and sorting into synonyms in the synonym dictionary to obtain a plurality of similar problems of the standard problem.
It should be noted that, since the professional words in a specific professional field cannot be replaced by synonyms as individuals, in order to guarantee the accuracy of replacement, classification replacement is performed on the professional words.
Preferably, the training of the corpus to train a machine learning model to obtain an intention recognition classification model includes the following steps:
performing word segmentation, part of speech tagging and stop word filtering on the training corpus to obtain a preprocessed training corpus;
constructing the characteristics of the preprocessed training corpus, wherein the construction of the characteristics comprises the construction of custom characteristics, the construction of word characteristics, the construction of semantic characteristics and the construction of syntactic characteristics;
training a machine learning model through the training corpus after feature construction, and determining the weight of features fitted with the intention labels in the training corpus in the machine learning model;
and fixing the weight to obtain an intention recognition classification model.
The intention recognition classification model can classify the intentions of the user problems, is beneficial to subsequent business process processing, and improves the user service experience, for example, if the user wants to transact the credit card, the user classifies the intention of transacting the credit card, and if the user does logout, the user classifies the intention of logout.
And (3) carrying out feature construction on the preprocessed word segmentation, wherein the feature construction process is as follows:
constructing a custom feature based on the keywords of the custom intention category and the keyword combination;
constructing word characteristics based on word frequency statistics of each participle;
constructing semantic features based on word vectors of each participle, wherein the word vectors can be obtained through the existing corpus training or by using a trained open source word vector file;
and constructing syntactic characteristics based on the syntactic analysis result.
Preferably, the method for receiving the user question and performing intention recognition on the user question through the intention recognition classification model to obtain an intention candidate set comprises the following steps:
performing word segmentation, part of speech tagging and stop word filtering on the user problem to obtain a preprocessed user problem;
constructing the characteristics of the preprocessed user questions;
calculating a confidence value of each intention category of the features under a fixed weight corresponding to the features through the intention recognition model, sequentially sorting and outputting standard problems under the intention categories corresponding to the related confidence values as candidate problems according to the confidence values from large to small, and forming an intention candidate set, wherein the intention candidate set comprises a preset number of candidate problems and a confidence value corresponding to each candidate problem.
The intention candidate set is formed by a preset number of standard questions which are obtained by an intention recognition and classification model and most related to the intention of the user question, the larger the confidence value is, the closer the candidate question corresponding to the confidence value is to the intention of the user question, wherein the values of the preset number are not limited in the process, and can be set by self according to actual conditions, and generally 10 to 20 standard questions are selected to form the intention candidate set. In one embodiment of the present invention, a plurality of intention candidate sets with different numbers can be selected through the statistics of the test question corpus, and the value of the preset number is determined according to the accuracy rate of each intention candidate set including the real intention, for example, there are 100 test question corpora, only one intention identification result (intention category) is taken as the intention candidate set, wherein the number of test questions including the correct intention is 85, and the accuracy rate can reach 85%; if the first five intention recognition results are selected as the intention candidate set, and the number of test problems containing correct intentions is 93, the accuracy can reach 93%. In the actual use process, a plurality of intention recognition results can be taken relatively as a candidate set, and then the actual intention of the user is found by combining with an industry semantic network.
Preferably, based on the intention candidate set, performing multiple rounds of human-computer interaction through the industry semantic network, determining a standard question matched with the intention of the user question, inquiring an answer of the standard question matched with the intention of the user question from the interaction data source, and outputting the answer, and the method comprises the following steps:
when the confidence value of the candidate question with the maximum confidence value in the intention candidate set is larger than a first preset threshold value, inquiring the answer corresponding to the candidate question with the maximum confidence value from the interaction data source, and outputting the answer corresponding to the candidate question with the maximum confidence value;
when the confidence value of the standard problem with the maximum confidence value in the intention candidate set is smaller than a second preset threshold value, judging the user problem as an unidentified problem;
when the confidence value of the candidate question with the maximum confidence value in the intention candidate set is smaller than the first preset threshold value and larger than the second preset threshold value, determining semantic attribute information missing from the user question according to the industry semantic network and the intention candidate set:
if the semantic attribute information missing from the user question corresponds to a unique candidate word based on the industry semantic network and the intention candidate set, completing the semantic attribute information missing from the user question through the unique candidate word, and querying a completed candidate question matched with the user question from the intention candidate set, and if the semantic attribute information missing from the user question corresponds to the unique candidate word, querying and outputting an answer of the completed candidate question matched with the user question from the interactive data source;
if the semantic attribute information missing the user question corresponds to more than one candidate word based on the industry semantic network and the intention candidate set, completing the semantic attribute missing the user question by each candidate word to form a query question; receiving feedback answers of the inquiry questions to complement the semantic attribute information missing from the user questions to form new user questions, inquiring candidate questions matched with the new user questions in the intention candidate set, and inquiring and outputting answers of the candidate questions matched with the new user questions from the interactive data source if the candidate questions are inquired; and if all the missing semantic attribute information is complemented through multi-round interaction, and the matched candidate problem is not inquired from the intention candidate set, and prompt information which does not understand the user problem is returned.
The embodiment solves the intention completion problem when the user intention is not clear through the semantic network and the intention recognition classification model, namely, realizes multiple rounds of man-machine interaction in a specific field by utilizing an intention candidate set output by the constructed industry semantic network and the intention recognition classification model, and completes the missing attribute information by asking the user reversely when the user question lacks certain attribute information.
And when the confidence value of the candidate question with the closest intention (the candidate question with the largest confidence value) in the intention candidate set is larger than a first preset threshold value, the intention is clear, and the answer corresponding to the closest candidate question is directly inquired from the interactive data source and returned. And if the confidence value of the candidate question with the closest intention (the candidate question with the highest confidence value) is smaller than a second preset threshold value, indicating that the user question is not identified, and marking the user question as an unidentified question.
When the confidence value of the candidate question is between the first preset threshold and the second preset threshold, the intention of the user question is ambiguous, the semantic attributes lacking in the user question need to be inquired from the industry semantic network, the specific semantic attribute information of the lacking semantic attributes is extracted from the candidate questions in the intention candidate set, and the true intention of the user question is determined through interaction by questioning (outputting the question) what the semantic attribute information lacking in the user is in combination with the specific semantic attribute information.
For example: the user question is 'I want to inquire', and about the intention of 'inquiring', more than one candidate words in the business semantic network identify the intention of the user question of the classification model based on the intention, and the intention category of more than one 'inquiring' necessarily exists in the acquired intention candidate set. At the moment, an inquiry question is output to the user according to the candidate words and the intention candidate set so as to clarify the intention of the user question, and as the inquiry is an action attribute, all corresponding individuals are determined in the business semantic network, and then the inquiry candidate set is matched with the individuals in the candidate sentences, an inquiry question can be generated, namely asking for inquiring about the gas fee, the meter reading time or the business hall address? "; and feeding back answers to the question by the user, thereby completing the missing individual information of the query, realizing the positioning of the question intentions of the user, forming standard questions with clear intentions, matching candidate questions consistent with the standard questions with clear intentions from the intention candidate set, and querying and returning the answers of the candidate questions from an interactive data source.
In other embodiments, the user may be further required to complete other semantic attribute information in sequence to form a standard problem with clear intent, generally, the semantic attribute information is completed in sequence according to the importance degree, and the completion sequence is as follows: individual, action attribute. And matching the intention candidate set once every time the zhuge, the data attribute and the fixed language are completed, wherein if the corresponding candidate sentence is matched, the intention is clear.
The values of the first preset threshold and the second preset threshold can be determined according to actual conditions, but the first preset threshold is larger than the second preset threshold, the first threshold is generally 0.8-0.9, and the second threshold is generally 0.2-0.3. In practical applications, the threshold may be determined by testing the confidence and accuracy of the first standard question identified by the corpus statistical intent of the question. For example, if there are 100 test questions related to the service, and 85 questions among them are correctly intended, and the confidence level of the 85 questions after intention recognition is in the range of 0.8 to 0.9, the mean value of the confidence level may be used as the first preset threshold. Similarly, the intention recognition is performed by using a problem unrelated to the service, and a value of the second preset threshold value can be obtained.
Further, the method also comprises the following steps:
acquiring an interaction log of the human-computer interaction; extracting unidentified problems in the interactive log to obtain an unidentified problem set;
preprocessing the unidentified problem, and constructing the features of the preprocessed unidentified problem;
clustering the unidentified problems through a K-means clustering algorithm to obtain a plurality of problems of intention categories;
screening questions in the intention categories, and processing the questions in the intention categories, wherein the processing comprises:
comparing the plurality of intention categories with the intention labels of the training corpus, and if the comparison is successful, adding the questions with the same intention categories as the intention labels into the training corpus;
otherwise, adding intention categories or discarding the unidentified problems.
Because the corpus is a similar problem of generating a standard problem by adopting a synonym replacement mode, the sentence pattern is single, user problems which cannot be identified exist in the interaction process, the user problems are summarized and summarized in the interaction process in a text clustering mode, and the corpus is adjusted, so that the corpus of the intention identification is enhanced, and the generalization capability of the intention identification classification model is further improved.
In other embodiments of the present invention, obviously meaningless questions in the unrecognized questions may be manually filtered to obtain an unrecognized question set, and the preprocessing of the unrecognized questions in this embodiment includes word segmentation and stop word filtering. And clustering the unidentified problems by using a K-means clustering algorithm to obtain K intention category problems, wherein the value of K is the optimal clustering number, and the K is obtained by traversing by using the contour coefficient as an index. Of course, in other implementations, the K-means clustering algorithm may also use any other clustering algorithm to implement the clustering of unidentified problems. And obtaining a clustering result of unidentified problems, comparing whether the clustering intention type is a business type in the industry or not, if not, discarding the unidentified problems under the clustering type, if so, comparing whether each clustering type is consistent with the intention type contained in the training corpus, if so, adding the unidentified problems with complete intention description under the corresponding clustering type into the training corpus to serve as similar problems of the original standard problems under the corresponding intention labels, if not, maintaining answers, summarizing new standard problems, and adding the problems with complete intention description under the clustering type to serve as similar problems of the new standard problems to the training corpus and an interactive data source. And performing iterative training on the intention recognition classification model through the adjusted training corpus. Of course, in some embodiments of the present invention, a manual judgment mode may also be adopted to judge, classify and discard the clustering result.
Example 2
Fig. 3 is a schematic structural diagram of an electronic device according to embodiment 2 of the present invention, as shown in fig. 3, the electronic device includes a processor 310, a memory 320, an input device 330, and an output device 340; the number of the processors 310 in the computer device may be one or more, and one processor 310 is taken as an example in fig. 3; the processor 310, the memory 320, the input device 330 and the output device 340 in the electronic apparatus may be connected by a bus or other means, and the connection by the bus is exemplified in fig. 3.
The memory 320 is a computer-readable storage medium for storing software programs, computer-executable programs, and modules, such as program instructions/modules corresponding to the semantic Web and intent recognition based human-computer interaction method in the embodiments of the present invention. The processor 310 executes various functional applications and data processing of the electronic device, i.e., implementing the man-machine interaction method based on semantic web and intention recognition of embodiment 1, by executing software programs, instructions and modules stored in the memory 320.
The memory 320 may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application program required for at least one function; the storage data area may store data created according to the use of the terminal, and the like. Further, the memory 320 may include high speed random access memory, and may also include non-volatile memory, such as at least one magnetic disk storage device, flash memory device, or other non-volatile solid state storage device. In some examples, the memory 320 may further include memory located remotely from the processor 310, which may be connected to the electronic device through a network. Examples of such networks include, but are not limited to, the internet, intranets, local area networks, mobile communication networks, and combinations thereof.
The input device 330 may receive input data or requests, etc. The output device 340 may output and display data.
Example 3
Embodiment 3 of the present invention further provides a storage medium containing computer-executable instructions, which when executed by a computer processor, are configured to implement a human-computer interaction method based on semantic web and intent recognition, the method including:
obtaining common problem solutions in the industry as an interactive data source;
performing semantic annotation on the standard questions in the common question answers to construct an industry semantic network;
acquiring a training corpus, wherein the training corpus comprises the standard problem, a similar problem of the standard problem and an intention label corresponding to the standard problem;
training a machine learning model through the training corpus to obtain an intention recognition classification model;
receiving a user question, and performing intention identification on the user question through the intention identification classification model to obtain an intention candidate set, wherein the intention candidate set comprises standard questions in a plurality of intention categories;
and performing multiple rounds of man-machine interaction through the industry semantic network based on the intention candidate set, determining a standard question matched with the intention of the user question, inquiring an answer of the standard question matched with the intention of the user question from the interaction data source, and outputting the answer.
Of course, the storage medium containing the computer-executable instructions provided by the embodiments of the present invention is not limited to the method operations described above, and may also perform related operations in the human-computer interaction method based on semantic web and intent recognition provided by any embodiments of the present invention.
From the above description of the embodiments, it is obvious for those skilled in the art that the present invention can be implemented by software and necessary general hardware, and certainly, can also be implemented by hardware, but the former is a better embodiment in many cases. Based on such understanding, the technical solutions of the present invention may be embodied in the form of a software product, which may be stored in a computer-readable storage medium, such as a floppy disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a FLASH Memory (FLASH), a hard disk or an optical disk of a computer, and includes instructions for enabling an electronic device (which may be a mobile phone, a personal computer, a server, or a network device) to execute the methods according to the embodiments of the present invention.
It should be noted that, in the embodiment of the human-computer interaction method or apparatus based on semantic web and intention recognition, the included units and modules are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be realized; in addition, specific names of the functional units are only for convenience of distinguishing from each other, and are not used for limiting the protection scope of the present invention.
Various other modifications and changes may be made by those skilled in the art based on the above-described technical solutions and concepts, and all such modifications and changes should fall within the scope of the claims of the present invention.
Claims (10)
1. A man-machine interaction method based on semantic web and intention recognition is characterized by comprising the following steps:
obtaining common problem solutions in the industry as an interactive data source;
performing semantic annotation on the standard questions in the common question answers to construct an industry semantic network;
acquiring a training corpus, wherein the training corpus comprises the standard problem, a similar problem of the standard problem and an intention label corresponding to the standard problem;
training a machine learning model through the training corpus to obtain an intention recognition classification model;
receiving a user question, and performing intention identification on the user question through the intention identification classification model to obtain an intention candidate set, wherein the intention candidate set comprises standard questions in a plurality of intention categories;
and performing multiple rounds of man-machine interaction through the industry semantic network based on the intention candidate set, determining a standard question matched with the intention of the user question, inquiring an answer of the standard question matched with the intention of the user question from the interaction data source, and outputting the answer.
2. The human-computer interaction method based on semantic web and intention recognition, as claimed in claim 1, wherein the standard questions in the common question solution are semantically labeled to construct an industry semantic network, comprising the following steps:
performing word segmentation, part-of-speech tagging and syntactic analysis on the standard problem according to a pre-constructed word segmentation dictionary to obtain a dependency syntactic relation among all the segmented words in the standard problem;
extracting semantic attributes of each participle in the standard problem according to a semantic labeling rule, wherein the semantic attributes comprise an individual, a idiom, an action attribute, a fixed language and a data attribute;
marking each participle in the standard problem according to the semantic attribute to obtain a semantic marking result of each participle in the standard problem;
according to the semantic annotation result, constructing a semantic network according to the following query sequence: querying individuals in the standard questions, and taking the individuals as nodes of the semantic network; inquiring action attributes corresponding to the individuals, and taking the action attributes as branch nodes of the individuals; inquiring the data attribute and the shape language corresponding to the action attribute, and respectively taking the inquired data attribute and the shape language as branch nodes of the action attribute; and inquiring a fixed language corresponding to the data attribute, and taking the inquired fixed language as a branch node of the data attribute.
3. The human-computer interaction method based on semantic web and intention recognition according to claim 2, wherein the semantic labeling rule comprises:
action attribute determination rule: if the core word is the first verb in the standard question, marking the core word as an action attribute; if the core word is a non-verb, searching a verb mark closest to the core word as an action attribute; if the core word is not the first verb, the verb which has a direct relation with the core word is searched and marked as an action attribute;
individual determination rules: when the number of the participles of the standard question is less than 3, marking the first unlabeled participle as an individual; when two unlabeled participles are not connected and have the closest distance, labeling the first unlabeled participle as an individual; when two unlabeled participles are connected and have a modification relation, merging the two unlabeled and connected participles and labeling as an individual;
word segmentation and combination rules: when the participles needing to be combined are connected and not marked as the semantic attributes and are not adjectives or adverbs, combining the fixed relation and the state-middle relation of the standard problem with the modified words; merging the participles parallel to the core word with the object; merging parallel objects in the standard questions; merging a core word and an object of the core word when an action attribute in the standard question is not a core word;
a data attribute determination rule; marking an object of an action attribute of the standard question as a data attribute; marking the participles which have a main and subordinate relation with the action attribute in the standard problem as data attributes;
determining rules of fixed language; marking the adjectives or adverbs and other name word modifiers for modifying the action attributes in the standard questions as the shape words; and marking the adjectives or adverbs and other name word modifiers for modifying the data attributes in the standard questions as fixed words.
4. The human-computer interaction method based on semantic web and intention recognition according to claim 1, wherein the obtaining of the training corpus comprises:
acquiring the standard problem and an intention label corresponding to the standard problem;
constructing a similar problem to the standard problem, comprising the steps of:
performing word segmentation and part-of-speech tagging on the standard problem, and extracting nouns, verbs and individual words of the standard problem;
searching synonyms of the nouns and the verbs in a universal synonym dictionary, and sequentially and circularly replacing the corresponding nouns and the corresponding verbs through the searched synonyms to obtain a plurality of new sentences;
scoring the new sentences through a language model;
and sequentially replacing the individual words in the new sentences with N positions before grading and sorting into synonyms in the individual synonym dictionary to obtain a plurality of similar problems of the standard problem.
5. The human-computer interaction method based on semantic web and intention recognition according to claim 4, wherein the training corpus is used for training a machine learning model to obtain an intention recognition classification model, and the method comprises the following steps:
performing word segmentation, part of speech tagging and stop word filtering on the training corpus to obtain a preprocessed training corpus;
constructing the characteristics of the preprocessed training corpus, wherein the construction of the characteristics comprises the construction of custom characteristics, the construction of word characteristics, the construction of semantic characteristics and the construction of syntactic characteristics;
training a machine learning model through the training corpus after feature construction, and determining the weight of features fitted with the intention labels in the training corpus in the machine learning model;
and fixing the weight to obtain an intention recognition classification model.
6. The human-computer interaction method based on semantic web and intention recognition of claim 5, wherein receiving a user question, and performing intention recognition on the user question through the intention recognition classification model to obtain an intention candidate set, comprises the following steps:
performing word segmentation, part of speech tagging and stop word filtering on the user problem to obtain a preprocessed user problem;
constructing the characteristics of the preprocessed user questions;
calculating a confidence value of each intention category of the features under a fixed weight corresponding to the features through the intention recognition model, sequentially sorting and outputting standard problems under the intention categories corresponding to the related confidence values as candidate problems according to the confidence values from large to small to form an intention candidate set, wherein the intention candidate set comprises a preset number of candidate problems and a confidence value corresponding to each candidate problem.
7. The human-computer interaction method based on semantic web and intention recognition as claimed in claim 6, wherein based on the intention candidate set, multiple rounds of human-computer interaction are performed through the industry semantic web, a standard question matched with the intention of the user question is determined, an answer of the standard question matched with the intention of the user question is inquired from the interaction data source, and the answer is output, and the method comprises the following steps:
when the confidence value of the candidate question with the maximum confidence value in the intention candidate set is larger than a first preset threshold value, inquiring the answer corresponding to the candidate question with the maximum confidence value from the interaction data source, and outputting the answer corresponding to the candidate question with the maximum confidence value;
when the confidence value of the candidate question with the maximum confidence value in the intention candidate set is smaller than a second preset threshold value, judging the user question as an unidentified question;
when the confidence value of the candidate question with the maximum confidence value in the intention candidate set is smaller than the first preset threshold value and larger than the second preset threshold value, determining semantic attribute information missing from the user question according to the industry semantic network and the intention candidate set:
if the semantic attribute information missing from the user question corresponds to a unique candidate word based on the industry semantic network and the intention candidate set, completing the semantic attribute information missing from the user question through the unique candidate word, and querying a completed candidate question matched with the user question from the intention candidate set, and if the semantic attribute information missing from the user question corresponds to the unique candidate word, querying and outputting an answer of the completed candidate question matched with the user question from the interactive data source;
if the semantic attribute information missing the user question corresponds to more than one candidate word based on the industry semantic network and the intention candidate set, completing the semantic attribute missing the user question by each candidate word to form a query question; receiving feedback answers of the inquiry questions to complement the semantic attribute information missing from the user questions to form new user questions, inquiring candidate questions matched with the new user questions in the intention candidate set, and inquiring and outputting answers of the candidate questions matched with the new user questions from the interactive data source if the candidate questions are inquired; and if all the missing semantic attribute information is complemented through multi-round interaction, and the matched candidate problem is not inquired from the intention candidate set, and prompt information which does not understand the user problem is returned.
8. The human-computer interaction method based on semantic web and intention recognition according to claim 1 or 7, further comprising the steps of:
acquiring an interaction log of the human-computer interaction; extracting unidentified problems in the interaction log;
preprocessing the unidentified problem, and constructing the features of the preprocessed unidentified problem;
clustering the unidentified problems through a K-means clustering algorithm to obtain a plurality of problems of intention categories;
screening questions in the intention categories, and processing the questions in the intention categories, wherein the processing comprises:
comparing the plurality of intention categories with the intention labels of the training corpus, and if the comparison is successful, adding the questions with the same intention categories as the intention labels into the training corpus;
otherwise, adding intention categories or discarding the unidentified problems.
9. An electronic device comprising a processor, a storage medium, and a computer program stored in the storage medium, wherein the computer program, when executed by the processor, implements the human-computer interaction method based on semantic web and intent recognition according to any one of claims 1 to 8.
10. A computer-readable storage medium, on which a computer program is stored, which, when being executed by a processor, implements the human-computer interaction method based on semantic web and intent recognition according to any one of claims 1 to 8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010756664.6A CN112069298A (en) | 2020-07-31 | 2020-07-31 | Human-computer interaction method, device and medium based on semantic web and intention recognition |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010756664.6A CN112069298A (en) | 2020-07-31 | 2020-07-31 | Human-computer interaction method, device and medium based on semantic web and intention recognition |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112069298A true CN112069298A (en) | 2020-12-11 |
Family
ID=73657762
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010756664.6A Pending CN112069298A (en) | 2020-07-31 | 2020-07-31 | Human-computer interaction method, device and medium based on semantic web and intention recognition |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112069298A (en) |
Cited By (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112487810A (en) * | 2020-12-17 | 2021-03-12 | 税友软件集团股份有限公司 | Intelligent customer service method, device, equipment and storage medium |
| CN112632991A (en) * | 2020-12-30 | 2021-04-09 | 北京久其软件股份有限公司 | Method and device for extracting feature information of Chinese language |
| CN112632220A (en) * | 2020-12-22 | 2021-04-09 | 平安普惠企业管理有限公司 | Loan fund destination analysis method based on AI model and related equipment |
| CN112650846A (en) * | 2021-01-13 | 2021-04-13 | 北京智通云联科技有限公司 | Question-answer intention knowledge base construction system and method based on question frame |
| CN112800778A (en) * | 2021-02-07 | 2021-05-14 | 北京智通云联科技有限公司 | Intention identification method and system based on word string length and storage medium |
| CN113095724A (en) * | 2021-04-30 | 2021-07-09 | 中国银行股份有限公司 | Data processing method and device |
| CN113486182A (en) * | 2021-08-02 | 2021-10-08 | 上海明略人工智能(集团)有限公司 | Intention identification method, system, electronic equipment and readable storage medium |
| CN113569578A (en) * | 2021-08-13 | 2021-10-29 | 上海淇玥信息技术有限公司 | User intention identification method and device and computer equipment |
| CN113571145A (en) * | 2021-08-11 | 2021-10-29 | 四川医枢科技有限责任公司 | Information matching method and system for health management decision |
| CN113590790A (en) * | 2021-07-30 | 2021-11-02 | 北京壹心壹翼科技有限公司 | Question retrieval method, device, equipment and medium applied to multiple rounds of question answering |
| CN113742469A (en) * | 2021-09-03 | 2021-12-03 | 科讯嘉联信息技术有限公司 | Pipeline processing and ES storage based question-answering system construction method |
| CN115238060A (en) * | 2022-09-20 | 2022-10-25 | 支付宝(杭州)信息技术有限公司 | Man-machine interaction method and device, medium and computing equipment |
| CN115270786A (en) * | 2022-09-27 | 2022-11-01 | 炫我信息技术(北京)有限公司 | Method, device and equipment for identifying question intention and readable storage medium |
| CN116383367A (en) * | 2023-06-06 | 2023-07-04 | 北京啄木鸟云健康科技有限公司 | Data processing method, device, equipment and medium for cold start stage of dialogue system |
| CN117765949A (en) * | 2024-02-22 | 2024-03-26 | 青岛海尔科技有限公司 | semantic dependency analysis-based statement multi-intention recognition method and device |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105589844A (en) * | 2015-12-18 | 2016-05-18 | 北京中科汇联科技股份有限公司 | Missing semantic supplementing method for multi-round question-answering system |
| CN111400453A (en) * | 2020-03-16 | 2020-07-10 | 上海百事通信息技术股份有限公司 | Intelligent interaction system and method based on legal consultation |
-
2020
- 2020-07-31 CN CN202010756664.6A patent/CN112069298A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105589844A (en) * | 2015-12-18 | 2016-05-18 | 北京中科汇联科技股份有限公司 | Missing semantic supplementing method for multi-round question-answering system |
| CN111400453A (en) * | 2020-03-16 | 2020-07-10 | 上海百事通信息技术股份有限公司 | Intelligent interaction system and method based on legal consultation |
Cited By (25)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112487810A (en) * | 2020-12-17 | 2021-03-12 | 税友软件集团股份有限公司 | Intelligent customer service method, device, equipment and storage medium |
| CN112487810B (en) * | 2020-12-17 | 2022-03-11 | 税友软件集团股份有限公司 | Intelligent customer service method, device, equipment and storage medium |
| CN112632220A (en) * | 2020-12-22 | 2021-04-09 | 平安普惠企业管理有限公司 | Loan fund destination analysis method based on AI model and related equipment |
| CN112632991A (en) * | 2020-12-30 | 2021-04-09 | 北京久其软件股份有限公司 | Method and device for extracting feature information of Chinese language |
| CN112632991B (en) * | 2020-12-30 | 2024-05-14 | 北京久其软件股份有限公司 | Method and device for extracting characteristic information of Chinese language |
| CN112650846A (en) * | 2021-01-13 | 2021-04-13 | 北京智通云联科技有限公司 | Question-answer intention knowledge base construction system and method based on question frame |
| CN112800778B (en) * | 2021-02-07 | 2023-07-18 | 北京智通云联科技有限公司 | Intent recognition method, system and storage medium based on word string length |
| CN112800778A (en) * | 2021-02-07 | 2021-05-14 | 北京智通云联科技有限公司 | Intention identification method and system based on word string length and storage medium |
| CN113095724A (en) * | 2021-04-30 | 2021-07-09 | 中国银行股份有限公司 | Data processing method and device |
| CN113095724B (en) * | 2021-04-30 | 2024-03-01 | 中国银行股份有限公司 | Data processing method and device |
| CN113590790A (en) * | 2021-07-30 | 2021-11-02 | 北京壹心壹翼科技有限公司 | Question retrieval method, device, equipment and medium applied to multiple rounds of question answering |
| CN113590790B (en) * | 2021-07-30 | 2023-11-28 | 北京壹心壹翼科技有限公司 | Question retrieval method, device, equipment and medium applied to multi-round question and answer |
| CN113486182A (en) * | 2021-08-02 | 2021-10-08 | 上海明略人工智能(集团)有限公司 | Intention identification method, system, electronic equipment and readable storage medium |
| CN113571145A (en) * | 2021-08-11 | 2021-10-29 | 四川医枢科技有限责任公司 | Information matching method and system for health management decision |
| CN113569578B (en) * | 2021-08-13 | 2024-03-08 | 上海淇玥信息技术有限公司 | User intention recognition method and device and computer equipment |
| CN113569578A (en) * | 2021-08-13 | 2021-10-29 | 上海淇玥信息技术有限公司 | User intention identification method and device and computer equipment |
| CN113742469A (en) * | 2021-09-03 | 2021-12-03 | 科讯嘉联信息技术有限公司 | Pipeline processing and ES storage based question-answering system construction method |
| CN113742469B (en) * | 2021-09-03 | 2023-12-15 | 科讯嘉联信息技术有限公司 | Method for constructing question-answering system based on Pipeline processing and ES storage |
| CN115238060B (en) * | 2022-09-20 | 2022-12-27 | 支付宝(杭州)信息技术有限公司 | Human-computer interaction method and device, medium and computing equipment |
| CN115238060A (en) * | 2022-09-20 | 2022-10-25 | 支付宝(杭州)信息技术有限公司 | Man-machine interaction method and device, medium and computing equipment |
| CN115270786A (en) * | 2022-09-27 | 2022-11-01 | 炫我信息技术(北京)有限公司 | Method, device and equipment for identifying question intention and readable storage medium |
| CN116383367A (en) * | 2023-06-06 | 2023-07-04 | 北京啄木鸟云健康科技有限公司 | Data processing method, device, equipment and medium for cold start stage of dialogue system |
| CN116383367B (en) * | 2023-06-06 | 2023-08-11 | 北京啄木鸟云健康科技有限公司 | Data processing method, device, equipment and medium for cold start stage of dialogue system |
| CN117765949A (en) * | 2024-02-22 | 2024-03-26 | 青岛海尔科技有限公司 | semantic dependency analysis-based statement multi-intention recognition method and device |
| CN117765949B (en) * | 2024-02-22 | 2024-05-24 | 青岛海尔科技有限公司 | Semantic dependency analysis-based statement multi-intention recognition method and device |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112069298A (en) | Human-computer interaction method, device and medium based on semantic web and intention recognition | |
| CN108804521B (en) | Knowledge graph-based question-answering method and agricultural encyclopedia question-answering system | |
| CN108363790B (en) | Method, device, equipment and storage medium for evaluating comments | |
| CN108304372B (en) | Entity extraction method and device, computer equipment and storage medium | |
| CN110222045B (en) | Data report acquisition method and device, computer equipment and storage medium | |
| WO2021093755A1 (en) | Matching method and apparatus for questions, and reply method and apparatus for questions | |
| CN110727779A (en) | Question-answering method and system based on multi-model fusion | |
| CN111241237B (en) | Intelligent question-answer data processing method and device based on operation and maintenance service | |
| CN110347787B (en) | Interview method and device based on AI auxiliary interview scene and terminal equipment | |
| CN112035599B (en) | Query method and device based on vertical search, computer equipment and storage medium | |
| CN109271524B (en) | Entity linking method in knowledge base question-answering system | |
| WO2017091985A1 (en) | Method and device for recognizing stop word | |
| CN112699645B (en) | Corpus labeling method, apparatus and device | |
| CN115470338B (en) | Multi-scenario intelligent question answering method and system based on multi-path recall | |
| CN109829045A (en) | A kind of answering method and device | |
| CN108287848B (en) | Method and system for semantic parsing | |
| CN109783806A (en) | A kind of text matching technique using semantic analytic structure | |
| CN113821605A (en) | Event extraction method | |
| CN111325033B (en) | Entity identification method, entity identification device, electronic equipment and computer readable storage medium | |
| CN114266256A (en) | Method and system for extracting new words in field | |
| CN111159381B (en) | Data searching method and device | |
| CN114003709A (en) | Intelligent question-answering system and method based on question matching | |
| CN110795942B (en) | Keyword determination method and device based on semantic recognition and storage medium | |
| CN109614493B (en) | Text abbreviation recognition method and system based on supervision word vector | |
| CN116628173B (en) | Intelligent customer service information generation system and method based on keyword extraction |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| CB02 | Change of applicant information |
Country or region after: China Address after: 23 / F, World Trade Center, 857 Xincheng Road, Binjiang District, Hangzhou City, Zhejiang Province, 310051 Applicant after: Hangzhou Yuanchuan Xinye Technology Co.,Ltd. Address before: 23 / F, World Trade Center, 857 Xincheng Road, Binjiang District, Hangzhou City, Zhejiang Province, 310051 Applicant before: Hangzhou Yuanchuan New Technology Co.,Ltd. Country or region before: China |
|
| CB02 | Change of applicant information |