CN112699909B - Information identification method, information identification device, electronic equipment and computer readable storage medium - Google Patents
Information identification method, information identification device, electronic equipment and computer readable storage medium Download PDFInfo
- Publication number
- CN112699909B CN112699909B CN201911012568.4A CN201911012568A CN112699909B CN 112699909 B CN112699909 B CN 112699909B CN 201911012568 A CN201911012568 A CN 201911012568A CN 112699909 B CN112699909 B CN 112699909B
- Authority
- CN
- China
- Prior art keywords
- concept
- class
- information
- feature vector
- identified
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 69
- 239000013598 vector Substances 0.000 claims abstract description 216
- 238000012549 training Methods 0.000 claims description 52
- 238000000605 extraction Methods 0.000 claims description 39
- 238000012937 correction Methods 0.000 claims description 12
- 238000004590 computer program Methods 0.000 claims description 8
- 238000013473 artificial intelligence Methods 0.000 description 35
- 238000013135 deep learning Methods 0.000 description 27
- 230000008569 process Effects 0.000 description 24
- 238000005516 engineering process Methods 0.000 description 16
- 241000282326 Felis catus Species 0.000 description 14
- 238000013507 mapping Methods 0.000 description 13
- 238000013528 artificial neural network Methods 0.000 description 11
- 241000282472 Canis lupus familiaris Species 0.000 description 8
- 238000004364 calculation method Methods 0.000 description 8
- 244000025254 Cannabis sativa Species 0.000 description 6
- 238000010586 diagram Methods 0.000 description 6
- 238000007477 logistic regression Methods 0.000 description 6
- 238000012545 processing Methods 0.000 description 6
- 241000282320 Panthera leo Species 0.000 description 5
- 230000006870 function Effects 0.000 description 5
- 230000006978 adaptation Effects 0.000 description 4
- 238000002372 labelling Methods 0.000 description 4
- 238000013527 convolutional neural network Methods 0.000 description 3
- 230000008878 coupling Effects 0.000 description 3
- 238000010168 coupling process Methods 0.000 description 3
- 238000005859 coupling reaction Methods 0.000 description 3
- 238000004422 calculation algorithm Methods 0.000 description 2
- 238000004891 communication Methods 0.000 description 2
- 238000001514 detection method Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 230000003993 interaction Effects 0.000 description 2
- 239000000463 material Substances 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 238000012706 support-vector machine Methods 0.000 description 2
- 230000004931 aggregating effect Effects 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 125000004122 cyclic group Chemical group 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000002093 peripheral effect Effects 0.000 description 1
- 230000000306 recurrent effect Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- General Health & Medical Sciences (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention provides an information identification method, an information identification device, electronic equipment and a computer readable storage medium, wherein the information identification method comprises the following steps: comparing the feature vector of the information to be identified with the comparison feature vector of each first concept class in the concept knowledge base to obtain a similarity judgment value of each first concept class; obtaining a target similarity judgment value from the similarity judgment values of the first concept classes; the target similarity judgment value is the highest similarity judgment value in the similarity judgment values of the first concept classes; if the target similarity judgment value is larger than a first preset threshold value, determining a first concept class corresponding to the target comparison feature vector as the concept class of the information to be identified; the target comparison feature vector is a comparison feature vector corresponding to the target similarity judgment value; if the target similarity judgment value is smaller than a second preset threshold value, creating a second concept class in the concept knowledge base, and determining the second concept class as the concept class of the information to be identified. The embodiment of the invention can improve the applicability of the multi-task.
Description
Technical Field
The embodiment of the invention relates to the technical field of artificial intelligence, in particular to an information identification method, an information identification device, electronic equipment and a computer readable storage medium.
Background
The artificial intelligence technology can realize the automatic detection and identification of various objects and can improve the social production efficiency, so that people pay attention to the technology. At present, the artificial intelligence technology mainly comprises technologies such as deep learning, knowledge graph and the like, the deep learning technology realizes extraction and identification of sample characteristics through neural network training through a large number of sample sets, and the knowledge graph technology realizes labeling, reasoning and inquiring of examples based on ontology definition by constructing an ontology.
In the related art, the main problem of the deep learning technology is that a large number of sample sets are needed for training, and a large amount of manpower and material resources are consumed for actually marking the samples; in addition, deep learning often requires specialized task training when performing network training, and different tasks often require different samples and networks to retrain.
Therefore, the information identification system based on artificial intelligence in the prior art has the problem of poor multi-task applicability.
Disclosure of Invention
The embodiment of the invention provides an information identification method, an information identification device, electronic equipment and a computer readable storage medium, which are used for solving the problem that an artificial intelligence-based information identification system in the prior art is poor in task applicability.
In a first aspect, an embodiment of the present invention provides an information identifying method, including:
comparing the obtained feature vector of the information to be identified with the comparison feature vector of each first concept class in the concept knowledge base to obtain a similarity judgment value of each first concept class;
obtaining a target similarity judgment value from the similarity judgment value of each first concept class; the target similarity judgment value is the highest similarity judgment value in the similarity judgment values of the first concept classes;
if the target similarity judgment value is larger than a first preset threshold value, determining a first concept class corresponding to the target comparison feature vector as the concept class of the information to be identified; the target comparison feature vector is a comparison feature vector corresponding to the target similarity judgment value;
if the target similarity judgment value is smaller than a second preset threshold value, creating a second concept class in the concept knowledge base, and determining the second concept class as the concept class of the information to be identified.
In a second aspect, an embodiment of the present invention provides an information identifying apparatus, including:
the feature comparison module is used for comparing the obtained feature vector of the information to be identified with the comparison feature vector of each first concept class in the concept knowledge base to obtain a similarity judgment value of each first concept class;
the first acquisition module is used for acquiring a target similarity judgment value from the similarity judgment values of the first concept classes; the target similarity judgment value is the highest similarity judgment value in the similarity judgment values of the first concept classes;
the first determining module is used for determining a first concept class corresponding to the target comparison feature vector as the concept class of the information to be identified if the target similarity judging value is larger than a first preset threshold value; the target comparison feature vector is a comparison feature vector corresponding to the target similarity judgment value;
and the second determining module is used for creating a second concept class in the concept knowledge base and determining the second concept class as the concept class of the information to be identified if the target similarity judgment value is smaller than a second preset threshold value.
In a third aspect, an embodiment of the present invention provides an electronic device, including a processor, a memory, and a computer program stored in the memory and executable on the processor, where the computer program implements the steps of the information identifying method described above when executed by the processor.
In a fourth aspect, embodiments of the present invention provide a computer readable storage medium having stored thereon a computer program which, when executed by a processor, implements the steps of the above-described information identification method.
The embodiment of the invention provides an information identification method, an information identification device, electronic equipment and a computer readable storage medium, wherein firstly, an obtained feature vector of information to be identified is compared with a comparison feature vector of each first concept class in a concept knowledge base to obtain a similarity judgment value of each first concept class; then, obtaining a target similarity judgment value from the similarity judgment value of each first concept class; the target similarity judgment value is the highest similarity judgment value in the similarity judgment values of the first concept classes; finally, if the target similarity judgment value is larger than a first preset threshold value, determining a first concept class corresponding to the target comparison feature vector as the concept class of the information to be identified; the target comparison feature vector is a comparison feature vector corresponding to the target similarity judgment value; if the target similarity judgment value is smaller than a second preset threshold value, creating a second concept class in the concept knowledge base, and determining the second concept class as the concept class of the information to be identified.
In the embodiment of the invention, the deep learning technology and the knowledge graph technology are combined, and meanwhile, the concept clustering is introduced in the deep learning, so that the deep learning has certain new feature capturing and distinguishing capability, and the new concepts derived in the recognition process can be recognized and distinguished, thereby realizing a semi-supervised learning mode. Compared with the existing deep learning mode, the embodiment of the invention can enable the deep learning network to have better generalization capability, less sample demand and stronger multi-task adaptation capability, can improve the multi-task applicability, and has certain growth.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings that are needed in the description of the embodiments of the present invention will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
Fig. 1 is a schematic flow chart of an information identification method according to an embodiment of the present invention;
FIG. 2 is a schematic diagram of an artificial intelligence system to which the information identification method according to the embodiment of the present invention is applied;
Fig. 3 is a schematic structural diagram of an information identifying apparatus according to an embodiment of the present invention;
fig. 4 is a schematic structural diagram of an electronic device according to an embodiment of the present invention.
Detailed Description
As can be seen from the background art, in the related art, the main problem of the deep learning technology is that a large number of sample sets are needed to perform training, and in practice, a large amount of manpower and material resources are needed to perform sample labeling, so that the problem of relatively large resource consumption exists in the information identification system based on artificial intelligence in the prior art.
Meanwhile, specialized task training is usually needed when deep learning is performed for network training, different tasks often need different samples and networks to perform retraining, and tasks which are not performed with specialized task training cannot be autonomously learned in practical application, so that the problem of relatively poor multi-task applicability of an information identification system based on artificial intelligence in the prior art also exists.
Based on this, the embodiment of the present invention proposes a new information recognition scheme, and the technical solution in the embodiment of the present invention will be clearly and completely described below with reference to the drawings in the embodiment of the present invention, and it is obvious that the described embodiment is a part of embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
The information identifying method provided by the embodiment of the invention is first described below.
It should be noted that the information identification method provided by the embodiment of the invention can be applied to electronic equipment. Here, the electronic device may include an artificial intelligence system for identifying information to be identified, such as identifying a conceptual class of the information to be identified.
The artificial intelligence system may be divided into a plurality of functional units based on implementation functions, where each functional unit is configured to implement a function in an information identification process, for example, the artificial intelligence system may include a feature extraction unit configured to perform feature extraction on information to be identified to obtain a feature vector of the information to be identified, and for example, the artificial intelligence system may further include a feature comparison unit configured to compare the feature vector of the information to be identified with a comparison feature vector of each concept class to determine the concept class of the information to be identified. In particular, the artificial intelligence system includes which functional units, as will be described in detail below.
Referring to fig. 1, a flow chart of an information identification method provided by an embodiment of the present invention is shown. As shown in fig. 1, the method may include the steps of:
And step 101, comparing the obtained feature vector of the information to be identified with the comparison feature vector of each first concept class in the concept knowledge base to obtain a similarity judgment value of each first concept class.
Here, the information to be identified may be a multimedia file, and the multimedia file may be audio, video, image, text, etc., and in the following embodiment, the information to be identified will be described in detail by taking an image as an example.
The concept knowledge base can be used for storing concept knowledge maps, concept classes, concept examples, feature vectors of the concept examples and comparison feature vectors of the concept classes, and is convenient for each functional unit of the artificial intelligent system to acquire.
The concept knowledge graph can be used for marking concept classes of new concepts through reasoning and query, and is also used for aggregating concept instances of the same concept class, feature vectors of the concept instances and comparison feature vectors through a clustering mechanism.
The concept instance may be a multimedia file that has been identified as a concept class, e.g., for image a, which has been identified as a concept class "dog", then image a may be stored as a concept instance of a concept class "dog" into a concept repository.
These concept classes, concept instances, feature vectors of concept instances, and aligned feature vectors of concept classes stored in the concept knowledge base may be used as samples of neural networks in artificial intelligence systems for retraining to update vector weights of the networks, the detailed process of which will be described below.
The first concept class is a concept class in which features have been captured and distinguished and identified in the recognition process, which already exists in the concept repository and has been annotated, for example the first concept class is "dog".
The comparison feature vector may be a feature vector of each concept instance in the first concept class that best embodies the concept class. Which can be determined from the feature vectors of the concept instances in the concept class, a detailed description of which will be provided below.
The artificial intelligence system comprises a feature comparison unit, wherein the feature comparison unit can be used for comparing the feature vector of the information to be identified with the comparison feature vector of each first concept class to obtain similarity judgment values, and the similarity judgment values are used for judging whether the information to be identified belongs to which concept class.
Each similarity judgment value can be in any number between 0 and 1, and when the similarity judgment value obtained by comparing the feature vector of the information to be identified with the comparison feature vector of a first concept class is 0, the information to be identified is least similar to the first concept class, and is least likely to be the first concept class, and when the similarity judgment value obtained by comparing the feature vector of the information to be identified with the comparison feature vector of another first concept class is 1, the information to be identified is most similar to the first concept class, and the information to be identified is classified to the first concept class.
In addition, the output of the feature comparison unit, that is, the similarity judgment value, may be stored in the concept knowledge base to form a training sample later, so as to update the corresponding vector weight in the feature comparison unit.
The feature comparison unit can be formed by a full-connection network and a logistic regression model SoftMAX or a full-connection network and a logistic regression classifier, and can be used for carrying out sample iteration based on the full-connection network and the logistic regression model in the pre-training process to realize logistic regression judgment so as to obtain vector weight of the comparison model.
The feature comparison unit can also be formed by a full-connection network and a support vector machine (Support Vector Machine, SVM) classifier, and in the pre-training process, the feature comparison unit can be based on the full-connection network and can carry out sample iteration by taking the comparison feature vector as an SVM vector based on the SVM classifier, so that the vector weight of the comparison model is obtained.
It should be noted that the feature comparing unit may further use a distance decision algorithm to compare the feature vector of the information to be identified with the comparison feature vector of each first concept class in the concept knowledge base, such as a euclidean distance, a manhattan distance, a chebyshev distance, and the like. If the distance value obtained after the comparison with the comparison feature vector of the first concept class is shortest, the information to be identified can be determined to be most similar to the first concept class, and the information to be identified can be attributed to the first concept class.
Of course, before step 101, the method further comprises:
acquiring the information to be identified;
and extracting the characteristics of the information to be identified to obtain the characteristic vector of the information to be identified.
The artificial intelligence system can comprise a manual interaction interface, wherein the manual interaction interface is used for acquiring the information to be identified and outputting a concept result of the information to be identified, and the concept result can comprise a concept class of the information to be identified.
The artificial intelligence system further comprises a feature extraction unit, wherein the feature extraction unit is used for extracting features of the information to be identified and obtaining feature vectors of the information to be identified.
The feature extraction unit may be based on a deep learning neural network, and may be configured by a convolutional neural network (Convolutional Neural Networks, CNN) and/or a cyclic neural network (Recurrent Neural Network, RNN), such as a convolutional network portion of a residual neural network (Residual Neural Network, resNET), a super-resolution test sequence (Visual Geometry Group, VGG), alexNet, or the like.
In the pre-training process, sample iteration can be performed based on the deep learning neural network, so that the vector weight of the neural network is obtained.
It should be noted that the feature extraction unit and the feature comparison unit may be connected in cascade, and after the information to be identified is obtained, feature extraction is performed on the information to be identified based on the feature extraction unit, a feature vector of the information to be identified is obtained, and then the feature vector of the information to be identified is compared with the comparison feature vector of each first concept class based on the feature comparison unit, so as to obtain a similarity judgment value of each first concept class.
In practical application, the feature extraction unit may select a basic neural network structure (such as VGG 16) to pretrain the feature extraction network, and the same pre-stage convolutional network is adopted during training, and different fully-connected networks are cascaded to correspond to different recognition tasks. And taking the trained front-stage convolution network as an initial feature extraction network. And (3) utilizing the initial feature extraction network to cascade the network of the feature comparison unit (such as a fully connected network and a logistic regression classifier), solidifying the initial feature extraction network weight, utilizing the training sample to extract the concept class and the comparison feature vector, and carrying out initial training on the network of the feature comparison unit.
Step 102, obtaining a target similarity judgment value from the similarity judgment value of each first concept class; the target similarity judgment value is the highest similarity judgment value in the similarity judgment values of the first concept classes.
The artificial intelligence system may include a concept mapping unit, and the concept mapping unit may select, as the concept class of the information to be identified, a concept class corresponding to the aligned feature vector having the highest similarity to the feature vector output by the feature extraction unit, according to the output of the feature alignment unit.
Step 103, if the target similarity judgment value is greater than a first preset threshold value, determining a first concept class corresponding to the target comparison feature vector as the concept class of the information to be identified; the target comparison feature vector is a comparison feature vector corresponding to the target similarity judgment value.
The first preset threshold may be set according to practical situations, and is generally greater than 0.5 and less than 1. Accordingly, the first preset threshold value may be set to 0.75, and in the following embodiment, the first preset threshold value will be described in detail by taking 0.75 as an example.
If the target similarity judgment value is greater than a first preset threshold value, the concept mapping unit can attribute the information to be identified to a first concept class corresponding to a target comparison feature vector, wherein the target comparison feature vector is a comparison feature vector corresponding to the target similarity judgment value.
Meanwhile, the information to be identified can be stored into a concept knowledge base as a concept instance of the first concept class, and the feature vector of the information to be identified is also stored into the concept knowledge base as a feature vector instance of the concept instance.
Step 104, if the target similarity judgment value is smaller than a second preset threshold value, creating a second concept class in the concept knowledge base, and determining the second concept class as the concept class of the information to be identified.
The second preset threshold may also be set according to practical situations, and is generally greater than 0 and less than 0.5. Accordingly, the second preset threshold value may be set to 0.25, and in the following embodiment, the second preset threshold value will be described in detail by taking 0.25 as an example.
If the target similarity judgment value is smaller than a second preset threshold value, the concept mapping unit can determine that the information to be identified is dissimilar to the existing first concept class in the concept knowledge base, the information to be identified is likely to be a new concept which is derived, at this time, the concept mapping unit can create a second concept class, mark the second concept class under the condition of concept knowledge graph and/or human knowledge intervention, and meanwhile, determine the second concept class as the concept class of the information to be identified.
Meanwhile, the information to be identified can be stored into a concept knowledge base as a concept instance of the second concept class, and the feature vector of the information to be identified is also stored into the concept knowledge base as a feature vector instance of the concept instance.
At this time, the concept mapping unit has new feature capturing and distinguishing capability, and can identify and distinguish the new concept derived in the identification process, so as to realize a semi-supervised learning mode.
According to the information identification method provided by the embodiment of the invention, the deep learning technology and the knowledge graph technology are combined, and meanwhile, the concept clustering is introduced in the deep learning, so that the method has a certain new feature capturing and distinguishing capability, and the new concepts which are derived can be identified and distinguished in the identification process, so that a semi-supervised learning mode is realized. Compared with the existing deep learning mode, the embodiment of the invention can enable the deep learning network to have better generalization capability, less sample demand and stronger multi-task adaptation capability, can improve the multi-task applicability, and has certain growth.
In order to realize continuous learning and growth of the artificial intelligence system, with the increase of concept examples, the concept examples stored in the concept knowledge base and other stored information can be collected to form samples, and some parameters in the artificial intelligence system are updated, so that the updated parameters are more accurate, the identification capability of the artificial intelligence system is stronger, and the identification task is wider.
The details of updating some parameters in an artificial intelligence system are described below.
First, the home concept class of the concept instance is modified. Each concept instance may have an error in recognition, for example, the data of the concept instance is less in the early stage, the recognition accuracy of the information to be recognized may not be enough, the concept class recognized by the information to be recognized may have an error, for example, the concept class is the information to be recognized of a "cat", the artificial intelligence system recognizes that the concept class is the "dog", and the concept class of the information to be recognized has an error. Along with the increase of the data of the concept examples, the recognition accuracy is higher, and if errors are found in the concept class of the information to be recognized in the detection process, the concept class of the information to be recognized can be corrected.
After determining the concept class of the information to be identified, the method further comprises:
storing the information to be identified into the concept knowledge base as a concept instance of the concept class of the information to be identified;
comparing the concept instance corresponding to the information to be identified with the concept instance associated with the concept instance corresponding to the information to be identified in the concept class based on the concept relation between the concept class of the information to be identified and the concept class of the concept class aiming at each concept class except for the concept class of the information to be identified in all concept classes of the concept knowledge base, and obtaining the similarity between the concept instance corresponding to the information to be identified and the concept class;
and correcting the concept class of the concept instance corresponding to the information to be identified if the similarity between the concept instance corresponding to the information to be identified and the concept class is greater than a third preset threshold value aiming at each concept class except for the concept class of the information to be identified in all concept classes of the concept knowledge base.
The third preset threshold may be set according to practical situations, and is generally greater than 0.5 and less than 1. Accordingly, the third preset threshold value may be set to 0.75, and in the following embodiment, the third preset threshold value will be described in detail taking 0.75 as an example.
Specifically, the artificial intelligence system may further include a concept correction unit, configured to correct the concept class of the concept instance when it is determined that the annotation of the concept class of the concept instance is problematic.
When the concept knowledge base does not include the corrected concept class of the concept instance, the correction mode can modify the concept class label of the concept instance, for example, the concept class of the concept instance is "dog", however, in reality, the concept class of the concept instance is "cat", and the concept knowledge base does not include the concept class of the "cat", the label of the concept class of the concept instance is corrected, and the label "dog" is modified to "cat".
When the concept knowledge base includes the concept class modified by the concept instance, the modification manner may re-cluster the concept instance to the concept class modified by the concept instance, for example, the concept class of the concept instance is "dog", however, in reality, the concept class of the concept instance is "cat", and the concept knowledge base includes the concept class of "cat", and then the concept class of the concept instance is re-clustered to the concept class of "cat".
The concept correction unit may compare various concept relationships according to the concept knowledge graph based on the concept instance and the concept instance data associated with the concept instance to obtain similarity between the concept instance and each concept class based on the concept relationship, and then correct the concept class of the concept instance based on the obtained similarity.
For example, for a concept instance, the belonging concept classes may include two, and according to the concept knowledge graph, the two concept classes often appear at the same time, and there is a concept relationship, for example, for an image a, the image a includes "tree" and "grass", and according to the concept knowledge graph, the "tree" and the "grass" often appear at the same time in a concept instance, and the concept relationship of the "tree" and the "grass" is always "tree" under the "grass", and for the labeling of the concept class of the concept instance, the upper-labeled concept class is "grass" and the lower-labeled concept class is "tree", at this time, the labeling of the concept class of the concept instance is wrong, the upper-labeled concept class is "tree", and the lower-labeled concept class is "grass".
The foregoing is merely an example, and of course, in practical application, the method is not limited to one concept example, but can also be applied to different concept examples, and the concept classes of the two concept examples have a concept relationship, for example, before correction, the concept class is "cat", aiming at a concept knowledge graph, some big cats are very similar to lions, the concept classes are "cat" and "lion" have a concept relationship, based on the concept relationship, the similarity between the concept example and the concept class is "lion", if the similarity is greater than a third preset threshold, the concept class of the concept example can be corrected, the concept class is "cat" can be corrected to lion ", or the concept example is reclustered to the concept class of lion.
The foregoing is merely an example, and the similarity between the concept instance and another concept class may be calculated based on other concept relationships, and a target similarity may be determined, where the target similarity is the maximum similarity between the concept instance and each concept class based on the similarity of the corresponding concept relationships, and the home concept class of the concept instance may be modified based on the concept class corresponding to the target similarity.
In addition, the concept correction unit can also correct the attribution concept class of the concept instance based on other knowledge interventions, such as manual knowledge interventions and posterior knowledge interventions, in use.
Further, after updating the comparison feature vector of each concept class in the concept knowledge base and determining the concept class of the information to be identified, the method further includes:
storing the feature vector of the information to be identified into the concept knowledge base as a feature vector instance of the concept class of the information to be identified;
for each of all concept classes of the concept knowledge base, determining a target feature vector of the concept class from all feature vector instances of the concept class; the target feature vector of the concept class is a feature vector instance with the largest average value result of similarity judgment values obtained after comparison with all feature vector instances of the concept class in all feature vector instances of the concept class;
For each concept class in all concept classes of the concept knowledge base, if the concept class corresponds to the comparison feature vector, updating the target feature vector of the concept class into the comparison feature vector of the concept class; and if the concept class does not correspond to the comparison feature vector, determining the target feature vector of the concept class as the comparison feature vector of the concept class.
Specifically, the artificial intelligence system further comprises a concept comparison feature calculation unit, wherein the concept comparison feature calculation unit is used for updating or determining comparison feature vectors of various concept classes. Updating the comparison feature vector of the concept class for which the comparison feature vector has been determined, and directly determining the comparison feature vector of the concept class for which the comparison feature vector has not been determined.
For example, if the information to be identified is a new feature identified by the artificial intelligence system, the new feature belongs to a second concept class just created by the concept mapping unit, that is, the concept instance of the concept class only has the information to be identified, that is, only has one feature vector instance in the concept class (that is, the feature vector of the information to be identified), in this case, the feature vector may be directly determined as the comparison feature vector of the concept class.
The concept comparison feature calculation unit may search a target feature vector for each concept class, so that feature vector instances of all concept instances belonging to the concept class pass through the feature comparison unit to maximize the average function F of similarity judgment values of the target feature vector, and then update the searched target feature vector to be a comparison feature vector of the concept class. Here, the average function F may be any one of arithmetic average, square sum average, geometric average, mean average, and the like.
Here, the manner of searching for the target feature vector may be solved based on a sub-gradient method, which will not be described herein since this technique belongs to a conventional technique in the art.
Further, based on the feature extraction unit and the feature comparison unit, the respective corresponding vector weights are updated.
If the concept class corresponds to the comparison feature vector, updating the target feature vector of the concept class into the comparison feature vector of the concept class; if the concept class does not correspond to the aligned feature vector, after determining the target feature vector of the concept class as the aligned feature vector of the concept class, the method further includes:
Generating a training sample based on at least one concept class, at least one concept instance of each concept class and a comparison feature vector of each concept class stored in the concept knowledge base;
and updating vector weights for carrying out feature extraction and/or feature comparison on the information to be identified based on the training samples.
Specifically, the artificial intelligence system further comprises a network training unit, wherein the network training unit is used for training the feature extraction unit and the feature comparison unit based on the generated samples so as to update the vector weights corresponding to the feature extraction unit and the feature comparison unit.
For the feature extraction unit, the training samples may be concept examples stored in the concept knowledge base and feature vectors of the concept examples, training is performed based on the training samples, and vector weights of feature extraction are updated.
For the feature comparison unit, the training samples may be feature vectors of concept examples stored in the concept knowledge base, comparison feature vectors of each concept class, and similarity judgment values after the feature vectors of the concept examples are compared with the comparison feature vectors of each concept class, training is performed based on the training samples, and vector weights of the feature comparison models are updated.
It should be noted that in the artificial intelligence system, the working periods of the concept correction unit, the concept comparison feature calculation unit and the network training unit for updating parameters are generally larger than those of the feature extraction unit, the feature comparison unit and the concept mapping unit for information identification.
An information recognition method is exemplified in detail below.
The application scene is an image A, the image A is specifically a cat picture, referring to FIG. 2, which shows a schematic diagram of an artificial intelligence system to which the information identification method provided by the embodiment of the invention is applied. As shown in fig. 2, the artificial intelligence system includes a feature extraction unit 201, a feature comparison unit 202, a concept mapping unit 203, a concept correction unit 204, a concept comparison feature calculation unit 205, and a network training unit 206. In addition, the artificial intelligence system may further include a concept knowledge base 207, where the concept knowledge base 207 is configured to store a concept knowledge graph, a concept class, a concept instance, a feature vector of the concept instance, and a comparison feature vector of the concept class, so as to facilitate each functional unit of the artificial intelligence system to acquire.
The learning using method of the artificial intelligence system comprises a pre-training process, a training process and a using process.
The pre-training process is as follows:
firstly, selecting a basic neural network structure (such as VGG 16) to pretrain a feature extraction network, adopting the same front-stage convolution network during training, cascading different full-connection networks to correspond to different recognition tasks, and taking the trained front-stage convolution network as an initial feature extraction network.
And then, utilizing the initial feature extraction network to cascade the network of the feature comparison unit (such as a fully connected network and a logistic regression classifier), solidifying the initial feature extraction network weight, utilizing the training sample to extract the concept class and the comparison feature vector, and carrying out initial training on the network of the feature comparison unit.
And finally, the cascade initial feature extraction unit network and the feature comparison unit network are subjected to joint training.
The training process is as follows:
the new training samples are input into the pre-trained artificial intelligent system for formal training, and are processed according to the feature extraction unit 201, the feature comparison unit 202, the concept mapping unit 203, the concept correction unit 204, the concept comparison feature calculation unit 205 and the network training unit 206. In the processing stage of the concept correction unit 204 during training, concept classes of concept instances are corrected randomly based on concept classes of input information in samples, meanwhile, in the processing stage of the concept comparison feature calculation unit 205, comparison feature vectors of each concept class are updated based on new training samples, and in the processing stage of the network training unit 206, vector weights of a network of the feature extraction unit and a network of the feature comparison unit are updated based on new training samples.
The use process is as follows:
the image a is input to the feature extraction unit 201, passes through the feature comparison unit 202 and the concept mapping unit 203, and the mapping result of the concept mapping unit 203 is taken as the concept result of the image a, which may include the concept class of the image a, that is, "cat" may be output. During use, other functional units of the system still perform closed loop operation, for example, the concept correction unit 204, the concept comparison feature calculation unit 205 and the network training unit 206 still perform update processing.
It should be noted that, in the use process, if the artificial intelligence system captures the new features, the capability expansion can be easily realized only by introducing the concept class corresponding to the new features into the concept knowledge graph or marking the new concept class, for example, the originally trained artificial intelligence system for identifying dogs can evolve the capability of identifying cats in the use process.
The information identifying apparatus provided by the embodiment of the invention is described below.
Referring to fig. 3, a schematic structural diagram of an information identifying apparatus according to an embodiment of the present invention is shown. As shown in fig. 3, the information identifying apparatus 300 includes:
the feature comparison module 301 is configured to compare the obtained feature vector of the information to be identified with a comparison feature vector of each first concept class in the concept knowledge base, so as to obtain a similarity judgment value of each first concept class;
A first obtaining module 302, configured to obtain a target similarity determination value from the similarity determination values of each of the first concept classes; the target similarity judgment value is the highest similarity judgment value in the similarity judgment values of the first concept classes;
a first determining module 303, configured to determine a first concept class corresponding to the target comparison feature vector as the concept class of the information to be identified, if the target similarity decision value is greater than a first preset threshold; the target comparison feature vector is a comparison feature vector corresponding to the target similarity judgment value;
and a second determining module 304, configured to create a second concept class in the concept repository if the target similarity determination value is smaller than a second preset threshold, and determine the second concept class as the concept class of the information to be identified.
Optionally, the apparatus further includes:
the second acquisition module is used for acquiring the information to be identified;
and the feature extraction module is used for carrying out feature extraction on the information to be identified to obtain a feature vector of the information to be identified.
Optionally, the apparatus further includes:
the first storage module is used for storing the information to be identified into the concept knowledge base as a concept instance of the concept class of the information to be identified;
The concept comparison module is used for comparing the concept instance corresponding to the information to be identified with the concept instance associated with the concept instance corresponding to the information to be identified in the concept class based on the concept relation between the concept class of the information to be identified and the concept class of the concept class aiming at each concept class except for the concept class of the information to be identified in all concept classes of the concept knowledge base, so as to obtain the similarity of the concept instance corresponding to the information to be identified and the concept class;
and the correction module is used for correcting the concept class of the concept instance corresponding to the information to be identified according to the similarity between the concept instance corresponding to the information to be identified and the concept class which is greater than a third preset threshold value in all the concept classes of the concept knowledge base except for the concept class of the information to be identified.
Optionally, the apparatus further includes:
the second storage module is used for storing the feature vector of the information to be identified into the concept knowledge base as a feature vector instance of the concept class of the information to be identified;
a third determining module, configured to determine, for each of all concept classes of the concept repository, a target feature vector of the concept class from all feature vector instances of the concept class; the target feature vector of the concept class is a feature vector instance with the largest average value result of similarity judgment values obtained after comparison with all feature vector instances of the concept class in all feature vector instances of the concept class;
The first updating module is used for updating the target feature vector of the concept class into the comparison feature vector of the concept class aiming at each concept class in all concept classes of the concept knowledge base if the concept class corresponds to the comparison feature vector; and if the concept class does not correspond to the comparison feature vector, determining the target feature vector of the concept class as the comparison feature vector of the concept class.
Optionally, the apparatus further includes:
the generation module is used for generating a training sample based on at least one concept class, at least one concept instance of each concept class and a comparison feature vector of each concept class stored in the concept knowledge base;
and the second updating module is used for updating the vector weight for carrying out feature extraction and/or feature comparison on the information to be identified based on the training sample.
The device provided by the embodiment of the invention can realize each process realized in the embodiment of the method, and in order to avoid repetition, the description is omitted here.
The information identification device provided by the embodiment of the invention combines the deep learning technology and the knowledge graph technology, and simultaneously introduces concept clustering in the deep learning to enable the deep learning to have a certain new feature capturing and distinguishing capability, so that the new concepts which are derived can be identified and distinguished in the identification process, and a semi-supervised learning mode is realized. Compared with the existing deep learning mode, the embodiment of the invention can enable the deep learning network to have better generalization capability, less sample demand and stronger multi-task adaptation capability, can improve the multi-task applicability, and has certain growth.
The electronic device provided by the embodiment of the invention is explained below.
Referring to fig. 4, a schematic structural diagram of an electronic device according to an embodiment of the present invention is shown. As shown in fig. 4, the electronic device 400 includes: a processor 401, a memory 402, a user interface 403 and a bus interface 404.
A processor 401 for reading the program in the memory 402, performing the following process:
comparing the obtained feature vector of the information to be identified with the comparison feature vector of each first concept class in the concept knowledge base to obtain a similarity judgment value of each first concept class;
obtaining a target similarity judgment value from the similarity judgment value of each first concept class; the target similarity judgment value is the highest similarity judgment value in the similarity judgment values of the first concept classes;
if the target similarity judgment value is larger than a first preset threshold value, determining a first concept class corresponding to the target comparison feature vector as the concept class of the information to be identified; the target comparison feature vector is a comparison feature vector corresponding to the target similarity judgment value;
if the target similarity judgment value is smaller than a second preset threshold value, creating a second concept class in the concept knowledge base, and determining the second concept class as the concept class of the information to be identified.
In fig. 4, a bus architecture may comprise any number of interconnected buses and bridges, with one or more processors, represented in particular by processor 401, and various circuits of memory, represented by memory 402, linked together. The bus architecture may also link together various other circuits such as peripheral devices, voltage regulators, power management circuits, etc., which are well known in the art and, therefore, will not be described further herein. Bus interface 404 provides an interface. The user interface 403 may also be an interface capable of interfacing with an inscribed desired device for a different user device, including but not limited to a keypad, display, speaker, microphone, joystick, etc.
The processor 401 is responsible for managing the bus architecture and general processing, and the memory 402 may store data used by the processor 401 in performing operations.
Optionally, the processor 401 is further configured to:
acquiring the information to be identified;
and extracting the characteristics of the information to be identified to obtain the characteristic vector of the information to be identified.
Optionally, the processor 401 is further configured to:
storing the information to be identified into the concept knowledge base as a concept instance of the concept class of the information to be identified;
Comparing the concept instance corresponding to the information to be identified with the concept instance associated with the concept instance corresponding to the information to be identified in the concept class based on the concept relation between the concept class of the information to be identified and the concept class of the concept class aiming at each concept class except for the concept class of the information to be identified in all concept classes of the concept knowledge base, and obtaining the similarity between the concept instance corresponding to the information to be identified and the concept class;
and correcting the concept class of the concept instance corresponding to the information to be identified if the similarity between the concept instance corresponding to the information to be identified and the concept class is greater than a third preset threshold value aiming at each concept class except for the concept class of the information to be identified in all concept classes of the concept knowledge base.
Optionally, the processor 401 is further configured to:
storing the feature vector of the information to be identified into the concept knowledge base as a feature vector instance of the concept class of the information to be identified;
for each of all concept classes of the concept knowledge base, determining a target feature vector of the concept class from all feature vector instances of the concept class; the target feature vector of the concept class is a feature vector instance with the largest average value result of similarity judgment values obtained after comparison with all feature vector instances of the concept class in all feature vector instances of the concept class;
For each concept class in all concept classes of the concept knowledge base, if the concept class corresponds to the comparison feature vector, updating the target feature vector of the concept class into the comparison feature vector of the concept class; and if the concept class does not correspond to the comparison feature vector, determining the target feature vector of the concept class as the comparison feature vector of the concept class.
Optionally, the processor 401 is further configured to:
generating a training sample based on at least one concept class, at least one concept instance of each concept class and a comparison feature vector of each concept class stored in the concept knowledge base;
and updating vector weights for carrying out feature extraction and/or feature comparison on the information to be identified based on the training samples.
In the embodiment of the invention, the deep learning technology and the knowledge graph technology are combined, and meanwhile, the concept clustering is introduced in the deep learning, so that the deep learning has certain new feature capturing and distinguishing capability, and the new concepts derived in the recognition process can be recognized and distinguished, thereby realizing a semi-supervised learning mode. Compared with the existing deep learning mode, the embodiment of the invention can enable the deep learning network to have better generalization capability, less sample demand and stronger multi-task adaptation capability, can improve the multi-task applicability, and has certain growth.
Preferably, the embodiment of the present invention further provides an electronic device, including a processor 401, a memory 402, and a computer program stored in the memory 402 and capable of running on the processor 401, where the computer program when executed by the processor 401 implements each process of the above embodiment of the information identifying method, and the same technical effects can be achieved, and for avoiding repetition, a description is omitted herein.
The embodiment of the invention also provides a computer readable storage medium, on which a computer program is stored, which when executed by a processor, implements the processes of the above-mentioned information identification method embodiment, and can achieve the same technical effects, and in order to avoid repetition, the description is omitted here. Wherein the computer readable storage medium is selected from Read-Only Memory (ROM), random access Memory (Random Access Memory, RAM), magnetic disk or optical disk.
Those of ordinary skill in the art will appreciate that the various illustrative elements and algorithm steps described in connection with the embodiments disclosed herein may be implemented as electronic hardware, or combinations of computer software and electronic hardware. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the solution. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present invention.
It will be clear to those skilled in the art that, for convenience and brevity of description, specific working procedures of the above-described systems, apparatuses and units may refer to corresponding procedures in the foregoing method embodiments, and are not repeated herein.
In the embodiments provided herein, it should be understood that the disclosed systems and methods may be implemented in other ways. For example, the system embodiments described above are merely illustrative, e.g., the division of the elements is merely a logical functional division, and there may be additional divisions when actually implemented, e.g., multiple elements or components may be combined or integrated into another system, or some features may be omitted or not performed. Alternatively, the coupling or direct coupling or communication connection shown or discussed with each other may be an indirect coupling or communication connection via some interfaces, devices or units, which may be in electrical, mechanical or other form.
The units described as separate units may or may not be physically separate, and units shown as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the embodiment of the present invention.
In addition, each functional unit in the embodiments of the present invention may be integrated in one processing unit, or each unit may exist alone physically, or two or more units may be integrated in one unit.
The functions, if implemented in the form of software functional units and sold or used as a stand-alone product, may be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention may be embodied essentially or in a part contributing to the prior art or in a part of the technical solution, in the form of a software product stored in a storage medium, comprising several instructions for causing a computer device (which may be a personal computer, a server, a network device, etc.) to perform all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a usb disk, a removable hard disk, a ROM, a RAM, a magnetic disk, or an optical disk, etc.
The foregoing is merely illustrative of the present invention, and the present invention is not limited thereto, and any person skilled in the art will readily recognize that variations or substitutions are within the scope of the present invention. Therefore, the protection scope of the invention is subject to the protection scope of the claims.
Claims (6)
1. An information identification method, characterized in that the method comprises:
comparing the obtained feature vector of the information to be identified with the comparison feature vector of each first concept class in the concept knowledge base to obtain a similarity judgment value of each first concept class, wherein the information to be identified is a multimedia file, and the multimedia file is audio, video, image and text;
obtaining a target similarity judgment value from the similarity judgment value of each first concept class; the target similarity judgment value is the highest similarity judgment value in the similarity judgment values of the first concept classes;
if the target similarity judgment value is larger than a first preset threshold value, determining a first concept class corresponding to the target comparison feature vector as the concept class of the information to be identified; the target comparison feature vector is a comparison feature vector corresponding to the target similarity judgment value;
if the target similarity judgment value is smaller than a second preset threshold value, creating a second concept class in the concept knowledge base, and determining the second concept class as the concept class of the information to be identified;
after determining the concept class of the information to be identified, the method further comprises:
Storing the feature vector of the information to be identified into the concept knowledge base as a feature vector instance of the concept class of the information to be identified;
for each of all concept classes of the concept knowledge base, determining a target feature vector of the concept class from all feature vector instances of the concept class; the target feature vector of the concept class is a feature vector instance with the largest average value result of similarity judgment values obtained after comparison with all feature vector instances of the concept class in all feature vector instances of the concept class;
for each concept class in all concept classes of the concept knowledge base, if the concept class corresponds to the comparison feature vector, updating the target feature vector of the concept class into the comparison feature vector of the concept class; if the concept class does not correspond to the comparison feature vector, determining the target feature vector of the concept class as the comparison feature vector of the concept class;
if the concept class corresponds to the comparison feature vector, updating the target feature vector of the concept class into the comparison feature vector of the concept class; if the concept class does not correspond to the aligned feature vector, after determining the target feature vector of the concept class as the aligned feature vector of the concept class, the method further includes:
Generating a training sample based on at least one concept class, at least one concept instance of each concept class and a comparison feature vector of each concept class stored in the concept knowledge base;
updating vector weights for feature extraction and/or feature comparison of the information to be identified based on the training samples;
after determining the concept class of the information to be identified, the method further comprises:
storing the information to be identified into the concept knowledge base as a concept instance of the concept class of the information to be identified;
comparing the concept instance corresponding to the information to be identified with the concept instance associated with the concept instance corresponding to the information to be identified in the concept class based on the concept relation between the concept class of the information to be identified and the concept class of each concept class except the concept class of the information to be identified in all concept classes of the concept knowledge base, and obtaining the similarity between the concept instance corresponding to the information to be identified and the concept class, wherein the concept relation is obtained by comparing a concept knowledge graph, and the concept knowledge graph is used for marking the concept class of a new concept by reasoning and inquiring;
And correcting the concept class of the concept instance corresponding to the information to be identified if the similarity between the concept instance corresponding to the information to be identified and the concept class is greater than a third preset threshold value aiming at each concept class except for the concept class of the information to be identified in all concept classes of the concept knowledge base.
2. The method according to claim 1, wherein the comparing the obtained feature vector of the information to be identified with the comparison feature vector of each first concept class in the concept repository, before obtaining the similarity judgment value of each first concept class, the method further comprises:
acquiring the information to be identified;
and extracting the characteristics of the information to be identified to obtain the characteristic vector of the information to be identified.
3. An information identifying apparatus, characterized in that the apparatus comprises:
the feature comparison module is used for comparing the obtained feature vector of the information to be identified with the comparison feature vector of each first concept class in the concept knowledge base to obtain a similarity judgment value of each first concept class, wherein the information to be identified is a multimedia file, and the multimedia file is audio, video, image and text;
The first acquisition module is used for acquiring a target similarity judgment value from the similarity judgment values of the first concept classes; the target similarity judgment value is the highest similarity judgment value in the similarity judgment values of the first concept classes;
the first determining module is used for determining a first concept class corresponding to the target comparison feature vector as the concept class of the information to be identified if the target similarity judging value is larger than a first preset threshold value; the target comparison feature vector is a comparison feature vector corresponding to the target similarity judgment value;
the second determining module is configured to create a second concept class in the concept knowledge base if the target similarity decision value is smaller than a second preset threshold, and determine the second concept class as the concept class of the information to be identified;
the apparatus further comprises:
the second storage module is used for storing the feature vector of the information to be identified into the concept knowledge base as a feature vector instance of the concept class of the information to be identified;
a third determining module, configured to determine, for each of all concept classes of the concept repository, a target feature vector of the concept class from all feature vector instances of the concept class; the target feature vector of the concept class is a feature vector instance with the largest average value result of similarity judgment values obtained after comparison with all feature vector instances of the concept class in all feature vector instances of the concept class;
The first updating module is used for updating the target feature vector of the concept class into the comparison feature vector of the concept class aiming at each concept class in all concept classes of the concept knowledge base if the concept class corresponds to the comparison feature vector; if the concept class does not correspond to the comparison feature vector, determining the target feature vector of the concept class as the comparison feature vector of the concept class;
the apparatus further comprises:
the generation module is used for generating a training sample based on at least one concept class, at least one concept instance of each concept class and a comparison feature vector of each concept class stored in the concept knowledge base;
the second updating module is used for updating vector weights for carrying out feature extraction and/or feature comparison on the information to be identified based on the training samples;
the apparatus further comprises:
the first storage module is used for storing the information to be identified into the concept knowledge base as a concept instance of the concept class of the information to be identified;
the concept comparison module is used for comparing the concept instance corresponding to the information to be identified with the concept instance associated with the concept instance corresponding to the information to be identified in the concept class based on the concept relation between the concept class of the information to be identified and the concept class of the concept class aiming at each concept class except for the concept class of the information to be identified in all concept classes of the concept knowledge base, so as to obtain the similarity between the concept instance corresponding to the information to be identified and the concept class, wherein the concept relation is obtained by comparing a concept knowledge graph, and the concept knowledge graph is used for marking the concept class of a new concept through reasoning and query;
And the correction module is used for correcting the concept class of the concept instance corresponding to the information to be identified according to the similarity between the concept instance corresponding to the information to be identified and the concept class which is greater than a third preset threshold value in all the concept classes of the concept knowledge base except for the concept class of the information to be identified.
4. A device according to claim 3, characterized in that the device further comprises:
the second acquisition module is used for acquiring the information to be identified;
and the feature extraction module is used for carrying out feature extraction on the information to be identified to obtain a feature vector of the information to be identified.
5. An electronic device comprising a processor, a memory, a computer program stored on the memory and executable on the processor, which when executed by the processor implements the steps of the information identification method according to any one of claims 1 to 2.
6. A computer-readable storage medium, characterized in that the computer-readable storage medium has stored thereon a computer program which, when executed by a processor, implements the steps of the information identification method according to any of claims 1 to 2.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911012568.4A CN112699909B (en) | 2019-10-23 | 2019-10-23 | Information identification method, information identification device, electronic equipment and computer readable storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911012568.4A CN112699909B (en) | 2019-10-23 | 2019-10-23 | Information identification method, information identification device, electronic equipment and computer readable storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112699909A CN112699909A (en) | 2021-04-23 |
| CN112699909B true CN112699909B (en) | 2024-03-19 |
Family
ID=75505126
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911012568.4A Active CN112699909B (en) | 2019-10-23 | 2019-10-23 | Information identification method, information identification device, electronic equipment and computer readable storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112699909B (en) |
Citations (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8315849B1 (en) * | 2010-04-09 | 2012-11-20 | Wal-Mart Stores, Inc. | Selecting terms in a document |
| CN103699663A (en) * | 2013-12-27 | 2014-04-02 | 中国科学院自动化研究所 | Hot event mining method based on large-scale knowledge base |
| CN104008090A (en) * | 2014-04-29 | 2014-08-27 | 河海大学 | Multi-subject extraction method based on concept vector model |
| CN106446093A (en) * | 2016-09-12 | 2017-02-22 | 百度在线网络技术（北京）有限公司 | Concept aggregation method and device |
| CN106650940A (en) * | 2016-12-26 | 2017-05-10 | 东软集团股份有限公司 | Field knowledge base establishment method and device |
| CN107895193A (en) * | 2017-11-13 | 2018-04-10 | 北京神州泰岳软件股份有限公司 | A kind of construction of knowledge base method, parameter setting method and device based on dialogue |
| CN108153818A (en) * | 2017-11-29 | 2018-06-12 | 成都东方盛行电子有限责任公司 | A kind of clustering method based on big data |
| CN108182245A (en) * | 2017-12-28 | 2018-06-19 | 北京锐安科技有限公司 | The construction method and device of people's object properties classificating knowledge collection of illustrative plates |
| CN108647264A (en) * | 2018-04-28 | 2018-10-12 | 北京邮电大学 | A kind of image automatic annotation method and device based on support vector machines |
| CN109492222A (en) * | 2018-10-31 | 2019-03-19 | 平安科技（深圳）有限公司 | Intension recognizing method, device and computer equipment based on conceptional tree |
| CN109783582A (en) * | 2018-12-04 | 2019-05-21 | 平安科技（深圳）有限公司 | A kind of knowledge base alignment schemes, device, computer equipment and storage medium |
| CN109829433A (en) * | 2019-01-31 | 2019-05-31 | 北京市商汤科技开发有限公司 | Facial image recognition method, device, electronic equipment and storage medium |
| CN109871446A (en) * | 2019-01-31 | 2019-06-11 | 平安科技（深圳）有限公司 | Rejection method for identifying, electronic device and storage medium in intention assessment |
| CN110070087A (en) * | 2019-05-05 | 2019-07-30 | 广东三维家信息科技有限公司 | Image identification method and device |
| CN110110800A (en) * | 2019-05-14 | 2019-08-09 | 长沙理工大学 | Automatic image marking method, device, equipment and computer readable storage medium |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10394829B2 (en) * | 2015-12-08 | 2019-08-27 | International Business Machines Corporation | Content authoring |
-
2019
- 2019-10-23 CN CN201911012568.4A patent/CN112699909B/en active Active
Patent Citations (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8315849B1 (en) * | 2010-04-09 | 2012-11-20 | Wal-Mart Stores, Inc. | Selecting terms in a document |
| CN103699663A (en) * | 2013-12-27 | 2014-04-02 | 中国科学院自动化研究所 | Hot event mining method based on large-scale knowledge base |
| CN104008090A (en) * | 2014-04-29 | 2014-08-27 | 河海大学 | Multi-subject extraction method based on concept vector model |
| CN106446093A (en) * | 2016-09-12 | 2017-02-22 | 百度在线网络技术（北京）有限公司 | Concept aggregation method and device |
| CN106650940A (en) * | 2016-12-26 | 2017-05-10 | 东软集团股份有限公司 | Field knowledge base establishment method and device |
| CN107895193A (en) * | 2017-11-13 | 2018-04-10 | 北京神州泰岳软件股份有限公司 | A kind of construction of knowledge base method, parameter setting method and device based on dialogue |
| CN108153818A (en) * | 2017-11-29 | 2018-06-12 | 成都东方盛行电子有限责任公司 | A kind of clustering method based on big data |
| CN108182245A (en) * | 2017-12-28 | 2018-06-19 | 北京锐安科技有限公司 | The construction method and device of people's object properties classificating knowledge collection of illustrative plates |
| CN108647264A (en) * | 2018-04-28 | 2018-10-12 | 北京邮电大学 | A kind of image automatic annotation method and device based on support vector machines |
| CN109492222A (en) * | 2018-10-31 | 2019-03-19 | 平安科技（深圳）有限公司 | Intension recognizing method, device and computer equipment based on conceptional tree |
| CN109783582A (en) * | 2018-12-04 | 2019-05-21 | 平安科技（深圳）有限公司 | A kind of knowledge base alignment schemes, device, computer equipment and storage medium |
| CN109829433A (en) * | 2019-01-31 | 2019-05-31 | 北京市商汤科技开发有限公司 | Facial image recognition method, device, electronic equipment and storage medium |
| CN109871446A (en) * | 2019-01-31 | 2019-06-11 | 平安科技（深圳）有限公司 | Rejection method for identifying, electronic device and storage medium in intention assessment |
| CN110070087A (en) * | 2019-05-05 | 2019-07-30 | 广东三维家信息科技有限公司 | Image identification method and device |
| CN110110800A (en) * | 2019-05-14 | 2019-08-09 | 长沙理工大学 | Automatic image marking method, device, equipment and computer readable storage medium |
Non-Patent Citations (2)
| Title |
|---|
| 基于语义树的中文词语相似度计算与分析;张亮;尹存燕;陈家骏;;中文信息学报(第06期);25-32 * |
| 面向地下管线空间数据匹配的概念语义相似性方法;房彩丽;张书亮;;计算机辅助设计与图形学学报(第04期);149-156 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112699909A (en) | 2021-04-23 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Chen et al. | Order-free rnn with visual attention for multi-label classification | |
| CN112199375B (en) | Cross-modal data processing method and device, storage medium and electronic device | |
| Wang et al. | Generalizing from a few examples: A survey on few-shot learning | |
| Meng et al. | Leveraging concept association network for multimedia rare concept mining and retrieval | |
| CN112819023B (en) | Sample set acquisition method, device, computer equipment and storage medium | |
| Wang et al. | Learning deep conditional neural network for image segmentation | |
| CN104915351A (en) | Picture sorting method and terminal | |
| CN111340112B (en) | Classification method, classification device and classification server | |
| CN114358188A (en) | Feature extraction model processing method, feature extraction model processing device, sample retrieval method, sample retrieval device and computer equipment | |
| Long et al. | Deep neural networks in fully connected crf for image labeling with social network metadata | |
| CN115131604A (en) | Multi-label image classification method and device, electronic equipment and storage medium | |
| Mishra et al. | Effectively leveraging attributes for visual similarity | |
| Dulac-Arnold et al. | Sequentially generated instance-dependent image representations for classification | |
| CN114048843A (en) | Small sample learning network based on selective feature migration | |
| CN113723426A (en) | Image classification method and device based on deep multi-flow neural network | |
| CN113704534A (en) | Image processing method and device and computer equipment | |
| CN113761291A (en) | Processing method and device for label classification | |
| Lv et al. | Retrieval oriented deep feature learning with complementary supervision mining | |
| US20230419170A1 (en) | System and method for efficient machine learning | |
| CN112699909B (en) | Information identification method, information identification device, electronic equipment and computer readable storage medium | |
| CN114462526A (en) | Classification model training method and device, computer equipment and storage medium | |
| CN113822130A (en) | Model training method, scene recognition method, computing device, and medium | |
| CN113591731A (en) | Knowledge distillation-based weak surveillance video time sequence behavior positioning method | |
| CN113392867A (en) | Image identification method and device, computer equipment and storage medium | |
| CN113590720A (en) | Data classification method and device, computer equipment and storage medium |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |