CN115601771A - Business order identification method, device, medium and terminal equipment based on multi-mode data - Google Patents
Business order identification method, device, medium and terminal equipment based on multi-mode data Download PDFInfo
- Publication number
- CN115601771A CN115601771A CN202211523510.8A CN202211523510A CN115601771A CN 115601771 A CN115601771 A CN 115601771A CN 202211523510 A CN202211523510 A CN 202211523510A CN 115601771 A CN115601771 A CN 115601771A
- Authority
- CN
- China
- Prior art keywords
- data
- feature vector
- text
- numerical
- category
- 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
- 238000000034 method Methods 0.000 title claims abstract description 41
- 239000013598 vector Substances 0.000 claims abstract description 200
- 230000004927 fusion Effects 0.000 claims abstract description 80
- 238000000605 extraction Methods 0.000 claims description 37
- 238000013528 artificial neural network Methods 0.000 claims description 25
- 230000004931 aggregating effect Effects 0.000 claims description 24
- 238000012935 Averaging Methods 0.000 claims description 22
- 238000004590 computer program Methods 0.000 claims description 17
- 230000000694 effects Effects 0.000 claims description 12
- 238000012549 training Methods 0.000 claims description 9
- 238000012545 processing Methods 0.000 description 16
- 230000002776 aggregation Effects 0.000 description 11
- 238000004220 aggregation Methods 0.000 description 11
- 238000004422 calculation algorithm Methods 0.000 description 8
- 238000010586 diagram Methods 0.000 description 7
- 230000006872 improvement Effects 0.000 description 6
- 238000010801 machine learning Methods 0.000 description 6
- 230000006870 function Effects 0.000 description 5
- 230000008569 process Effects 0.000 description 4
- 238000007499 fusion processing Methods 0.000 description 3
- 230000010354 integration Effects 0.000 description 3
- 238000005457 optimization Methods 0.000 description 3
- 230000004913 activation Effects 0.000 description 2
- 238000007635 classification algorithm Methods 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000006116 polymerization reaction Methods 0.000 description 2
- 230000006399 behavior Effects 0.000 description 1
- 230000002457 bidirectional effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000013500 data storage Methods 0.000 description 1
- 230000010076 replication Effects 0.000 description 1
- 230000000717 retained effect Effects 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
- 239000007787 solid Substances 0.000 description 1
- 230000002269 spontaneous effect Effects 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 238000000844 transformation Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
- G06V30/40—Document-oriented image-based pattern recognition
- G06V30/41—Analysis of document content
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
- G06Q30/0201—Market modelling; Market analysis; Collecting market data
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
- G06V30/10—Character recognition
- G06V30/19—Recognition using electronic means
- G06V30/191—Design or setup of recognition systems or techniques; Extraction of features in feature space; Clustering techniques; Blind source separation
- G06V30/19173—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
- G06V30/10—Character recognition
- G06V30/19—Recognition using electronic means
- G06V30/191—Design or setup of recognition systems or techniques; Extraction of features in feature space; Clustering techniques; Blind source separation
- G06V30/1918—Fusion techniques, i.e. combining data from various sources, e.g. sensor fusion
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
- G06V30/10—Character recognition
- G06V30/26—Techniques for post-processing, e.g. correcting the recognition result
- G06V30/262—Techniques for post-processing, e.g. correcting the recognition result using context analysis, e.g. lexical, syntactic or semantic context
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Multimedia (AREA)
- Business, Economics & Management (AREA)
- Development Economics (AREA)
- Accounting & Taxation (AREA)
- Strategic Management (AREA)
- Finance (AREA)
- Evolutionary Computation (AREA)
- Entrepreneurship & Innovation (AREA)
- Databases & Information Systems (AREA)
- Artificial Intelligence (AREA)
- Data Mining & Analysis (AREA)
- Computational Linguistics (AREA)
- General Engineering & Computer Science (AREA)
- Software Systems (AREA)
- Medical Informatics (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Health & Medical Sciences (AREA)
- Game Theory and Decision Science (AREA)
- Economics (AREA)
- Marketing (AREA)
- General Business, Economics & Management (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a business order identification method, a device, a medium and terminal equipment based on multi-mode data, wherein the method comprises the following steps: acquiring data to be identified from a network platform, and extracting multi-modal data in the data to be identified, wherein the multi-modal data comprises text data, category data and numerical data; respectively extracting features of the text data, the category data and the numerical data to correspondingly obtain a text feature vector, a category feature vector and a numerical feature vector; performing feature fusion on the text feature vector, the category feature vector and the numerical feature vector to obtain a fusion feature vector; the method comprises the steps of identifying fusion characteristic vectors according to a preset multi-model integrated learning system to obtain a quotient list identification result, wherein the multi-model integrated learning system is generated in a deep layer stacking mode and comprises at least two network layers, and each network layer comprises at least two basic models. By adopting the technical scheme of the invention, the accuracy of the identification result is improved.
Description
Technical Field
The invention relates to the technical field of internet big data processing, in particular to a business order identification method and device based on multi-mode data, a computer readable storage medium and terminal equipment.
Background
The content marketing can be divided into a business list and user generated content according to the nature of sources, wherein the business list is a mode of exchanging consideration provided by a business or a brand party to enable a user to conduct text transmission and publicity on social media, and the user generated content is a spontaneous sharing behavior of the user. The business form is correctly identified according to the user text content, a business company or a brand can be helped to know the marketing strategy of a competitor, the marketing effect of a specific brand is monitored, and the marketing strategy is formulated on the data level so as to achieve higher return on investment.
The recognition of the business order can be converted into a common two-classification problem in the field of machine learning, namely, a given text on a social media is judged whether the text is the business order or not by utilizing a classification algorithm. At present, a supervised learning algorithm is adopted in a mainstream classification algorithm, namely positive and negative samples are provided, data characteristics are fitted through a model of the supervised learning algorithm until the model has generalization capability, and then quotient and order identification is carried out through the model.
However, the supervised learning algorithm needs to adopt different algorithm models to be adapted to the data types according to different data types, because the different algorithm models have great differences in the aspects of model construction, data input, parameter fitting and the like. Input data corresponding to the business order identification generally comprises multiple data types, and the existing business order identification scheme usually adopts a single-class algorithm model for identification, cannot be suitable for the input data comprising the multiple data types, and is easy to generate a phenomenon of model overfitting, so that the accuracy of an identification result is influenced.
Disclosure of Invention
Embodiments of the present invention provide a business order recognition method and apparatus based on multimodal data, a computer-readable storage medium, and a terminal device, which can ensure integrity of data features by performing feature extraction and effective feature fusion on multimodal data, and can be applied to multimodal data processing by performing business order recognition with a multimodal integrated learning system, thereby improving accuracy of recognition results.
In order to achieve the above object, an embodiment of the present invention provides a quotient sheet identification method based on multimodal data, including:
acquiring data to be identified from a network platform, and extracting multi-modal data in the data to be identified; wherein the multimodal data comprises textual data, category data, and numerical data;
respectively extracting the features of the text data, the category data and the numerical data to correspondingly obtain a text feature vector, a category feature vector and a numerical feature vector;
performing feature fusion on the text feature vector, the category feature vector and the numerical feature vector to obtain a fusion feature vector;
identifying the fusion characteristic vector according to a preset multi-model ensemble learning system to obtain a business order identification result; the multi-model ensemble learning system is generated in a deep stacking mode and comprises at least two network layers, and each network layer comprises at least two basic models.
Further, the multi-model ensemble learning system comprises two network layers; then, the identifying the fusion feature vector according to a preset multi-model ensemble learning system to obtain a quotient list identification result, specifically including:
respectively inputting the fusion feature vectors into each basic model of a first layer network layer of the multi-model ensemble learning system for recognition, and correspondingly obtaining at least two first recognition results;
aggregating the at least two first recognition results to obtain an initial recognition result;
inputting the initial recognition results into each basic model of a second layer network layer of the multi-model integrated learning system respectively for recognition, and correspondingly obtaining at least two second recognition results;
and aggregating the at least two second identification results to obtain the business order identification result.
Further, the aggregating the at least two first recognition results to obtain an initial recognition result specifically includes:
and aggregating the at least two first recognition results in an averaging or splicing mode to obtain the initial recognition result.
Further, the aggregating the at least two second recognition results to obtain the quotient sheet recognition result specifically includes:
and aggregating the at least two second identification results in an averaging mode to obtain the quotient sheet identification result.
Further, the first network layer comprises five basic models, wherein the five basic models comprise a fully connected neural network, two LightGBM models with different parameters, an extreme stochastic tree and a castboost gradient tree.
Further, the second layer network layer comprises five basic models, wherein the five basic models comprise a fully connected neural network, a LightGBM model with two different parameters, an extreme random tree and a Catboost gradient tree.
Further, the respectively performing feature extraction on the text data, the category data, and the numerical data to correspondingly obtain a text feature vector, a category feature vector, and a numerical feature vector specifically includes:
inputting the text data into a pre-training language model for feature extraction to obtain the text feature vector;
inputting the category data into an embedding layer and a full-connection neural network in sequence for feature extraction to obtain the category feature vector;
and splicing the numerical data into a numerical vector, and inputting the numerical vector into a fully-connected neural network for feature extraction to obtain the numerical feature vector.
Further, the performing feature fusion on the text feature vector, the category feature vector, and the numerical feature vector to obtain a fusion feature vector specifically includes:
and performing feature fusion on the text feature vector, the category feature vector and the numerical feature vector in an averaging or splicing mode to obtain the fusion feature vector.
Further, the text data comprises a text sending title, a text sending body and text sending comment contents, the category data comprises the place where the text sending belongs and whether the text sending belongs to business activities, and the numerical data comprises the reading number of the text sending and the fan number of a text sending account.
In order to achieve the above object, an embodiment of the present invention further provides a business order recognition apparatus based on multimodal data, which is used for implementing any one of the business order recognition methods based on multimodal data described above, and the apparatus includes:
the multi-mode data acquisition module is used for acquiring data to be identified from a network platform and extracting multi-mode data in the data to be identified; wherein the multimodal data comprises text data, category data, and numerical data;
the multi-mode feature extraction module is used for respectively extracting features of the text data, the category data and the numerical data to correspondingly obtain a text feature vector, a category feature vector and a numerical feature vector;
the multi-mode feature fusion module is used for performing feature fusion on the text feature vector, the category feature vector and the numerical feature vector to obtain a fusion feature vector;
the business order identification module is used for identifying the fusion characteristic vector according to a preset multi-model integrated learning system to obtain a business order identification result; the multi-model ensemble learning system is generated in a deep stacking mode and comprises at least two network layers, and each network layer comprises at least two basic models.
An embodiment of the present invention further provides a computer-readable storage medium, where the computer-readable storage medium includes a stored computer program; wherein the computer program controls a device where the computer readable storage medium is located to execute any one of the above methods for identifying a business order based on multimodal data.
The embodiment of the present invention further provides a terminal device, which includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, where the processor implements any one of the above-mentioned methods for identifying a business order based on multimodal data when executing the computer program.
Compared with the prior art, the embodiment of the invention provides a business order identification method and device based on multi-modal data, a computer readable storage medium and terminal equipment, firstly, data to be identified are obtained from a network platform, and multi-modal data in the data to be identified are extracted, wherein the multi-modal data comprise text data, category data and numerical data; then, respectively extracting the characteristics of the text data, the category data and the numerical data to correspondingly obtain a text characteristic vector, a category characteristic vector and a numerical characteristic vector, and performing characteristic fusion on the text characteristic vector, the category characteristic vector and the numerical characteristic vector to obtain a fusion characteristic vector; finally, identifying the fusion characteristic vector according to a preset multi-model ensemble learning system to obtain a quotient list identification result, wherein the multi-model ensemble learning system is generated in a deep layer stacking mode and comprises at least two network layers, and each network layer comprises at least two basic models; the embodiment of the invention can ensure the integrity of data characteristics by respectively extracting the characteristics of the multi-mode data and carrying out effective characteristic fusion, and can be suitable for multi-mode data processing by adopting a multi-model integrated learning system to carry out business order recognition, thereby improving the accuracy of a recognition result.
Drawings
FIG. 1 is a flow chart of a preferred embodiment of a business order recognition method based on multimodal data provided by the present invention;
FIG. 2 is a block diagram of a preferred embodiment of a business form identification using a multi-model ensemble learning system according to the present invention;
FIG. 3 is a block diagram of a preferred embodiment of a business order recognition apparatus based on multi-modal data according to the present invention;
fig. 4 is a block diagram of a preferred embodiment of a terminal device according to the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without any inventive step, are within the scope of the present invention.
The embodiment of the invention provides a business form recognition method based on multi-modal data, which is a flow chart of a preferred embodiment of the business form recognition method based on multi-modal data, shown in fig. 1, and the method comprises steps S11 to S14:
s11, acquiring data to be identified from a network platform, and extracting multi-modal data in the data to be identified; wherein the multimodal data includes text data, category data, and numerical data.
Specifically, data to be identified (for example, account data and text data of a certain account collected from a social media platform) may be collected and acquired from a network platform such as an e-commerce platform, and the obtained data to be identified is classified to divide the data to be identified into a plurality of data types, and multimodal data corresponding to the data to be identified is extracted accordingly, where the extracted multimodal data includes, but is not limited to, text data (the corresponding data type is a text type), category data (the corresponding data type is a category type), and numerical data (the corresponding data type is a numerical value type).
As a preferred scheme, the text data comprises a text sending title, a text sending body and text sending comment contents, the category data comprises the place where the text sending belongs and whether the text sending belongs to business activities, and the numerical data comprises the reading number of the text sending and the fan number of a text sending account.
Specifically, the text data of the text type includes, but is not limited to, at least one of a text title, a text body and text comment content, the category data of the category type includes, but is not limited to, at least one of a place where the text belongs to and whether the text belongs to a merchant activity (or platform activity), and the numerical data of the numerical type includes, but is not limited to, at least one of a reading number of the text and a fan number of a text account.
And S12, respectively carrying out feature extraction on the text data, the category data and the numerical data to correspondingly obtain a text feature vector, a category feature vector and a numerical feature vector.
Specifically, after obtaining the multi-modal data, feature extraction is respectively performed on text data in the multi-modal data to correspondingly obtain text feature vectors, feature extraction is performed on category data in the multi-modal data to correspondingly obtain category feature vectors, feature extraction is performed on numerical data in the multi-modal data to correspondingly obtain numerical feature vectors.
And S13, performing feature fusion on the text feature vector, the category feature vector and the numerical feature vector to obtain a fusion feature vector.
Specifically, after the text feature vector, the category feature vector and the numerical feature vector are obtained, feature fusion processing can be performed on the text feature vector, the category feature vector and the numerical feature vector, and fusion feature vectors are correspondingly obtained, so that feature vectors corresponding to different data types are converted into uniform feature types, and effective feature fusion is realized.
S14, identifying the fusion feature vector according to a preset multi-model integrated learning system to obtain a business form identification result; the multi-model ensemble learning system is generated in a deep stacking mode and comprises at least two network layers, and each network layer comprises at least two basic models.
Specifically, a deep-layer stacking mode is adopted in advance to generate a multi-model ensemble learning system, the multi-model ensemble learning system is composed of at least two network layers, each network layer is composed of at least two basic models, the obtained fusion feature vectors are input into the multi-model ensemble learning system to be recognized, a business order recognition result is correspondingly obtained, and whether the data to be recognized is a business order can be judged according to the obtained business order recognition result.
For example, assuming that the obtained quotient sheet recognition result is represented by a quotient sheet confidence level, when the obtained quotient sheet confidence level is greater than a preset confidence level threshold (for example, preset confidence level threshold = 0.5), it may be determined that the data to be recognized is the quotient sheet, and when the obtained quotient sheet confidence level is less than or equal to the preset confidence level threshold, it may be determined that the data to be recognized is not the quotient sheet.
It should be noted that, in the field of machine learning, the recognition capability of using multiple models simultaneously is usually stronger than that of using a single model, and since the business order identification has unique multi-modal data, the single model is difficult to directly fit the integral data to be recognized, or cannot be free from various constraints of machine learning, such as long parameter optimization time, process flexibility, and the like, in the embodiment of the present invention, a multi-model integrated learning system on the multi-modal data level is provided, and after the fusion feature vector is obtained, the fusion feature vector can be input into the multi-model integrated learning system for recognition, so that the multiple machine learning models can be effectively utilized to realize high-efficiency and high-quality recognition.
Furthermore, in the existing scheme using multi-model integration, the currently used model integration mode is generally a relatively shallow fusion mode, and because the complexity in the aspects of the parameter quantity, the model architecture and the like of a multi-modal algorithm model is continuously increased, the fusion effect of different models is poor, and the integration requirement of deep fusion of the multi-modal algorithm model cannot be met.
According to the business order identification method based on the multi-modal data, provided by the embodiment of the invention, through respectively extracting the features of the multi-modal data and carrying out effective feature fusion, even if the data of different types are mutually separated, the whole data is divided into the data of multiple types, the integrity of the data features can be ensured through the feature fusion of the multi-modal data, and through adopting the multi-model integrated learning system to carry out business order identification, the thought of integrated learning is utilized, the defect that a single model consumes time and labor in the aspects of model design, parameter optimization and the like is overcome, the business order identification method can be suitable for multi-modal data processing, the phenomenon of model overfitting is avoided, and therefore the accuracy of an identification result is improved.
In another preferred embodiment, the multi-model ensemble learning system includes two network layers; then, the identifying the fusion feature vector according to a preset multi-model ensemble learning system to obtain a quotient list identification result, specifically including:
respectively inputting the fusion feature vectors into each basic model of a first layer network layer of the multi-model ensemble learning system for recognition, and correspondingly obtaining at least two first recognition results;
aggregating the at least two first recognition results to obtain an initial recognition result;
inputting the initial recognition results into each basic model of a second layer network layer of the multi-model integrated learning system for recognition, and correspondingly obtaining at least two second recognition results;
and aggregating the at least two second identification results to obtain the business order identification result.
Specifically, in combination with the above embodiment, the multi-model integrated learning system is composed of at least two network layers, in this embodiment, the multi-model integrated learning system specifically includes two network layers, namely a first network layer and a second network layer, and similarly, the first network layer is composed of at least two basic models, and the second network layer is composed of at least two basic models, then, when identifying the obtained fusion feature vector according to the multi-model integrated learning system, the fusion feature vector may be input into each basic model of the first network layer, each basic model in the first network layer is used for identification, and one basic model correspondingly outputs one first identification result, so as to obtain at least two first identification results correspondingly, and aggregate the obtained at least two first identification results, so as to obtain an initial identification result correspondingly; and then inputting the obtained initial recognition results into each basic model of the second layer network layer, respectively recognizing by using each basic model in the second layer network layer, correspondingly outputting a second recognition result by using one basic model, correspondingly obtaining at least two second recognition results, and carrying out aggregation processing on the obtained at least two second recognition results to correspondingly obtain a business order recognition result.
The first layer network layer is equivalent to a basic model of the second layer network layer, the input of the first layer network layer is a fusion feature vector obtained through feature fusion processing, the output results (namely, at least two first recognition results) of the first layer network layer are used as the input of the second layer network layer after aggregation processing, and the output results (namely, at least two second recognition results) of the second layer network layer are used as the final quotient list recognition result after aggregation processing.
It should be noted that the number of network layers in the multi-model ensemble learning system is preferably two, and on the premise of satisfying at least two network layers, three layers, four layers, and the like may also be set, however, the more the number of layers is, the more complicated the system is, the longer the time is consumed in the recognition process, and the better the recognition effect is, so the specific number of layers may be set according to actual needs, and the embodiment of the present invention is not specifically limited.
As an improvement of the above scheme, the aggregating the at least two first recognition results to obtain an initial recognition result specifically includes:
and aggregating the at least two first recognition results in an averaging or splicing mode to obtain the initial recognition result.
Specifically, with reference to the foregoing embodiment, when performing aggregation processing on at least two first identification results output by the first network layer, aggregation processing may be performed on the at least two first identification results in an averaging manner or a splicing manner, so as to obtain an initial identification result correspondingly.
As an improvement of the foregoing solution, the aggregating the at least two second recognition results to obtain the quotient sheet recognition result specifically includes:
and aggregating the at least two second identification results in an averaging mode to obtain the quotient list identification result.
Specifically, with reference to the foregoing embodiment, when aggregating at least two second identification results output by the second layer network layer, the aggregation processing may be performed on the at least two second identification results in an averaging manner, so as to obtain the quotient sheet identification result correspondingly.
Exemplarily, referring to fig. 2, it is a block diagram of a preferred embodiment of the present invention for identifying a business order by using a multi-model ensemble learning system, where the multi-model ensemble learning system in fig. 2 includes two network layers, a first network layer includes N basic models, which are respectively a basic model 11, a basic model 12, and a basic model 8230, a second network layer also includes N basic models, which are respectively a basic model 21, a basic model 22, and a basic model 2N, where N is greater than or equal to 2 and N is an integer; when the business form recognition is carried out, firstly, obtained fusion feature vectors are respectively input into a basic model 11, a basic model 12, 8230, N first recognition results are correspondingly obtained in a basic model 1N, aggregation (averaging or splicing) processing is carried out on the N first recognition results, initial recognition results are correspondingly obtained, then the initial recognition results are respectively input into a basic model 21, a basic model 22, 8230, and a basic model 2N, N second recognition results are correspondingly obtained, aggregation (averaging) processing is carried out on the N second recognition results, and accordingly, the business form recognition results are obtained.
As an improvement of the above scheme, the first network layer includes five basic models, and the five basic models include a fully-connected neural network, two LightGBM models with different parameters, an extremely random tree, and a castboost gradient tree.
Specifically, with reference to the foregoing embodiment, the first network layer is formed by at least two basic models, and in this embodiment, the first network layer specifically includes five basic models, which are respectively a fully-connected neural network, two Light Gradient Boosting Machine (Light gbm) models, an extreme random tree (extreme random Trees) and a tree Gradient boosted tree (tree Trees).
The fully-connected neural network can adopt a ReLU activation function, the parameters of the LightGBM model can use model default parameters, and the two LightGBM models adopt different model parameters.
It should be noted that, the number of the basic models in the first network layer is preferably five, and may also be set to be three, four, six, and the like on the premise that at least two basic models are included, however, the number of the basic models is different, and the influence on the final recognition result is also different, and the specific number of the basic models may be set according to the actual requirement, and the embodiment of the present invention is not particularly limited.
As an improvement of the above scheme, the second layer network layer includes five basic models, and the five basic models include a fully-connected neural network, two LightGBM models with different parameters, an extreme random tree, and a castboost gradient tree.
Specifically, in combination with the above embodiment, the second layer network layer is formed by at least two basic models, and in this embodiment, the second layer network layer specifically includes five basic models, which are respectively a fully connected neural network, two LightGBM models, an extreme random tree, and a castboost gradient tree.
The fully-connected neural network can adopt a ReLU activation function, the parameters of the LightGBM model can use model default parameters, and the two LightGBM models adopt different model parameters.
It should be noted that, the number of the basic models in the second layer network layer is preferably five, and on the premise that the number of the basic models includes at least two basic models, the basic models may also be set to be three, four, six, and the like, however, the number of the basic models is different, the influence on the final recognition result is also different, the specific number of the basic models may be set according to the actual requirement, and the embodiment of the present invention is not particularly limited.
It can be understood that the configuration of the first network layer and the configuration of the second network layer may be the same or different, and the influence of different configurations on the final recognition result is also different.
The embodiment of the invention effectively utilizes the existing basic model to generate the multi-model integrated learning system, and can realize the characteristics of easy use and high efficiency of the system.
In another preferred embodiment, the performing feature extraction on the text data, the category data, and the numerical data respectively to obtain a text feature vector, a category feature vector, and a numerical feature vector correspondingly specifically includes:
inputting the text data into a pre-training language model for feature extraction to obtain the text feature vector;
inputting the category data into an embedding layer and a full-connection neural network in sequence for feature extraction to obtain the category feature vector;
and splicing the numerical data into a numerical vector, and inputting the numerical vector into a fully-connected neural network for feature extraction to obtain the numerical feature vector.
Specifically, in combination with the above embodiment, when feature extraction is performed on text data in the multimodal data, the text data may be input into a pre-training language model for feature extraction, and a text feature vector is correspondingly obtained; the pre-training language model can adopt a general BERT (Bidirectional Encoder replication from transformations) type pre-training language model, and for text data, corresponding processing is performed sequentially through word segmentation, a word embedding layer and a self-attention layer of a sentence level, so that a numerical Representation of the text data is obtained, namely a text feature vector is obtained.
It can be understood that, in order to enable the pre-training language model to better adapt to the language in the specific field, the pre-training language model may be finely tuned by using the corpus in the specific field, and the language model after the fine tuning may better characterize the text data in the specific field.
When feature extraction is carried out on category data in multi-modal data, the category data can be input into an independent embedding layer, a vector representation is output through the embedding layer, then the vector representation is input into a full-connection neural network for feature extraction, and category feature vectors are correspondingly obtained; in most machine learning projects, vectors obtained by an embedding layer generally represent unique heat vectors, and trainable embedding layers are adopted, parameters of the embedding layer are obtained by setting a specific machine learning task, and then the parameters are applied to other projects.
When the feature extraction is performed on numerical data in the multi-modal data, the numerical data can be spliced into a numerical vector, and then the numerical vector is input into a fully-connected neural network for feature extraction, so that a numerical feature vector is correspondingly obtained.
It should be noted that, in the embodiment of the present invention, each type of data in the multi-modal data adopts the most suitable feature extraction manner, so that features of various types of data can be effectively retained, and an accurate feature extraction result can be obtained; for example, if the multimodal data also includes picture data, a network model suitable for pictures can be used for feature extraction.
In another preferred embodiment, the performing feature fusion on the text feature vector, the category feature vector, and the numerical feature vector to obtain a fused feature vector specifically includes:
and performing feature fusion on the text feature vector, the category feature vector and the numerical feature vector in an averaging or splicing mode to obtain the fusion feature vector.
Specifically, with reference to the foregoing embodiment, when performing feature fusion processing on the obtained text feature vector, category feature vector, and numeric feature vector, feature fusion may be performed on the text feature vector, the category feature vector, and the numeric feature vector in an averaging manner or a splicing manner, so as to obtain a fusion feature vector accordingly.
It should be noted that, if the feature fusion is performed in an averaging manner, it is necessary to ensure that the text feature vector, the category feature vector, and the numerical feature vector have the same dimension, and if the feature fusion is performed in a splicing manner, there is no requirement for the vector dimension, and there is no requirement for the splicing order.
For example, assuming that a text feature vector is represented as V1, a category feature vector is represented as V2, and a numeric feature vector is represented as V3, and the three have the same dimension, if feature fusion is performed in an averaging manner, a fusion feature vector V = (V1 + V2+ V3)/3 is obtained, and if feature fusion is performed in a splicing manner, a fusion feature vector V = (V1, V2, V3) is obtained.
It should be noted that, by aggregating the feature vectors corresponding to the data of the three data types in an averaging or splicing manner, the feature fusion manner is simple and easy to use, and a better balance is achieved between the type independence and the global fusion of the data.
For example, with reference to all the embodiments described above, the following takes a specific data to be identified as an example to describe the overall flow of the business form identification method provided by the embodiment of the present invention:
the text data in the data to be recognized is:
(1) Text: the activity of the merchant is very preferential and the merchant is going to buy the product at a high speed. ";
the category data in the data to be identified is as follows:
(1) Whether the text sending account is a merchant authentication account is as follows: if not;
(2) Whether the message participates in the platform activity: is that;
numerical data in the data to be identified are as follows:
(1) The number of the fans of the text sending account is as follows: 100 000;
(1) Number of reads of the text: 50 000;
respectively extracting characteristics of text data, category data and numerical data in data to be identified, and correspondingly obtaining a text characteristic vector, a category characteristic vector and a numerical characteristic vector; wherein the text feature vector is represented as array1 ([ [ a1, a2, a3, \8230 ]; (middle omitted, which is a (1, 768) -dimensional vector), a767, a768] ], dtype = float 32); each class is represented as a unique heat vector shaped as [0,1] which is transformed via a fully connected neural network into a class feature vector, represented as array2 ([ [ b1, b2, b3, \8230; (omitted in the middle, this is a vector of dimension (1, 768)), b767, b768] ], dtype = float 32); the numeric feature vector is denoted array3 ([ [ c1, c2, c3, \ 8230; (omitted in the middle, this is a (1, 768) -dimensional vector), c767, c768] ], dtype = float 32);
performing feature fusion on the text feature vector array1, the category feature vector array2 and the numerical feature vector array3 by adopting an averaging mode to correspondingly obtain a fusion feature vector which is expressed as array4 ([ [ d1, d2, d3, \ 8230 ], (omitted in the middle, which is a vector with (1, 768) dimensions), d767, d768] ], dtype = float 32);
taking the fusion feature vector array4 as the input of the multi-model ensemble learning system, and obtaining the following five second recognition results through forward calculation of two network layers:
the second recognition result output by the fully-connected neural network is as follows: 0.90, which indicates that the confidence of 'being the quotient order' in the result of 'being the quotient order' identified by the fully-connected neural network is 0.90;
the second recognition result output by the LightGBM model is: 0.73, which indicates whether the result of "being a quotient order" identified by the LightGBM model has a confidence of 0.73;
the second recognition result output by the second LightGBM model is: 0.51, which indicates that the confidence of "being a quotient sheet" in the result of "being a quotient sheet" identified by the LightGBM model is 0.51;
the second recognition result of the extremely random tree output is: 0.45, indicating that in the result of "whether it is a quotient order" identified by an extremely random tree, the confidence of "being a quotient order" is 0.45;
the second recognition result output by the Catboost gradient boosting tree is as follows: 0.54, indicating that in the result of "being a quotient sheet" identified by a Catboost gradient tree, the confidence of "being a quotient sheet" is 0.54;
and performing aggregation processing on the five second identification results by adopting an averaging mode, and finally obtaining a quotient list identification result of 0.62, wherein if the preset confidence threshold =0.5 and 0.62 is greater than 0.5, the data to be identified is judged to be the quotient list.
An embodiment of the present invention further provides a device for identifying a business form based on multi-modal data, which is used to implement the method for identifying a business form based on multi-modal data according to any of the above embodiments, and is shown in fig. 3, which is a structural block diagram of an preferred embodiment of the device for identifying a business form based on multi-modal data according to the present invention, where the device includes:
the multi-modal data acquisition module 11 is configured to acquire data to be identified from a network platform and extract multi-modal data in the data to be identified; wherein the multimodal data comprises text data, category data, and numerical data;
the multi-modal feature extraction module 12 is configured to perform feature extraction on the text data, the category data, and the numerical data, respectively, and correspondingly obtain a text feature vector, a category feature vector, and a numerical feature vector;
a multi-modal feature fusion module 13, configured to perform feature fusion on the text feature vector, the category feature vector, and the numerical feature vector to obtain a fusion feature vector;
the business order recognition module 14 is used for recognizing the fusion feature vector according to a preset multi-model integrated learning system to obtain a business order recognition result; the multi-model integrated learning system is generated in a deep stacking mode and comprises at least two network layers, and each network layer comprises at least two basic models.
Preferably, the multi-model ensemble learning system includes two network layers; then, the business form identification module 14 specifically includes:
the first identification unit is used for respectively inputting the fusion feature vectors into each basic model of a first layer network layer of the multi-model ensemble learning system for identification, and correspondingly obtaining at least two first identification results;
the first aggregation unit is used for aggregating the at least two first identification results to obtain an initial identification result;
the second identification unit is used for respectively inputting the initial identification results into each basic model of a second layer network layer of the multi-model ensemble learning system for identification, and correspondingly obtaining at least two second identification results;
and the second aggregation unit is used for aggregating the at least two second identification results to obtain the quotient sheet identification result.
Preferably, the first polymerization unit is specifically configured to:
and aggregating the at least two first recognition results in an averaging or splicing mode to obtain the initial recognition result.
Preferably, said second polymerization unit is used in particular for:
and aggregating the at least two second identification results in an averaging mode to obtain the quotient list identification result.
Preferably, the first network layer includes five basic models including a fully-connected neural network, two parametrically different LightGBM models, an extreme stochastic tree, and a castboost gradient tree.
Preferably, the second layer network layer comprises five basic models, wherein the five basic models comprise a fully connected neural network, two LightGBM models with different parameters, an extreme random tree and a castboost gradient tree.
Preferably, the multi-modal feature extraction module 12 specifically includes:
the text feature vector extraction unit is used for inputting the text data into a pre-training language model for feature extraction to obtain the text feature vector;
the category characteristic vector extraction unit is used for sequentially inputting the category data into an embedding layer and a full-connection neural network for characteristic extraction to obtain the category characteristic vector;
and the numerical value feature vector extraction unit is used for splicing the numerical value data into a numerical value vector, inputting the numerical value vector into a fully-connected neural network for feature extraction, and obtaining the numerical value feature vector.
Preferably, the multimodal feature fusion module 13 is specifically configured to:
and performing feature fusion on the text feature vector, the category feature vector and the numerical feature vector in an averaging or splicing mode to obtain the fusion feature vector.
Preferably, the text data comprises a text sending title, a text sending body and text sending comment contents, the category data comprises the place where the text sending belongs and whether the text sending belongs to business activities, and the numerical data comprises the reading number of the text sending and the fan number of a text sending account.
It should be noted that, the business order recognition apparatus based on multimodal data provided in the embodiment of the present invention can implement all processes of the business order recognition method based on multimodal data described in any one of the embodiments above, and the functions and implemented technical effects of each module and unit in the apparatus are respectively the same as those of the business order recognition method based on multimodal data described in the embodiment above, and are not described herein again.
An embodiment of the present invention further provides a computer-readable storage medium, where the computer-readable storage medium includes a stored computer program; when the computer program runs, the device where the computer readable storage medium is located is controlled to execute the method for identifying a quotient list based on multi-modal data according to any one of the above embodiments.
An embodiment of the present invention further provides a terminal device, as shown in fig. 4, which is a block diagram of a preferred embodiment of the terminal device provided in the present invention, the terminal device includes a processor 10, a memory 20, and a computer program stored in the memory 20 and configured to be executed by the processor 10, and when the computer program is executed, the processor 10 implements the business order identification method based on multimodal data according to any of the above embodiments.
Preferably, the computer program can be divided into one or more modules/units (e.g. computer program 1, computer program 2,) which are stored in the memory 20 and executed by the processor 10 to accomplish the present invention. The one or more modules/units may be a series of computer program instruction segments capable of performing specific functions, which are used for describing the execution process of the computer program in the terminal device.
The Processor 10 may be a Central Processing Unit (CPU), other general purpose Processor, a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field Programmable Gate Array (FPGA) or other Programmable logic device, a discrete Gate or transistor logic device, a discrete hardware component, etc., the general purpose Processor may be a microprocessor, or the Processor 10 may be any conventional Processor, the Processor 10 is a control center of the terminal device, and various interfaces and lines are used to connect various parts of the terminal device.
The memory 20 mainly includes a program storage area that may store an operating system, an application program required for at least one function, and the like, and a data storage area that may store related data and the like. In addition, the memory 20 may be a high speed random access memory, a non-volatile memory such as a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) Card, a Flash Card (Flash Card), and the like, or the memory 20 may be other volatile solid state memory devices.
It should be noted that the terminal device may include, but is not limited to, a processor and a memory, and those skilled in the art will understand that the structural block diagram in fig. 4 is only an example of the terminal device, and does not constitute a limitation of the terminal device, and may include more or less components than those shown in the figure, or may combine some components, or may include different components.
To sum up, according to the business order identification method and apparatus, the computer-readable storage medium and the terminal device based on the multi-modal data provided by the embodiment of the present invention, firstly, data to be identified is obtained from a network platform, and multi-modal data in the data to be identified is extracted, where the multi-modal data includes text data, category data and numerical data; then, respectively extracting the characteristics of the text data, the category data and the numerical data to correspondingly obtain a text characteristic vector, a category characteristic vector and a numerical characteristic vector, and performing characteristic fusion on the text characteristic vector, the category characteristic vector and the numerical characteristic vector to obtain a fusion characteristic vector; finally, identifying the fusion characteristic vector according to a preset multi-model ensemble learning system to obtain a quotient list identification result, wherein the multi-model ensemble learning system is generated in a deep layer stacking mode and comprises at least two network layers, and each network layer comprises at least two basic models; according to the embodiment of the invention, through respectively extracting the features of the multi-mode data and effectively fusing the features, even if the different types of data are separated from each other and the overall data is divided into the multiple types of data, the integrity of the data features can be ensured through the feature fusion of the multi-mode data, and through adopting a multi-model integrated learning system to carry out business order identification, and utilizing the thought of integrated learning, the defects of time and labor consumption of a single model in the aspects of model design, parameter optimization and the like are overcome, the multi-mode data processing can be suitable, the phenomenon of over-fitting of the model is avoided, so that the accuracy of the identification result is improved, and meanwhile, the multi-model integrated learning system adopts a deep stacking mode and has stronger learning capability compared with the existing shallow integrated learning mode.
The above description is only a preferred embodiment of the present invention, and it should be noted that, for those skilled in the art, it is possible to make various improvements and modifications without departing from the technical principle of the present invention, and those improvements and modifications should be considered as the protection scope of the present invention.
Claims (12)
1. A business order recognition method based on multi-modal data is characterized by comprising the following steps:
acquiring data to be identified from a network platform, and extracting multi-modal data in the data to be identified; wherein the multimodal data comprises text data, category data, and numerical data;
respectively extracting the characteristics of the text data, the category data and the numerical data to correspondingly obtain a text characteristic vector, a category characteristic vector and a numerical characteristic vector;
performing feature fusion on the text feature vector, the category feature vector and the numerical value feature vector to obtain a fusion feature vector;
identifying the fusion feature vector according to a preset multi-model integrated learning system to obtain a business form identification result; the multi-model integrated learning system is generated in a deep stacking mode and comprises at least two network layers, and each network layer comprises at least two basic models.
2. The multi-modal data-based business form identification method of claim 1, wherein the multi-model ensemble learning system comprises two network layers; then, the identifying the fusion feature vector according to a preset multi-model ensemble learning system to obtain a quotient list identification result, specifically including:
respectively inputting the fusion feature vectors into each basic model of a first layer network layer of the multi-model ensemble learning system for recognition, and correspondingly obtaining at least two first recognition results;
aggregating the at least two first recognition results to obtain an initial recognition result;
inputting the initial recognition results into each basic model of a second layer network layer of the multi-model integrated learning system respectively for recognition, and correspondingly obtaining at least two second recognition results;
and aggregating the at least two second identification results to obtain the business form identification result.
3. The quotient and bill identification method based on multi-modal data as claimed in claim 2, wherein said aggregating the at least two first identification results to obtain an initial identification result comprises:
and aggregating the at least two first recognition results in an averaging or splicing mode to obtain the initial recognition result.
4. The quotient sheet recognition method based on multi-modal data as recited in claim 2, wherein the aggregating the at least two second recognition results to obtain the quotient sheet recognition result specifically comprises:
and aggregating the at least two second identification results in an averaging mode to obtain the quotient sheet identification result.
5. The method of claim 2, wherein the first network layer comprises five base models, the five base models comprising a fully connected neural network, two parametrically distinct LightGBM models, an extreme stochastic tree, and a castboost gradient tree.
6. The method of claim 2, wherein the second tier network layer comprises five base models, the five base models comprising a fully connected neural network, a LightGBM model with two different parameters, an extreme stochastic tree, and a castboost gradient tree.
7. The method for identifying quotient sheet based on multi-modal data according to claim 1, wherein the performing feature extraction on the text data, the category data and the numerical data respectively to obtain a text feature vector, a category feature vector and a numerical feature vector correspondingly comprises:
inputting the text data into a pre-training language model for feature extraction to obtain the text feature vector;
inputting the category data into an embedding layer and a full-connection neural network in sequence for feature extraction to obtain the category feature vector;
and splicing the numerical data into a numerical vector, and inputting the numerical vector into a fully-connected neural network for feature extraction to obtain the numerical feature vector.
8. The method for identifying a quotient list based on multi-modal data according to claim 1, wherein the performing feature fusion on the text feature vector, the category feature vector and the numerical feature vector to obtain a fused feature vector specifically comprises:
and performing feature fusion on the text feature vector, the category feature vector and the numerical feature vector in an averaging or splicing mode to obtain the fusion feature vector.
9. The method for business order recognition based on multi-modal data as claimed in any one of claims 1 to 8, wherein the text data comprises a text title, a text body and text comment contents, the category data comprises the location of the text and whether the text belongs to business activities, and the numerical data comprises the reading number of the text and the fan number of a text account.
10. A device for identifying a business order based on multi-modal data, which is used for implementing the method for identifying the business order based on multi-modal data according to any one of claims 1 to 9, and comprises:
the multi-mode data acquisition module is used for acquiring data to be identified from a network platform and extracting multi-mode data in the data to be identified; wherein the multimodal data comprises text data, category data, and numerical data;
the multi-mode feature extraction module is used for respectively extracting features of the text data, the category data and the numerical data to correspondingly obtain a text feature vector, a category feature vector and a numerical feature vector;
the multi-mode feature fusion module is used for performing feature fusion on the text feature vector, the category feature vector and the numerical feature vector to obtain a fusion feature vector;
the business form identification module is used for identifying the fusion feature vector according to a preset multi-model integrated learning system to obtain a business form identification result; the multi-model integrated learning system is generated in a deep stacking mode and comprises at least two network layers, and each network layer comprises at least two basic models.
11. A computer-readable storage medium, characterized in that the computer-readable storage medium comprises a stored computer program; wherein the computer program controls a device on which the computer-readable storage medium is located to execute the method for identifying a business order based on multimodal data as claimed in any one of claims 1 to 9 when running.
12. A terminal device comprising a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor, when executing the computer program, implements the multi-modal data based commerce identification method according to any one of claims 1 to 9.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211523510.8A CN115601771A (en) | 2022-12-01 | 2022-12-01 | Business order identification method, device, medium and terminal equipment based on multi-mode data |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211523510.8A CN115601771A (en) | 2022-12-01 | 2022-12-01 | Business order identification method, device, medium and terminal equipment based on multi-mode data |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115601771A true CN115601771A (en) | 2023-01-13 |
Family
ID=84852463
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211523510.8A Pending CN115601771A (en) | 2022-12-01 | 2022-12-01 | Business order identification method, device, medium and terminal equipment based on multi-mode data |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115601771A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116071740A (en) * | 2023-03-06 | 2023-05-05 | 深圳前海环融联易信息科技服务有限公司 | Invoice identification method, computer equipment and storage medium |
| CN116452241A (en) * | 2023-04-17 | 2023-07-18 | 广西财经学院 | User loss probability calculation method based on multi-mode fusion neural network |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20170024186A1 (en) * | 2015-03-02 | 2017-01-26 | Slyce Holdings Inc. | System and method for clustering data |
| CN109522917A (en) * | 2018-09-10 | 2019-03-26 | 中山大学 | A method of fusion forecasting is stacked based on multi-model |
| CN111047413A (en) * | 2019-12-17 | 2020-04-21 | 秒针信息技术有限公司 | Text content acquisition method and device, computer equipment and readable storage medium |
| CN111489065A (en) * | 2020-03-27 | 2020-08-04 | 北京理工大学 | Node risk assessment integrating ICT supply chain network topology and product business information |
| CN112613544A (en) * | 2020-12-16 | 2021-04-06 | 北京迈格威科技有限公司 | Target detection method, device, electronic equipment and computer readable medium |
| CN113630302A (en) * | 2020-05-09 | 2021-11-09 | 阿里巴巴集团控股有限公司 | Junk mail identification method and device and computer readable storage medium |
| CN114003700A (en) * | 2021-10-27 | 2022-02-01 | 浙江诺诺网络科技有限公司 | Method and system for processing session information, electronic device and storage medium |
-
2022
- 2022-12-01 CN CN202211523510.8A patent/CN115601771A/en active Pending
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20170024186A1 (en) * | 2015-03-02 | 2017-01-26 | Slyce Holdings Inc. | System and method for clustering data |
| CN109522917A (en) * | 2018-09-10 | 2019-03-26 | 中山大学 | A method of fusion forecasting is stacked based on multi-model |
| CN111047413A (en) * | 2019-12-17 | 2020-04-21 | 秒针信息技术有限公司 | Text content acquisition method and device, computer equipment and readable storage medium |
| CN111489065A (en) * | 2020-03-27 | 2020-08-04 | 北京理工大学 | Node risk assessment integrating ICT supply chain network topology and product business information |
| CN113630302A (en) * | 2020-05-09 | 2021-11-09 | 阿里巴巴集团控股有限公司 | Junk mail identification method and device and computer readable storage medium |
| CN112613544A (en) * | 2020-12-16 | 2021-04-06 | 北京迈格威科技有限公司 | Target detection method, device, electronic equipment and computer readable medium |
| CN114003700A (en) * | 2021-10-27 | 2022-02-01 | 浙江诺诺网络科技有限公司 | Method and system for processing session information, electronic device and storage medium |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116071740A (en) * | 2023-03-06 | 2023-05-05 | 深圳前海环融联易信息科技服务有限公司 | Invoice identification method, computer equipment and storage medium |
| CN116452241A (en) * | 2023-04-17 | 2023-07-18 | 广西财经学院 | User loss probability calculation method based on multi-mode fusion neural network |
| CN116452241B (en) * | 2023-04-17 | 2023-10-20 | 广西财经学院 | User loss probability calculation method based on multi-mode fusion neural network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110188202B (en) | Training method and device of semantic relation recognition model and terminal | |
| CN110168535B (en) | Information processing method and terminal, computer storage medium | |
| CN112711953B (en) | Text multi-label classification method and system based on attention mechanism and GCN | |
| CN115601771A (en) | Business order identification method, device, medium and terminal equipment based on multi-mode data | |
| CN112507704B (en) | Multi-intention recognition method, device, equipment and storage medium | |
| CN112070138B (en) | Construction method of multi-label mixed classification model, news classification method and system | |
| CN112231485A (en) | Text recommendation method and device, computer equipment and storage medium | |
| CN113254649A (en) | Sensitive content recognition model training method, text recognition method and related device | |
| CN113704623A (en) | Data recommendation method, device, equipment and storage medium | |
| CN113723426A (en) | Image classification method and device based on deep multi-flow neural network | |
| CN110209772B (en) | Text processing method, device and equipment and readable storage medium | |
| CN116524261A (en) | Image classification method and product based on multi-mode small sample continuous learning | |
| CN117649567B (en) | Data labeling method, device, computer equipment and storage medium | |
| CN117558270B (en) | Voice recognition method and device and keyword detection model training method and device | |
| Juyal et al. | Multilabel image classification using the CNN and DC-CNN model on Pascal VOC 2012 dataset | |
| CN114330476A (en) | Model training method for media content recognition and media content recognition method | |
| CN117131222A (en) | Semi-automatic labeling method and device based on open world large model | |
| CN116070642A (en) | Text emotion analysis method and related device based on expression embedding | |
| CN116186257A (en) | Method and system for classifying short texts based on mixed features | |
| Yang et al. | Synergy of sight and semantics: Visual intention understanding with clip | |
| CN116227496B (en) | Deep learning-based electric public opinion entity relation extraction method and system | |
| CN118093839B (en) | Knowledge operation question-answer dialogue processing method and system based on deep learning | |
| CN116975298B (en) | NLP-based modernized society governance scheduling system and method | |
| CN117573956B (en) | Metadata management method, device, equipment and storage medium | |
| CN117669493A (en) | Intelligent image-text typesetting method and system based on significance detection |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20230113 |