CN112182020A - Financial behavior identification and classification method, device and computer readable storage medium - Google Patents

Financial behavior identification and classification method, device and computer readable storage medium Download PDF

Info

Publication number
CN112182020A
CN112182020A CN202011184465.9A CN202011184465A CN112182020A CN 112182020 A CN112182020 A CN 112182020A CN 202011184465 A CN202011184465 A CN 202011184465A CN 112182020 A CN112182020 A CN 112182020A
Authority
CN
China
Prior art keywords
text
data
vector
preset
financial
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.)
Granted
Application number
CN202011184465.9A
Other languages
Chinese (zh)
Other versions
CN112182020B (en
Inventor
唐积强
吴震
杨菁林
施力
陈梓瑄
吴莉莉
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
National Computer Network and Information Security Management Center
Original Assignee
National Computer Network and Information Security Management Center
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by National Computer Network and Information Security Management Center filed Critical National Computer Network and Information Security Management Center
Publication of CN112182020A publication Critical patent/CN112182020A/en
Application granted granted Critical
Publication of CN112182020B publication Critical patent/CN112182020B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/24—Querying
    • G06F16/242—Query formulation
    • G06F16/243—Natural language query formulation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F40/00—Handling natural language data
    • G06F40/20—Natural language analysis
    • G06F40/279—Recognition of textual entities
    • G06F40/289—Phrasal analysis, e.g. finite state techniques or chunking
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • 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
    • 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
    • G06Q30/06—Buying, selling or leasing transactions
    • G06Q30/0645—Rental transactions; Leasing transactions
    • 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
    • G06Q30/06—Buying, selling or leasing transactions
    • G06Q30/08—Auctions
    • 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
    • G06Q40/00—Finance; Insurance; Tax strategies; Processing of corporate or income taxes
    • G06Q40/03—Credit; Loans; Processing thereof
    • 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
    • G06Q40/00—Finance; Insurance; Tax strategies; Processing of corporate or income taxes
    • G06Q40/04—Trading; Exchange, e.g. stocks, commodities, derivatives or currency exchange
    • 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
    • G06Q40/00—Finance; Insurance; Tax strategies; Processing of corporate or income taxes
    • G06Q40/06—Asset management; Financial planning or analysis

Landscapes

  • Engineering & Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Theoretical Computer Science (AREA)
  • Accounting & Taxation (AREA)
  • Physics & Mathematics (AREA)
  • Finance (AREA)
  • General Physics & Mathematics (AREA)
  • Development Economics (AREA)
  • Economics (AREA)
  • Marketing (AREA)
  • Strategic Management (AREA)
  • General Business, Economics & Management (AREA)
  • General Engineering & Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Mathematical Physics (AREA)
  • Technology Law (AREA)
  • Computational Linguistics (AREA)
  • General Health & Medical Sciences (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Data Mining & Analysis (AREA)
  • Health & Medical Sciences (AREA)
  • Computing Systems (AREA)
  • Molecular Biology (AREA)
  • Evolutionary Computation (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Software Systems (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Game Theory and Decision Science (AREA)
  • Human Resources & Organizations (AREA)
  • Operations Research (AREA)
  • Databases & Information Systems (AREA)
  • Audiology, Speech & Language Pathology (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

The application relates to a financial behavior identification and classification method, a financial behavior identification and classification device and a computer readable storage medium. The method comprises the following steps: the method comprises the steps of obtaining multi-source data texts from at least two data sources, preprocessing the data texts according to a preset data processing mode to obtain vectors of the data texts, inputting the vectors of the data texts into a pre-trained multi-scale convolution neural network model, determining convolution kernels of the vectors of the data texts according to the types of the data sources, extracting semantic features of the vectors by utilizing the convolution kernels, determining the probability of each preset financial behavior according to the semantic features of the vectors, and finally determining financial behaviors corresponding to the multi-source data texts according to the probability of each preset financial behavior. The multiple data sources can reflect the actually developed business of each aspect of the target financial institution, so that the actual financial behavior of the target financial institution can be more accurately identified, and the supervision is more convenient.

Description

Financial behavior identification and classification method, device and computer readable storage medium
Technical Field
The present application relates to the field of financial data processing technologies, and in particular, to a method and an apparatus for identifying and classifying financial behaviors, and a computer-readable storage medium.
Background
With the vigorous development of the financial industry, various local financial institutions are also diversified, and can be roughly divided into ten types which are responsible for monitoring financial institutions such as small loan companies, financing guarantee companies, regional equity markets, pawn lines, financing lease companies, commercial insurance companies, local asset management companies and the like, and strengthening investment companies, farmer professional cooperative agencies, social crowd funding institutions and local various trading places. However, at present, the local financial institutions are classified based on the business data, and the business operation range contained in the business data may be inconsistent with the actual financial behavior, which causes a problem that the local financial supervision authority is difficult to find out the overall situation and classification situation of the local financial institutions, and further causes the local financial supervision authority to be difficult to supervise the local financial supervision authority. Therefore, how to finely classify the financial behaviors of the local financial institutions becomes the key point of supervision.
Disclosure of Invention
To overcome, at least in part, the problems in the related art, the present application provides a method, apparatus, and computer-readable storage medium for financial behavior identification and classification.
According to a first aspect of the present application, there is provided a method of financial activity identification and classification, the method comprising:
acquiring a multi-source data text of a target financial institution, wherein the multi-source data text comprises data texts from at least two types of data sources respectively;
determining the probability that the target financial institution has each preset financial behavior according to each type of the data text by utilizing a pre-trained multi-scale convolutional neural network model;
and identifying and classifying the financial behaviors of the target financial institution according to the probability of each preset financial behavior so as to determine the type of the target financial institution.
Optionally, the determining, by using a pre-trained multi-scale convolutional neural network model, the probability of each preset financial behavior according to each data text includes:
preprocessing each data text according to a preset data preprocessing mode to obtain a text vector of each data text;
inputting each text vector into a pre-trained multi-scale convolution neural network model, and determining the convolution kernel size of each text vector;
and correspondingly extracting semantic features of each text vector according to the convolution kernel size of each text vector, and determining the probability of each preset financial behavior according to each semantic feature.
Optionally, the determining the convolution kernel size of each text vector includes:
and determining the corresponding convolution kernel size of each text vector according to a preset convolution kernel size mapping table and the data source type corresponding to the text vector, wherein the convolution kernel size mapping table comprises the mapping relation between each data source type and the convolution kernel size.
Optionally, the preprocessing each data text according to a preset data preprocessing manner to obtain a text vector of each data text includes:
performing word segmentation on the data text by using a preset word segmentation method to obtain at least one word of the data text;
determining an index of each of the words in a preset dictionary;
and determining a text vector of the data text according to the index.
Optionally, the determining a text vector of each data text according to the index includes:
sequencing the indexes of the words according to the sequence of each word in the data text to obtain an initial vector of each data text;
and representing each initial vector as a text vector with a preset length according to a preset vector length representation rule.
Optionally, the extracting semantic features of each text vector according to the convolution kernel size of each text vector correspondingly includes:
mapping words corresponding to each index in each text vector into word vectors;
arranging the word vectors according to a preset array arrangement mode to obtain a vector matrix with a preset size;
and correspondingly extracting semantic features of the vector matrix according to the convolution kernel size of each text vector.
Optionally, the determining the probability of each preset financial behavior according to the semantic features of each vector includes:
performing maximum pooling operation on the semantic features of each vector to obtain target semantic features of each vector;
and splicing the target semantic features of the vectors, and inputting the spliced target semantic features into a preset function to obtain the probability of each preset financial behavior.
According to a second aspect of the present application, there is provided an apparatus for financial behavior identification and classification, comprising:
the system comprises an acquisition module, a display module and a display module, wherein the acquisition module is used for acquiring a multi-source data text of a target financial institution, and the multi-source data text comprises data texts from at least two types of data sources respectively;
the first determining module is used for determining the probability of each preset financial behavior according to each data text by utilizing a pre-trained multi-scale convolutional neural network model;
and the second determining module is used for determining the financial behaviors of the target financial institution according to the probability of each preset financial behavior so as to determine the type of the target financial institution.
Optionally, the first determining module includes:
the preprocessing unit is used for preprocessing each data text according to a preset data preprocessing mode to obtain a text vector of each data text;
the determining unit is used for inputting the text vectors into a pre-trained multi-scale convolutional neural network model, determining the convolutional kernel size of each text vector, correspondingly extracting the semantic features of each text vector according to the convolutional kernel size of each text vector, and determining the probability of each preset financial behavior according to the semantic features.
Optionally, the determining unit includes:
the first determining subunit is configured to determine, according to a preset convolution kernel size mapping table and a data source type corresponding to the text vector, a convolution kernel size corresponding to each text vector, where the convolution kernel size mapping table includes a mapping relationship between each data source type and a convolution kernel size.
Optionally, the preprocessing unit includes:
the word segmentation subunit is used for performing word segmentation on the data text by using a preset word segmentation method to obtain at least one word of the data text;
a second determining subunit, configured to determine an index of each of the words in a preset dictionary;
and the third determining subunit is used for determining the text vector of the data text according to the index.
Optionally, the third determining subunit includes:
the sorting subunit is configured to sort the indexes of the words according to the order of each word in the data text, so as to obtain an initial vector of each data text;
and the vector representing subunit is used for representing each initial vector as a vector with a preset length according to a preset vector length representing rule.
Optionally, the determining unit includes:
the mapping subunit is configured to map words corresponding to each index in each text vector into word vectors, so as to obtain a vector matrix of a preset size;
and the extraction subunit is used for correspondingly extracting the semantic features of the vector matrix according to the convolution kernel size of each text vector.
Optionally, the determining unit includes:
the pooling subunit is used for performing maximum pooling operation on the semantic features of each vector to obtain target semantic features of each vector;
and the probability calculation subunit is used for splicing the target semantic features of the vectors and inputting the spliced target semantic features into a preset function to obtain the probability of each preset financial behavior.
According to a third aspect of the present application, there is provided a computer readable storage medium storing one or more programs which, when executed, implement the method of financial activity identification and classification of the first aspect of the present application.
The technical scheme provided by the application can comprise the following beneficial effects: according to the scheme, firstly, data texts of a target financial institution, namely multi-source data texts, are obtained from at least two types of data sources, then, the probability of each preset financial behavior is determined according to each data text by utilizing a pre-trained multi-scale convolution neural network model, and finally, the financial behavior of the target financial institution is determined according to the probability of each preset financial behavior so as to determine the type of the target financial institution. Based on the method, the probability of each preset financial behavior can be determined according to the data texts of different data sources in a targeted manner by using the multi-source data texts and the multi-scale convolutional neural network model, so that the probability of the identification and classification basis of the method is more accurate, and in addition, the multiple data sources can embody the actually developed business of each aspect of the target financial institution, so that the actual financial behavior of the target financial institution can be identified more accurately, and the method is more convenient to supervise.
It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the application.
Drawings
The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and together with the description, serve to explain the principles of the application.
FIG. 1 is a schematic flow chart diagram of a method for financial activity identification and classification according to an embodiment of the present application;
FIG. 2 is a schematic flow chart illustrating the determination of the probability of each predetermined financial behavior using a multi-scale convolutional neural network model according to the present application;
FIG. 3 is a schematic flow chart illustrating the preprocessing of data text according to the present application;
FIG. 4 is a schematic flow chart of semantic feature extraction of vectors using convolution kernels in the present application;
FIG. 5 is a schematic flow chart illustrating the process of determining the probability of each predetermined financial behavior according to semantic features;
FIG. 6 is a schematic structural diagram of a multi-scale convolutional neural network model in the present application;
FIG. 7 is a schematic diagram of a training process of a multi-scale convolutional neural network model in the present application;
fig. 8 is a schematic structural diagram of an apparatus for financial behavior identification and classification according to another embodiment of the present application.
Detailed Description
Reference will now be made in detail to the exemplary embodiments, examples of which are illustrated in the accompanying drawings. When the following description refers to the accompanying drawings, like numbers in different drawings represent the same or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present application. Rather, they are merely examples of apparatus and methods consistent with certain aspects of the present application, as detailed in the appended claims.
With the vigorous development of the financial industry, various local financial institutions are also diversified, and can be roughly divided into ten types which are responsible for monitoring financial institutions such as small loan companies, financing guarantee companies, regional equity markets, pawn lines, financing lease companies, commercial insurance companies, local asset management companies and the like, and strengthening investment companies, farmer professional cooperative agencies, social crowd funding institutions and local various trading places. However, the local financial institutions have the problems of fast change, inconsistent industrial and commercial operation range and actual service development, online and offline mixing and the like, so that the overall situation and the classification situation of the local financial institutions are difficult to find out, and the local financial supervision and management bureau is difficult to supervise the local financial institutions. Therefore, how to finely classify the local financial institutions becomes the key point of supervision.
The refined classification of the local financial institutions aims to identify the classes to which the institutions belong, facilitates the supervision of the institutions, and is a multi-classification problem in nature. Different from the traditional classification based on single industrial and commercial data, the method is based on multi-source heterogeneous data modeling, integrates the multi-source data such as industrial and commercial, introduction, products, patents, websites, public opinions, advertisements and the like from the financial behavior and the business essence of the organization, avoids the problem that the operation range of the industrial and commercial is inconsistent with the actual business development, and adopts a multi-scale convolution neural network to establish a refined classification model of the local financial organization in consideration of different semantic granularities of the data of each channel, thereby improving the discovery efficiency of the local financial organization.
Aiming at the traditional local financial institution refined classification method based on single industrial and commercial data, the scheme provides the local financial institution refined classification method based on multi-source heterogeneous data, and avoids the problem that the industrial and commercial operation range is inconsistent with the actual business development; the method adopts the multi-scale convolution neural network to extract the semantic features of different granularities of the data of each channel, thereby improving the discovery efficiency of local financial institutions and better serving the financial supervision work. The following describes a method, an apparatus and a computer-readable storage medium for identifying and classifying financial behaviors, which are provided by the present application, by way of example.
Referring to fig. 1, fig. 1 is a flowchart illustrating a method for identifying and classifying financial behaviors according to an embodiment of the present application.
Step S101, multi-source data texts of a target financial institution are obtained, wherein the multi-source data texts comprise data texts from at least two types of data sources respectively.
It should be noted that the data source referred to in this step is a source of a data text, such as a business, an introduction, a product, a patent, a website, public opinion, an advertisement, and the like, and the data text has different sources, so the data structure is also different, that is, the multi-source data text obtained in this step is a multi-source and heterogeneous data text.
And S102, determining the probability that the target financial institution has each preset financial behavior according to each type of data text by using a pre-trained multi-scale convolutional neural network model.
It should be noted that, because the text types corresponding to various data sources are different, the semantic granularities of the data sources are also different, for example, public sentiment texts whose expression is more likely to be the semantics of sentiment classes, so that the expression capability of the neural network model is enhanced only by semantic features with finer granularity.
In addition, the preset financial behaviors refer to financial behaviors corresponding to different types of financial institutions, such as financial behaviors of a petty loan company, financial behaviors of a financing guarantee company, financial behaviors of a regional equity market, financial behaviors of a pawn, financial behaviors of a financing lease company, financial behaviors of a business insurance company, financial behaviors of a local asset management company, financial behaviors of an investment company, financial behaviors of a farmer professional cooperative, financial behaviors of a social crowd funding institution, financial behaviors of local various exchanges, and the like. The model in this step determines probability values corresponding to the financial behaviors according to the data text. Specifically, referring to fig. 2, a process of determining the probability of each preset financial behavior by using the multi-scale convolutional neural network model, where fig. 2 is a schematic flow chart of determining the probability of each preset financial behavior by using the multi-scale convolutional neural network model according to the present application.
As shown in fig. 2, the process of determining the probability of each preset financial behavior may include:
step S201, preprocessing each data text according to a preset data preprocessing method to obtain a text vector of each data text.
Since the multi-source data text obtained in step S101 is heterogeneous, and the data text may contain some unrecognizable contents, such as a special symbol or url, or may also have a complex text, in this step, the multi-source data text may be normalized first, the unrecognizable contents are removed, and the complex text is converted into a simple text, and the like.
In addition, in order to more easily process the data texts, each data text is further processed into a vector in this step, that is, each data text is preprocessed according to a preset data preprocessing manner, so as to obtain a text vector of each data text, specifically, refer to fig. 3, where fig. 3 is a schematic flow diagram illustrating preprocessing of the data text in this application.
As shown in fig. 3, taking one of the data texts as an example for description, the process of preprocessing the data text may include:
step S301, performing word segmentation on each data text by using a preset word segmentation method to obtain at least one word of each data text.
In this step, the preset word segmentation method may be, but is not limited to, a jieba word segmentation method, which is also called a jieba word segmentation method, and can support three word segmentation modes: the accurate mode (trying to cut the sentence most accurately and being suitable for text analysis), the full mode (quickly scanning all words that can be word in the sentence) and the search engine mode (on the basis of the accurate mode, cutting long words again and improving recall rate). In this step, a corresponding mode may be adopted according to a requirement, in an example, for example, an accurate mode may be adopted, and for a data text "i comes to the university of beijing qinghua", a jieba word segmentation method is used to obtain "i/comes to/beijing/qinghua university", where "/" is a word segmentation identifier, that is, a jieba word segmentation method is used to segment "i comes to the university of beijing qinghua" into: "i", "come", "Beijing", "Qinghua university" are 4 words.
Step S302, determining the index of each word in the preset dictionary.
Based on the word segmentation result in step S301, this step determines the index of each word obtained by the word segmentation in step S301 according to the mapping relationship between the word and the index in the preset dictionary. It should be noted that the preset dictionary is a preset set including all words that may be related, and each word in the dictionary is provided with a corresponding index, where the index may be a number or a number composed of letters, and in this embodiment, a number composed of numbers is preferred.
In a specific example, still based on the word segmentation result in step S301, 4 words "i", "come", "beijing", "qinghua university" are output in step S301, and in the preset dictionary, the index of "i" is "1", the index of "come" is "3", the index of "beijing" is "2", and the index of "qinghua university" is "5".
Step S303, determining a text vector of each data text according to the index.
In this step, after the index of each word is determined in step S302, the index is used to represent each data text to obtain an initial vector of each data text, specifically, the index of each word is used to represent each word, and the indexes of the words are ordered according to the sequence of each word in the data text, so that the data text can be represented in an index manner, that is, the initial vector of the data text.
In one specific example, the initial vector of the data text "i am coming to the university of qinghua in beijing" is (1, 3, 2, 5).
In addition, since the length of each data text may not be the same, and the vector length of the model input must be the same, after the initial vectors are obtained, each initial vector may be represented as a vector of a preset length according to a preset vector length representation rule. In a specific example, the length of each data text may be set to a fixed value, which may be denoted as max _ length, and in this case, a case where the vector length does not reach the fixed value or a case where the vector length is greater than the fixed value may be encountered. If the length of the vector does not reach the fixed value, 0 (or other labels without index meaning) can be added in front of the initial index until the length reaches the fixed value; if the vector length is greater than the fixed value, portions exceeding the fixed value may be truncated.
In a specific example, if max _ length is 5, the length of the vector (1, 3, 2, 5) obtained in step S203 is less than 5, and in this case, it is sufficient to add "0" before "1", to obtain (0, 1, 3, 2, 5); if max _ length is 3, the length of the vector (1, 3, 2, 5) obtained in step S203 is greater than 3, and in this case, a portion exceeding a fixed value may be deleted to obtain (1, 3, 2). Based on the above operations, the data texts can be mapped to vectors with equal lengths.
Step S202, inputting each text vector into a pre-trained multi-scale convolution neural network model, determining the convolution kernel size of each text vector, correspondingly extracting the semantic features of each text vector according to the convolution kernel size of each text vector, and determining the probability of each preset financial behavior according to each semantic feature.
In this step, the pre-trained multi-scale convolutional neural network model may calculate probabilities of the preset financial behaviors according to vectors of the data text, where the preset financial behaviors are multiple financial behaviors preset in the application, such as financial behavior of a small loan company, financial behavior of a financing guarantee company, financial behavior of a regional equity market, financial behavior of a classic, financial behavior of a financing lease company, financial behavior of a business guarantee company, financial behavior of a local asset management company, financial behavior of an investment company, financial behavior of a farmer professional cooperation society, financial behavior of a social funding institution, and financial behavior of various local exchanges, and in order to express a place to be more convenient, the financial behaviors of the different types of institutions may be mapped with numbers, for example, the financial behavior of a small loan company is "1" The financial behavior of the financing guarantee company is "2", the financial behavior of the regional equity market is "3", and so on.
In this step, because the semantic granularities of different data texts are different, for example, the expression of a public opinion text is more likely to be the semantic of an emotion class, so that semantic features with finer granularity are needed to enhance the expression capability of the model. Texts such as industry, brief introduction, products and the like usually contain semantic representations with coarser granularity, but some special words are separated through jieba participle, which weakens the expression capability of semantic features, so that the feature extraction is carried out on convolution kernel vectors with different sizes according to different data sources. Specifically, after receiving the vector of each data text, the multi-scale convolution neural network model firstly determines the convolution kernel size of each text vector according to the type of the data source of each data text, and then correspondingly extracts the semantic feature of each text vector according to the convolution kernel size of each text vector.
It should be noted that the size of the convolution kernel refers to the size of the convolution kernel, and in a specific example, the size of the convolution kernel corresponding to a text vector may be determined by using a preset convolution kernel size mapping table, and specifically, the size of the convolution kernel corresponding to each text vector may be determined according to the preset convolution kernel size mapping table and a data source type corresponding to the text vector, where the convolution kernel size mapping table includes a mapping relationship between each data source type and the convolution kernel size.
In addition, as for the process of extracting the semantic features of each text vector according to the convolution kernel size of each text vector, refer to fig. 4, where fig. 4 is a schematic flow chart of extracting the semantic features of the vectors by using convolution kernels in the present application.
As shown in fig. 4, the process of extracting semantic features of a vector using a convolution kernel may include:
step S401, mapping the words corresponding to the indexes in each text vector into word vectors.
In this embodiment, after the data text and the words therein are obtained, the data text and the words may be subjected to vocabulary model training through a skip-gram algorithm in a Word2Vec tool to generate corresponding Word vectors. That is, suppose the vocabulary of the data text is X, X is an ordered sequence, X ═ i, come, beijing, qinghua university]There are 4 words, i.e. the word X with the first position1The word vector of "I" is [1, 0, 0, 0]Where the word vector is 4 in length, the second position vocabulary X2The word vector for "come to" is [0, 1, 0]And by analogy, 4 word vectors are calculated, and each word vector is independent of the rest word vectors. Then, setting the length of each word vector to be M through a skip-gram algorithm, and finding a matrix M with shape being (4, M) so as to input a word vector X each timeiThen find the corresponding (X)i-k，……，Xi-2，Xi-1，Xi+1，Xi+2，……，Xi+k) AppearThe probability is the greatest, for example, the word "come to" appears before and after [ I, Beijing, Qinghua university]Should be maximized according to all input XiAnd constructing a joint probability to maximize the joint probability, so that the matrix M is a corresponding word vector matrix solved by the data text and has the characteristics of the website word vectors. Applying the matrix M, assuming that a word vector corresponding to "I" is desired to be found, the one-hot vector corresponding to "I" is [1, 0, 0, 0]May be [1, 0, 0, 0 ]]And expressing the word vectors with the vector multiplier of M to finally obtain word vectors corresponding to the word M, and analogizing other word vectors. The Word2Vec tool is a tool for Word vector calculation. Finally, the words are converted from a high-dimensional sparse vector to a low-dimensional dense vector, so that the words with similar characteristics are closer to each other in space.
And S402, arranging the word vectors according to a preset array arrangement mode to obtain a vector matrix with a preset size.
Since each number in the vector represents a word, in this step, each word in the vector may be mapped into a word vector, specifically, the size of each word vector is the same, for example, the size may be an Embedding _ size, and since the length of the vector is the same, that is, max _ length, at this time, the vector of each data text may be further represented as a vector matrix with a shape of (max _ length, Embedding _ size), that is, a vector matrix with a preset size. In a multi-scale convolutional network, the width of the convolution kernel is consistent with the dimension of the word vector. This is because each line vector of the input represents a word, which is the smallest granularity of the text in the process of extracting the features. And the height can be set by itself as CNN, and the height is similar to n-gram.
And S403, correspondingly extracting semantic features of the vector matrix according to the convolution kernel size of each text vector.
Because the data in this embodiment is based on the data text, which may be a segment of speech, a sentence, etc., and the relevance between adjacent words in the data text is high, in order to consider the word sense, word order, and context at the same time, the embodiment adopts a convolution kernel mode to perform feature extraction. That is, after the vector matrix is reached, semantic features are extracted by using a convolution kernel corresponding to the vector.
In the above steps, since the sizes of the convolution kernels are different, the vector dimensions of the output semantic features are also different, so that in the process of determining the probability of each preset financial behavior, a pooling operation is performed, specifically, refer to fig. 5, where fig. 5 is a schematic flow chart of determining the probability of each preset financial behavior according to the semantic features.
As shown in fig. 5, the process of determining the probability of each preset financial behavior according to the semantic features may include:
step S501, performing maximum pooling operation on the semantic features of each vector to obtain target semantic features of each vector.
The maximum pooling operation is generally performed at a pooling layer, and after the semantic features are obtained, feature vectors in the semantic features are pooled into a value, that is, a value capable of most representing the semantic features, and in a specific example, 1-Max-pooling may be used, that is, a maximum value of a vector in each semantic feature is extracted to represent the semantic features.
And S502, splicing the target semantic features of the vectors, and inputting the spliced target semantic features into a preset function to obtain the probability of each preset financial behavior.
Since the vectors of a plurality of data texts correspondingly obtain the same number of target semantic features, after the maximum pooling operation, the target semantic features need to be spliced together, and in addition, in order to prevent the phenomenon of overfitting of the splicing type, dropout can be added before the pooling layer to the full connection layer to prevent overfitting. The fully-connected layer is a layer for determining the probability of each preset financial behavior according to the spliced target semantic features, and the layer is provided with a preset function for calculating the probability.
For a more concrete description of the process of determining the probability of each preset financial behavior, refer to fig. 6, where fig. 6 is a schematic structural diagram of the multi-scale convolutional neural network model in the present application.
As shown in fig. 6, after the vector is input into the multi-scale convolutional neural network model, the vector is represented as a vector matrix, then convolution operations are performed through convolutional kernels of different sizes of the convolutional layer, that is, semantic feature extraction is performed, then a target semantic feature is obtained through maximum pooling operations and splicing of the pooling layer, and finally the probability of each preset financial behavior is calculated through the dropout reaching the full connection layer, that is, the softmax activation function.
Step S103, identifying and classifying the financial behaviors of the target financial institution according to the probability of each preset financial behavior so as to determine the type of the target financial institution.
In step S102, probability values of the preset financial behaviors are obtained, and the preset financial behavior with the highest probability value may be determined as the financial behavior of the target financial institution, so as to determine the type of the target financial institution.
In a specific example, the probability of financial behavior of a petty loan company is 0.9, the probability of financial behavior of a financing guarantee company is 0.3, the probability of financial behavior of a regional equity market is 0.2, the probability of financial behavior of an accrual company is 0.3, the probability of financial behavior of a financing lease company is 01, the probability of financial behavior of a business guarantee company is 0.3, the probability of financial behavior of a local asset management company is 0.2, the probability of financial behavior of an investment company is 0.1, the probability of financial behavior of a farmer professional cooperation is 0.1, the probability of financial behavior of a social funding institution is 0.2, and the probability of financial behavior of a local exchange is 0.1. At this time, the financial behavior with the highest probability value is the financial behavior of the small loan company, and thus the financial behavior of the target financial institution is the financial behavior of the small loan company, and accordingly, the type of the target financial institution should be the small loan company.
In this embodiment, first, a data text of a target financial institution, that is, a multi-source data text, is obtained from at least two types of data sources, then, a pre-trained multi-scale convolutional neural network model is used to determine probabilities of preset financial behaviors according to the data texts, and finally, financial behaviors of the target financial institution are determined according to the probabilities of the preset financial behaviors to determine the type of the target financial institution. Based on the method, the probability of each preset financial behavior can be determined according to the data texts of different data sources in a targeted manner by using the multi-source data texts and the multi-scale convolutional neural network model, so that the probability of the identification and classification basis of the method is more accurate, and in addition, the multiple data sources can embody the actually developed business of each aspect of the target financial institution, so that the actual financial behavior of the target financial institution can be identified more accurately, and the method is more convenient to supervise.
In addition, when the multi-scale convolutional neural network model is trained, the training sample can be divided into a training set, a verification set and a test set according to the ratio of 7:2:1, the multi-scale convolutional neural network model is trained through the training set, verification is carried out on the verification set, parameters of the model are adjusted, and the generalization capability of the model is tested on the test set.
Referring to fig. 7, fig. 7 is a schematic diagram of a training process of the multi-scale convolutional neural network model in the present application.
As shown in fig. 7, the training process of the multi-scale convolutional neural network model may include preprocessing the training samples, and the preprocessing process may refer to the process shown in fig. 2, which is not described herein again.
After preprocessing, dividing the preprocessed training samples into a training set, a verification set and a test set, verifying the effectiveness of the model by using five-fold cross verification during training, and performing cross verification by using layered sampling because the data has the problem of unbalance of positive and negative samples.
Specifically, the training sample may be T { (x)1,y1),(x2,y2),...,(xN,yN) In which xkRepresenting the corresponding multi-source data text set of the local financial institutions such as the industry, the introduction, the products, the patents, the websites, the public sentiments, the advertisements and the like, ykN denotes the category of the local financial institution, i.e. the tag value (the tag value may be indicated by a number, e.g. 1,2,3, etc., each referring to a different fundConvergence behavior), k is 1,2,3, … …, N, T is input data of the model, i.e. text data of different channels and sample y to be predictedk。
It should be noted that the divided verification set can be used to adjust parameters of the model, and the test set is used to verify the generalization ability of the model, specifically, the trained multi-scale convolutional neural network model is used to the test set, the F1 score on the test set is calculated, and the score is used to verify the generalization ability of the model.
The F1 score (also called F1 score) represents the harmonic mean of the accuracy and recall of the model, and for the specific calculation process and verification process, reference may be made to related technologies, which are not described herein again.
Referring to fig. 8, fig. 8 is a schematic structural diagram of an apparatus for financial behavior identification and classification according to another embodiment of the present application.
As shown in fig. 8, the apparatus provided in this embodiment may include:
an obtaining module 801, configured to obtain a multi-source data text of a target financial institution, where the multi-source data text includes data texts from at least two types of data sources respectively;
a first determining module 802, configured to determine, according to each data text, a probability of each preset financial behavior by using a pre-trained multi-scale convolutional neural network model;
a second determining module 803, configured to determine the financial behaviors that the target financial institution has according to the probabilities of the preset financial behaviors, so as to determine the type of the target financial institution.
Optionally, the first determining module includes:
the preprocessing unit is used for preprocessing each data text according to a preset data preprocessing mode to obtain a text vector of each data text;
the determining unit is used for inputting the text vectors into a pre-trained multi-scale convolutional neural network model, determining the convolutional kernel size of each text vector, correspondingly extracting the semantic features of each text vector according to the convolutional kernel size of each text vector, and determining the probability of each preset financial behavior according to the semantic features.
Optionally, the determining unit includes:
the first determining subunit is configured to determine, according to a preset convolution kernel size mapping table and a data source type corresponding to the text vector, a convolution kernel size corresponding to each text vector, where the convolution kernel size mapping table includes a mapping relationship between each data source type and a convolution kernel size.
Optionally, the preprocessing unit includes:
the word segmentation subunit is used for performing word segmentation on the data text by using a preset word segmentation method to obtain at least one word of the data text;
a second determining subunit, configured to determine an index of each of the words in a preset dictionary;
and the third determining subunit is used for determining the text vector of the data text according to the index.
Optionally, the third determining subunit includes:
the sorting subunit is configured to sort the indexes of the words according to the order of each word in the data text, so as to obtain an initial vector of each data text;
and the vector representing subunit is used for representing each initial vector as a vector with a preset length according to a preset vector length representing rule.
Optionally, the determining unit includes:
the mapping subunit is configured to map words corresponding to each index in each text vector into word vectors, so as to obtain a vector matrix of a preset size;
and the extraction subunit is used for correspondingly extracting the semantic features of the vector matrix according to the convolution kernel size of each text vector.
Optionally, the determining unit includes:
the pooling subunit is used for performing maximum pooling operation on the semantic features of each vector to obtain target semantic features of each vector;
and the probability calculation subunit is used for splicing the target semantic features of the vectors and inputting the spliced target semantic features into a preset function to obtain the probability of each preset financial behavior.
In addition, another embodiment of the present application further provides a computer-readable storage medium, in which one or more programs are stored, and when the one or more programs are executed, the method for identifying and classifying financial behaviors provided by the foregoing embodiment of the present application is implemented.
With regard to the apparatus in the above-described embodiment, the specific manner in which each module performs the operation has been described in detail in the embodiment related to the method, and will not be elaborated here.
It is understood that the same or similar parts in the above embodiments may be mutually referred to, and the same or similar parts in other embodiments may be referred to for the content which is not described in detail in some embodiments.
It should be noted that, in the description of the present application, the terms "first", "second", etc. are used for descriptive purposes only and are not to be construed as indicating or implying relative importance. Further, in the description of the present application, the meaning of "a plurality" means at least two unless otherwise specified.
Any process or method descriptions in flow charts or otherwise described herein may be understood as representing modules, segments, or portions of code which include one or more executable instructions for implementing specific logical functions or steps of the process, and the scope of the preferred embodiments of the present application includes other implementations in which functions may be executed out of order from that shown or discussed, including substantially concurrently or in reverse order, depending on the functionality involved, as would be understood by those reasonably skilled in the art of the present application.
It should be understood that portions of the present application may be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, the various steps or methods may be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, any one or combination of the following techniques, which are known in the art, may be used: a discrete logic circuit having a logic gate circuit for implementing a logic function on a data signal, an application specific integrated circuit having an appropriate combinational logic gate circuit, a Programmable Gate Array (PGA), a Field Programmable Gate Array (FPGA), or the like.
It will be understood by those skilled in the art that all or part of the steps carried by the method for implementing the above embodiments may be implemented by hardware that is related to instructions of a program, and the program may be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.
In addition, functional units in the embodiments of the present application may be integrated into one processing module, or each unit may exist alone physically, or two or more units are integrated into one module. The integrated module can be realized in a hardware mode, and can also be realized in a software functional module mode. The integrated module, if implemented in the form of a software functional module and sold or used as a separate product, may also be stored in a computer readable storage medium.
The storage medium mentioned above may be a read-only memory, a magnetic or optical disk, etc.
In the description herein, reference to the description of the term "one embodiment," "some embodiments," "an example," "a specific example," or "some examples," etc., means that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the application. In this specification, the schematic representations of the terms used above do not necessarily refer to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.
Although embodiments of the present application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present application, and that variations, modifications, substitutions and alterations may be made to the above embodiments by those of ordinary skill in the art within the scope of the present application.

Claims (10)

1. A method for financial activity identification and classification, the method comprising:
acquiring a multi-source data text of a target financial institution, wherein the multi-source data text comprises data texts from at least two types of data sources respectively;
determining the probability that the target financial institution has each preset financial behavior according to each type of the data text by utilizing a pre-trained multi-scale convolutional neural network model;
and identifying and classifying the financial behaviors of the target financial institution according to the probability of each preset financial behavior so as to determine the type of the target financial institution.
2. The method of claim 1, wherein determining the probability of each pre-determined financial activity from each data text using a pre-trained multi-scale convolutional neural network model comprises:
preprocessing each data text according to a preset data preprocessing mode to obtain a text vector of each data text;
inputting each text vector into a pre-trained multi-scale convolution neural network model, and determining the convolution kernel size of each text vector;
and correspondingly extracting semantic features of each text vector according to the convolution kernel size of each text vector, and determining the probability of each preset financial behavior according to each semantic feature.
3. The method of claim 2, wherein said determining a convolution kernel size for each of said text vectors comprises:
and determining the corresponding convolution kernel size of each text vector according to a preset convolution kernel size mapping table and the data source type corresponding to the text vector, wherein the convolution kernel size mapping table comprises the mapping relation between each data source type and the convolution kernel size.
4. The method of claim 2, wherein the preprocessing each of the data texts according to a preset data preprocessing manner to obtain a text vector of each of the data texts comprises:
performing word segmentation on the data text by using a preset word segmentation method to obtain at least one word of the data text;
determining the index of each word in a preset dictionary;
and determining a text vector of the data text according to the index.
5. The method of claim 4, wherein determining a text vector for each of the data texts based on the index comprises:
sequencing the indexes of the words according to the sequence of each word in the data text to obtain an initial vector of each data text;
and representing each initial vector as a text vector with a preset length according to a preset vector length representation rule.
6. The method according to claim 4, wherein said extracting semantic features of each of said text vectors according to said convolution kernel size correspondence of each of said text vectors comprises:
mapping words corresponding to each index in each text vector into word vectors;
arranging the word vectors according to a preset array arrangement mode to obtain a vector matrix with a preset size;
and correspondingly extracting semantic features of the vector matrix according to the convolution kernel size of each text vector.
7. The method of claim 2, wherein determining the probability of each predetermined financial action based on each semantic feature comprises:
performing maximum pooling operation on the semantic features of each vector to obtain target semantic features of each vector;
and splicing the target semantic features of the vectors, and inputting the spliced target semantic features into a preset function to obtain the probability of each preset financial behavior.
8. An apparatus for financial activity identification and classification, comprising:
the system comprises an acquisition module, a display module and a display module, wherein the acquisition module is used for acquiring a multi-source data text of a target financial institution, and the multi-source data text comprises data texts from at least two types of data sources respectively;
the first determining module is used for determining the probability of each preset financial behavior according to each data text by utilizing a pre-trained multi-scale convolutional neural network model;
and the second determining module is used for determining the financial behaviors of the target financial institution according to the probability of each preset financial behavior so as to determine the type of the target financial institution.
9. The apparatus of claim 8, wherein the first determining module comprises:
the preprocessing unit is used for preprocessing each data text according to a preset data preprocessing mode to obtain a text vector of each data text;
the determining unit is used for inputting the text vectors into a pre-trained multi-scale convolutional neural network model, determining the convolutional kernel size of each text vector, correspondingly extracting the semantic features of each text vector according to the convolutional kernel size of each text vector, and determining the probability of each preset financial behavior according to the semantic features.
10. A computer-readable storage medium, wherein the computer storage medium stores one or more programs which, when executed, implement the method of financial activity identification and classification of any of claims 1-7.
CN202011184465.9A 2020-09-25 2020-10-29 Financial behavior identification and classification method, device and computer readable storage medium Active CN112182020B (en)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202011028994 2020-09-25
CN202011028994X 2020-09-25

Publications (2)

Publication Number Publication Date
CN112182020A true CN112182020A (en) 2021-01-05
CN112182020B CN112182020B (en) 2023-04-18

Family

ID=73916144

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202011184465.9A Active CN112182020B (en) 2020-09-25 2020-10-29 Financial behavior identification and classification method, device and computer readable storage medium

Country Status (1)

Country Link
CN (1) CN112182020B (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112926631A (en) * 2021-02-01 2021-06-08 大箴(杭州)科技有限公司 Financial text classification method and device and computer equipment

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20140249977A1 (en) * 2013-03-04 2014-09-04 Rawllin International Inc. Interactive financial institution interface
CN108009911A (en) * 2017-11-29 2018-05-08 上海出版印刷高等专科学校 A kind of method of identification P2P network loan borrower's default risks
CN111460146A (en) * 2020-03-23 2020-07-28 南京邮电大学 Short text classification method and system based on multi-feature fusion
CN111680225A (en) * 2020-04-26 2020-09-18 国家计算机网络与信息安全管理中心 WeChat financial message analysis method and system based on machine learning

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20140249977A1 (en) * 2013-03-04 2014-09-04 Rawllin International Inc. Interactive financial institution interface
CN108009911A (en) * 2017-11-29 2018-05-08 上海出版印刷高等专科学校 A kind of method of identification P2P network loan borrower's default risks
CN111460146A (en) * 2020-03-23 2020-07-28 南京邮电大学 Short text classification method and system based on multi-feature fusion
CN111680225A (en) * 2020-04-26 2020-09-18 国家计算机网络与信息安全管理中心 WeChat financial message analysis method and system based on machine learning

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112926631A (en) * 2021-02-01 2021-06-08 大箴(杭州)科技有限公司 Financial text classification method and device and computer equipment

Also Published As

Publication number Publication date
CN112182020B (en) 2023-04-18

Similar Documents

Publication Publication Date Title
CN109858010B (en) Method and device for recognizing new words in field, computer equipment and storage medium
CN110209805B (en) Text classification method, apparatus, storage medium and computer device
Li et al. News impact on stock price return via sentiment analysis
CN112163424B (en) Data labeling method, device, equipment and medium
CN109829629A (en) Generation method, device, computer equipment and the storage medium of risk analysis reports
CN112613324A (en) Semantic emotion recognition method, device, equipment and storage medium
CN112347254B (en) Method, device, computer equipment and storage medium for classifying news text
Haque et al. Opinion mining from bangla and phonetic bangla reviews using vectorization methods
EP3876137A1 (en) System for identifying named entities with dynamic parameters
KR102280490B1 (en) Training data construction method for automatically generating training data for artificial intelligence model for counseling intention classification
CN110888983A (en) Positive and negative emotion analysis method, terminal device and storage medium
CN112182020B (en) Financial behavior identification and classification method, device and computer readable storage medium
CN116882414B (en) Automatic comment generation method and related device based on large-scale language model
CN117764039A (en) Bid file generation method, system, terminal and storage medium based on large model
KR20210001625A (en) A method and an apparatus for providing services of corporate insolvencies information based on automatic sentiment information labelings to news articles
KR20210001649A (en) A program for predicting corporate default
CN116795978A (en) Complaint information processing method and device, electronic equipment and medium
CN115757799A (en) Data storage method and system based on artificial intelligence and cloud platform
CN113468308B (en) Conversation behavior classification method and device and electronic equipment
CN114741501A (en) Public opinion early warning method and device, readable storage medium and electronic equipment
CN112257416A (en) Inspection new word discovery method and system
KR20210001686A (en) A program for providing a corporate insolvencies information based on automatic sentiment information labelings
KR20210001693A (en) A rcording media for recording program for providing a corporate insolvencies information based on automatic sentiment information labelings
KR20210001666A (en) A method for predicting corporate default
KR20210001645A (en) A method for predicting corporate default

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