Training methods, risk warning methods, devices, equipment, media and program products

Abstract

The present disclosure provides a method, apparatus, device, storage medium and program product for training a risk warning model. The method comprises: obtaining N first training samples from M data sources; obtaining a text vector based on the training text, and obtaining a data source vector based on the information of the data source to which it belongs; obtaining N training vectors corresponding to the N first training samples; inputting each of the training vectors into a risk warning model to obtain a prediction result; and training the risk warning model based on the error between the prediction result and the label information. The embodiment of the present disclosure can combine the data source and the training text to provide the model with a priori information, so that the model can learn the training text characteristics and risk characteristics related to the data source during the training process. The present disclosure also provides a risk warning method, apparatus, device, storage medium and program product.

Description

Technical Field

[0001] The present disclosure relates to the field of artificial intelligence, and more specifically, to a training method, a risk warning method, an apparatus, a device, a medium, and a program product. Background Art

[0002] With the development of the internet, it is increasingly important to identify risky public opinion in the vast amount of public information and to promptly address these risks. For example, we can quickly and accurately identify content related to risky public opinion in financial news texts and determine the event subject to which the risky public opinion relates, thereby achieving the purpose of risk early warning.

[0003] Manually monitoring network information to provide risk warnings is no longer sufficient. Related technologies utilize machine learning and other methods to develop risk warning models for automated risk monitoring and identification. However, during the training of risk warning models, it was found that the model training efficiency was poor, making it difficult to learn effective information, resulting in poor risk warning effectiveness. Summary of the Invention

[0004] In view of the above problems, the present disclosure provides a training method, a risk warning method, an apparatus, a device, a medium and a program product for improving training efficiency and warning effect.

[0005] One aspect of an embodiment of the present disclosure provides a method for training a risk warning model, comprising: obtaining N first training samples from M data sources, wherein each first training sample includes a training text and information about its data source, and label information of the training text, the label information including a risk entity label and a risk event label, and N and M are integers greater than or equal to 1, respectively; obtaining a text vector based on the training text, and obtaining a data source vector based on information about the data source; merging the text vector and the data source vector corresponding to each first training sample to obtain N training vectors corresponding to the N first training samples; inputting each of the training vectors into a risk warning model to obtain a prediction result, wherein the prediction result includes a predicted risk entity and a predicted risk event; and training the risk warning model based on the error between the prediction result and the label information.

[0006] According to an embodiment of the present disclosure, the information of the belonging data source includes risk event distribution information, and the method also includes obtaining the risk event distribution information, specifically including: obtaining S first training samples from the belonging data source from the N first training samples, where S is an integer greater than or equal to 1 and S is less than or equal to N; determining the risk event label corresponding to each first training sample in the S first training samples; and obtaining the risk event distribution information based on the proportion of each risk event label in the S first training samples.

[0007] According to an embodiment of the present disclosure, obtaining a data source vector based on the information of the belonging data source includes: determining the first dimension of the data source vector based on the number of types of risk event labels in the N first training samples, wherein each risk event label corresponds to a position in the data source vector; and taking the proportion of each risk event label in the S first training samples as the vector value of the corresponding position in the data source vector.

[0008] According to an embodiment of the present disclosure, the information of the belonging data source includes a data source identifier of the belonging data source, the M data sources correspond to M data source identifiers, and obtaining the data source vector based on the information of the belonging data source includes: performing vector encoding on each data source identifier of the M data source identifiers; and obtaining the data source vector based on the encoding result of each data source identifier.

[0009] According to an embodiment of the present disclosure, obtaining a text vector based on the training text includes: inputting the training text into a pre-training model to obtain the text vector.

[0010] According to an embodiment of the present disclosure, the pre-training model includes an encoding layer, which is used to obtain the text vector based on the training text. The method also includes: constructing a vector conversion model based on the encoding layer; inputting the data source vector into the vector conversion model for conversion; wherein, merging the text vector and the data source vector corresponding to each first training sample includes: adding or splicing the converted data source vector and the text vector.

[0011] According to an embodiment of the present disclosure, after obtaining the N first training samples, the method further includes: determining the number of each risk event label in the N first training samples, wherein each first training sample corresponds to a risk event label; determining a first risk event label based on the number of each risk event label, wherein the first risk event label is the risk event label with the largest number; performing data enhancement on the P first training samples corresponding to the second risk event label to obtain Q second training samples, wherein the Q second training samples are used to train the risk warning model, and the second risk event label is any label other than the first risk event label, and P and Q are integers greater than or equal to 1, respectively.

[0012] According to an embodiment of the present disclosure, the data enhancement of the P first training samples corresponding to the second risk event label includes: performing the following operation on the training text of each first training sample in the P first training samples, inserting a first preset number of punctuation marks into the training text according to the number of characters in the training text, wherein the first preset number is less than or equal to the number of characters.

[0013] According to an embodiment of the present disclosure, the method further includes: replacing the risk entity in the first training sample and / or the second training sample.

[0014] According to an embodiment of the present disclosure, training the risk warning model includes training the risk warning model X times, where X is an integer greater than or equal to 2. The method further includes obtaining a new second training sample after each training of the risk warning model until the X training times are completed, specifically including: performing the data augmentation and / or risk entity replacement operations before each training of the risk warning model.

[0015] Another aspect of an embodiment of the present disclosure provides a risk warning method, including: obtaining text to be recognized, and information about a data source to which the text to be recognized belongs; obtaining a first vector to be recognized based on the text to be recognized, and obtaining a second vector to be recognized based on the information about the data source to which it belongs; merging the first vector to be recognized and the second vector to be recognized to obtain a third vector to be recognized; inputting the third vector to be recognized into a risk warning model to obtain a prediction result, wherein the prediction result includes a predicted risk entity and a predicted risk event, and the risk warning model is obtained by training using the method described above.

[0016] Another aspect of an embodiment of the present disclosure provides a training device for a risk warning model, comprising: a first acquisition module, used to acquire N first training samples from M data sources, wherein each of the first training samples includes a training text and information about its data source, and label information of the training text, the label information includes a risk entity label and a risk event label, and N and M are integers greater than or equal to 1 respectively; a first vector module, used to obtain a text vector based on the training text, and to obtain a data source vector based on information about the data source; a first merging module, used to merge the text vector and the data source vector corresponding to each of the first training samples to obtain N training vectors corresponding to the N first training samples; a first input module, used to input each of the training vectors into a risk warning model to obtain a prediction result, wherein the prediction result includes a predicted risk entity and a predicted risk event; a model training module, used to train the risk warning model based on the error between the prediction result and the label information.

[0017] Another aspect of an embodiment of the present disclosure provides a risk warning model device, including: a second acquisition module, used to acquire text to be recognized, and information about the data source to which the text to be recognized belongs; a second vector module, used to obtain a first vector to be recognized based on the text to be recognized, and to obtain a second vector to be recognized based on the information about the data source to which it belongs; a second merging module, used to merge the first vector to be recognized and the second vector to be recognized to obtain a third vector to be recognized; a second input module, used to input the third vector to be recognized into a risk warning model to obtain a prediction result, wherein the prediction result includes a predicted risk entity and a predicted risk event, and the risk warning model is obtained by training the device as described above.

[0018] Another aspect of an embodiment of the present disclosure provides an electronic device, comprising: one or more processors; a storage device for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors execute the method described above.

[0019] Another aspect of the embodiments of the present disclosure further provides a computer-readable storage medium having executable instructions stored thereon, which, when executed by a processor, causes the processor to execute the method described above.

[0020] Another aspect of the embodiments of the present disclosure further provides a computer program product, including a computer program, which implements the above method when executed by a processor.

[0021] One or more of the above embodiments have the following beneficial effects:

[0022] 1) Adding data source information to the training text content to generate training vectors. This allows the data source and training text to be combined to provide the model with a priori information. This allows the model to learn the characteristics of the training text and the risk characteristics associated with the data source during training. This provides additional guidance for parameter updates, allowing the model to quickly learn effective information even with complexly distributed corpora, thereby improving training efficiency.

[0023] 2) The risk warning model obtained by the training method of the embodiment of the present disclosure can take into account the text to be identified and the data source information to which it belongs, and output predicted risk entities and predicted risk events more accurately, thereby improving the risk warning effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] The above contents and other objects, features and advantages of the present disclosure will become more apparent through the following description of the embodiments of the present disclosure with reference to the accompanying drawings, in which:

[0025] Figure 1Schematically illustrates an application scenario diagram of a training method or a risk warning method according to an embodiment of the present disclosure;

[0026] Figure 2 The flowchart of the training method according to the embodiment of the present disclosure is schematically shown;

[0027] Figure 3 Schematically shows a flow chart of a training method according to another embodiment of the present disclosure;

[0028] Figure 4 Schematically shows a flow chart for obtaining risk event distribution information according to an embodiment of the present disclosure;

[0029] Figure 5 Schematically shows a flow chart of obtaining a data source vector according to an embodiment of the present disclosure;

[0030] Figure 6 Schematically shows a flow chart of obtaining a data source vector according to another embodiment of the present disclosure;

[0031] Figure 7 Schematically shows a flowchart of merging a text vector and a data source vector according to an embodiment of the present disclosure;

[0032] Figure 8 The following schematically shows a flow chart of a risk warning method according to an embodiment of the present disclosure;

[0033] Figure 9 The following schematically shows a structural block diagram of a training device according to an embodiment of the present disclosure;

[0034] Figure 10 The following schematically shows a structural block diagram of a risk warning device according to an embodiment of the present disclosure;

[0035] Figure 11 A block diagram of an electronic device suitable for implementing a training method or a risk warning method according to an embodiment of the present disclosure is schematically shown. DETAILED DESCRIPTION

[0036] Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are merely exemplary and are not intended to limit the scope of the present disclosure. In the detailed description below, for ease of explanation, many specific details are set forth to provide a comprehensive understanding of the embodiments of the present disclosure. However, it is apparent that one or more embodiments may also be implemented without these specific details. In addition, in the following description, descriptions of well-known structures and technologies are omitted to avoid unnecessary confusion of the concepts of the present disclosure.

[0037] The terms used herein are only for describing specific embodiments and are not intended to limit the present disclosure. The terms "comprise," "include," etc. used herein indicate the presence of the features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0038] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein should be interpreted as having a meaning consistent with the context of this specification and should not be interpreted in an idealized or overly rigid manner.

[0039] When expressions such as "at least one of A, B and C, etc." are used, they should generally be interpreted in accordance with the meaning of the expression commonly understood by those skilled in the art (for example, "a system having at least one of A, B and C" should include but is not limited to a system having A alone, B alone, C alone, A and B, A and C, B and C, and / or A, B, C, etc.).

[0040] When using natural language processing technology to extract information and train risk warning models, the following problems arise:

[0041] Whether in model training or risk warning, input data comes from web crawling or data access. The data sources themselves potentially contain information about a variety of risk entities that warrant warnings. For example, news websites may cover a wide variety of risk entities, while stock market commentary websites contain even more risk entities related to stock prices. Treating data from all data sources equally could significantly increase training costs.

[0042] Furthermore, different data sources produce text corpora in different formats. When using natural language processing techniques to extract semantic vectors, the resulting semantic vectors may differ depending on the data source. For example, the format of the text corpus from website A is "[Kweichow Moutai] stock price rose 1.4%, [Tiandi Bank] stock price fell 2.9%...", while the format of the text corpus from website B is "Kweichow Moutai's stock price rose 1.4% after today's opening, while Tiandi Bank's stock price fell 2.9%."

[0043] The embodiments of the present disclosure provide a training method that can add data source information of the text to the content of the training text, and use this to perform feature fusion to obtain training vectors. In this way, the combination of the data source and the training text can provide a priori information to the model. The model can learn the data source-related training text characteristics (such as the format characteristics of the text corpus) and risk characteristics (such as the risk entity or event characteristics that each data source focuses on) during the training process, and output more confident risk prediction results. It can provide additional parameter update guidance during the model training process, allowing the model to quickly learn effective information even when faced with complex distribution corpus, thereby improving training efficiency.

[0044] The embodiments of the present disclosure also provide a risk warning method. The risk warning model obtained using the training method of the embodiments of the present disclosure can take into account the text to be identified and the data source information to which it belongs, and output predicted risk entities and predicted risk events more accurately, thereby improving the risk warning effect.

[0045] In the technical solution disclosed herein, the collection, storage, use, processing, transmission, provision, disclosure and application of the relevant data involved have been authorized or agreed to by each data source, comply with the provisions of relevant laws and regulations, take necessary confidentiality measures, and do not violate public order and good morals.

[0046] Figure 1 The application scenario diagram of the training method or risk warning method according to the embodiment of the present disclosure is schematically shown.

[0047] like Figure 1 As shown, the application scenario 100 according to this embodiment may include terminal devices 101, 102, 103, a network 104, and a server 105. The network 104 is used as a medium for providing a communication link between the terminal devices 101, 102, 103 and the server 105. The network 104 may include various connection types, such as wired or wireless communication links or optical fiber cables.

[0048] Users can use terminal devices 101, 102, and 103 to interact with server 105 via network 104 to receive or send messages, etc. Various communication client applications can be installed on terminal devices 101, 102, and 103, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, social platform software, etc. (only as examples).

[0049] The terminal devices 101 , 102 , and 103 may be various electronic devices having a display screen and supporting web browsing, including but not limited to smart phones, tablet computers, laptop computers, and desktop computers.

[0050] The server 105 may be a server that provides various services, such as a background management server (for example only) that supports websites browsed by users using the terminal devices 101, 102, and 103. The background management server may analyze and process received data such as user requests, and feed back processing results (such as web pages, information, or data obtained or generated according to user requests) to the terminal device.

[0051] It should be noted that the training method or risk warning method provided in the embodiment of the present disclosure can generally be executed by the server 105. Accordingly, the training device or risk warning device provided in the embodiment of the present disclosure can generally be set in the server 105. The training method or risk warning method provided in the embodiment of the present disclosure can also be executed by a server or server cluster that is different from the server 105 and can communicate with the terminal devices 101, 102, 103 and / or the server 105. Accordingly, the training device or risk warning device provided in the embodiment of the present disclosure can also be set in a server or server cluster that is different from the server 105 and can communicate with the terminal devices 101, 102, 103 and / or the server 105.

[0052] It should be understood that Figure 1 The number of terminal devices, networks and servers in the embodiment is merely illustrative. Any number of terminal devices, networks and servers may be provided as required.

[0053] The following will be based on Figure 1 The scene described by Figures 2 to 8 The training method and risk warning method of the embodiments of the present disclosure are described in detail.

[0054] Figure 2 The flowchart of the training method according to the embodiment of the present disclosure is schematically shown.

[0055] like Figure 2 As shown, the training method of this embodiment includes operations S210 to S250.

[0056] In operation S210, N first training samples are obtained from M data sources, where each first training sample includes information about a training text and its data source, and label information of the training text, the label information includes a risk entity label and a risk event label, and N and M are integers greater than or equal to 1, respectively.

[0057] Exemplarily, the data source is the source of text content. For example, news media, company official websites, social networking sites or government websites. Text content includes news texts, company financial reports, personal information or policy documents, etc. The training text is at least one sentence or at least one paragraph belonging to the above text content. The data source is the source of each training text. The information of the data source may include data source identification, risk event distribution information, author information, writing style or text format data, etc. The risk entity label includes the name of the entity involved in the risk in the training text, such as the name of a legal person or individual. Risk events include, for example, loss risk events, default risk events, bankruptcy risk events, leadership change risk events, rating deterioration risk events or financial fraud risk events.

[0058] Stylistic style refers to the fact that different texts have different styles. For example, news articles and company financial reports have distinct narrative styles. Converting stylistic style into vectors can further enable the model to learn effective information. Furthermore, different authors may have distinct narrative styles when describing the same type of text. Alternatively, different authors may focus on different types of risk events. For example, Author A may often publish articles focusing on loss-making events.

[0059] The following are examples of training text:

[0060] "However, affected by financial industry policies and the overall environment, Company A and its subsidiaries all suffered losses in 2018. Accounts receivable from commercial poly business, loan assistance business, and auto loan business totaling 1.972 billion yuan were uncollectible, and provisions for bad debts reached 112 million yuan. Furthermore, some loan assistance business was overdue, resulting in overdue liabilities of 397 million yuan."

[0061] For example, the risk entity label of the training text is "Company A", and the risk event label is "Loss Event Subject". The data source is financial media website A.

[0062] In operation S220 , a text vector is obtained according to the training text, and a data source vector is obtained according to information of the corresponding data source.

[0063] According to an embodiment of the present disclosure, obtaining a text vector based on a training text includes: inputting the training text into a pre-training model to obtain a text vector.

[0064] Pre-trained models can include the BERT model, the Roberta-wwm-ext-large model, the ERNIE model, the NEZHA model, or the XLNet model. Taking the BERT model as an example, BERT (Bidirectional Encoder Representation From Transformers) is a pre-trained language model that uses a multi-layer bidirectional transformer encoder to generate deep, bidirectional language representations that incorporate both left and right context information.

[0065] In operation S230 , the text vector and the data source vector corresponding to each first training sample are merged to obtain N training vectors corresponding to the N first training samples.

[0066] Exemplarily, each first training sample first obtains a text vector and a data source vector respectively, and then merges them to obtain a training vector.

[0067] In operation S240 , each training vector is input into the risk warning model to obtain a prediction result, wherein the prediction result includes a predicted risk entity and a predicted risk event.

[0068] For example, the risk warning model can be a neural network model, such as a convolutional neural network, a long short-term memory neural network, or a bidirectional long short-term memory network. The risk warning model can process training vectors and output a classification result via at least one of sigmoid, softmax, conditional random field (CRF), etc.

[0069] In operation S250 , a risk warning model is trained based on the error between the prediction result and the label information.

[0070] For example, positive samples (such as N first training samples) and negative samples can be input during the training of the model. Among them, the negative samples can be training texts without entities. Specifically, no entity means that this text cannot mean that a certain entity has risks within the specified range, such as "Founder Securities: Semiconductor profit growth is accelerating across the board, and domestic substitution has entered the 2.0 era. Founder Securities pointed out that the performance of domestic integrated circuit design companies has accelerated across the board. The turning point of the packaging and testing link has appeared." Among them, Founder Securities is not a risk entity.

[0071] For example, the error between the prediction result and the label information is calculated using a cross-entropy loss function, an exponential loss function, or a mean squared error loss function. The process of training the model is the process of making the loss function converge. For example, if the error is less than or equal to a preset threshold, the training can be considered complete.

[0072] The training method provided by the embodiment of the present disclosure can add the data source information of the text to the content of the training text, and thereby obtain the training vector. Thus, the combination of the data source and the training text can provide the model with a priori information. The model can learn the data source-related training text characteristics (such as the format characteristics of the text corpus and the author's style characteristics) and risk characteristics (such as the risk entity or event characteristics focused on by each data source) during the training process, and output more confident risk prediction results. It can provide additional parameter update guidance during the model training process, allowing the model to quickly learn effective information even when faced with complex distribution corpus, thereby improving training efficiency.

[0073] Figure 3 The flowchart of a training method according to another embodiment of the present disclosure is schematically shown.

[0074] like Figure 3 As shown, the training method of this embodiment includes not only operations S210 to S250, but also operations S310 to S330 after operation S210. Operations S210 to S250 can refer to the above embodiment and are not described in detail here.

[0075] In operation S310 , the number of each risk event label in N first training samples is determined, wherein each first training sample corresponds to a risk event label.

[0076] In operation S320 , a first risk event label is determined based on the number of each risk event label, wherein the first risk event label is a risk event label with the largest number.

[0077] For example, among the 100 first training samples, there are three types of risk event labels, such as 20 loss risk events, 30 default risk events, and 50 rating downgrade risk events (just an example). The first risk event label is a rating downgrade risk event.

[0078] In operation S330, data enhancement is performed on the P first training samples corresponding to the second risk event label to obtain Q second training samples, wherein the Q second training samples are used to train the risk warning model, the second risk event label is any label except the first risk event label, and P and Q are integers greater than or equal to 1 respectively.

[0079] For example, the second risk event label can be either the aforementioned loss risk event or the default risk event. Here, data augmentation is performed on the number of labels for both loss risk events and default risk events. For example, by augmenting the first training samples of 20 loss risk events, 30 second training samples of loss risk events are obtained. By augmenting the first training samples of 30 default risk events, 20 second training samples of default risk events are obtained. Ultimately, 50 training samples of loss risk events, 50 default risk events, and 50 rating downgrade risk events (including the first and second training samples) are obtained.

[0080] It should be noted that, in operation S220 , a text vector and a corresponding data source vector may be obtained according to the training text and the newly added second training text.

[0081] In related technologies, the collected training samples often have uneven risk distribution, which can make it difficult for the model to converge during training. Therefore, by performing data augmentation to obtain new training samples, so that various risk event labels have the same or similar distribution, training efficiency can be improved.

[0082] According to an embodiment of the present disclosure, data enhancement of the P first training samples corresponding to the second risk event label includes: performing the following operations on the training text of each first training sample in the P first training samples: inserting a first preset number of punctuation marks into the training text based on the number of characters in the training text, wherein the first preset number is less than or equal to the number of characters.

[0083] For example, punctuation marks are randomly inserted into non-entity locations (between characters in the risk entity name) in the training text. Punctuation marks may include at least one of a space, a semicolon, a period, a comma, a quotation mark, an exclamation mark, and an ellipsis. The first preset number may be a certain proportion of the number of characters. For example, for every 10 characters, one-fifth of the number of punctuation marks is inserted.

[0084] An example of a newly obtained second training sample is as follows:

[0085] However, affected by the financial industry policies and the overall environment, Company A and its subsidiaries all suffered losses in 2018. The commercial, poly business, loan assistance business, auto loan business, etc. amounted to 1.972 billion yuan. Accounts receivable could not be recovered, and provisions for bad debts amounted to 112 million yuan. In addition, some loan assistance businesses were overdue, and overdue liabilities of 397 million yuan were recognized.

[0086] Taking the Bert model as an example, after the first training sample is input into the Bert model, a semantic vector is generated based on the contextual semantic information. During the process of extracting semantic information, the position information of each character is taken into account. Therefore, by inserting punctuation, the position information of at least some characters changes relative to the original punctuation structure. Therefore, compared to the Bert model, the second training sample is a new training sample and can generate a different semantic vector. This overcomes the problem of uneven distribution of risk event labels.

[0087] According to the embodiment of the present disclosure, the training method further includes replacing the risk entity in the first training sample and / or the second training sample.

[0088] For example, replacing risk entities can be done independently of inserting punctuation marks, replacing some or all risk entities in N first training samples to obtain a new second training sample. Alternatively, after obtaining the second training sample by inserting punctuation marks, randomly selecting the first training sample and the second training sample for entity replacement. Alternatively, each entity in the first training sample and the second training sample can be randomly replaced with a probability of 50% (for example only).

[0089] If we replace Company A in the above training text, the examples after replacing the entities are as follows:

[0090] However, affected by the financial industry policies and the overall environment, Company B and its subsidiaries all suffered losses in 2018. The commercial, Poly business, loan assistance business, auto loan business, etc. amounted to 1.972 billion yuan. Accounts receivable could not be recovered, and bad debt provisions amounted to 112 million yuan. In addition, some loan assistance businesses were overdue, and overdue liabilities of 397 million yuan were recognized.

[0091] In related technologies, there may be a situation where a certain entity accounts for a large proportion of a certain type of risk, resulting in the model actually only learning about this entity for this type of risk event, and the model's robustness is ultimately low. For example, in news related to financial fraud, more than half of the news is about Company B. In this case, it is very likely that the model will use Company B as a feature of financial fraud. Even if Company B is the subject of a loss event in the future, it will be mistakenly judged as financial fraud. According to the embodiments of the present disclosure, by replacing entities, the imbalance of entity distribution can be avoided, thereby improving the robustness of the risk warning model.

[0092] Figure 4 The flowchart of obtaining risk event distribution information according to an embodiment of the present disclosure is schematically shown.

[0093] like Figure 4As shown, the information of the above-mentioned data source includes risk event distribution information. The acquisition of risk event distribution information in this embodiment includes operations S410 to S430.

[0094] In operation S410 , S first training samples from a corresponding data source are obtained from N first training samples, where S is an integer greater than or equal to 1 and S is less than or equal to N.

[0095] In operation S420 , a risk event label corresponding to each of the S first training samples is determined.

[0096] In operation S430 , risk event distribution information is obtained based on the proportion of each risk event label in the S first training samples.

[0097] For example, risk event distribution information includes the distribution ratio of risk events within the data source. For example, of the 2,000 first training samples, 100 come from financial media A. There are four types of risk event labels: 40 training samples for loss risk events, 30 training samples for leadership change risk events, 20 training samples for financial fraud risk events, and 10 training samples for bankruptcy risk events. The proportions of these four risk event labels are 40%, 30%, 20%, and 10%, respectively.

[0098] According to an embodiment of the present disclosure, by obtaining the risk event distribution information of each data source, the risk warning model can learn the different risk event weights under the corresponding data source through the risk event distribution information, thereby updating the model parameters more accurately.

[0099] Figure 5 The flowchart of obtaining a data source vector according to an embodiment of the present disclosure is schematically shown.

[0100] like Figure 5 As shown, the process of obtaining the data source vector in this embodiment includes operations S510 to S520.

[0101] In operation S510 , a first dimension of a data source vector is determined according to the number of types of risk event labels in N first training samples, wherein each risk event label corresponds to a position in the data source vector.

[0102] For example, if there are six risk event labels, the first dimension is 1*6 (1 row, 6 columns). For example, a loss risk event, a default risk event, a bankruptcy risk event, a leadership change risk event, a rating deterioration risk event, or a financial fraud risk event corresponds to a position in the 1 row, 6 columns vector.

[0103] In operation S520 , the proportion of each risk event label in the S first training samples is used as a vector value of a corresponding position in the data source vector.

[0104] The training samples from financial media A contain four risk event labels: loss risk events, leadership change risk events, financial fraud risk events, and bankruptcy risk events, with a ratio of 40%, 30%, 20%, and 10%, respectively. The data source vector is [0.4, 0, 0.1, 0.3, 0, 0.2].

[0105] Figure 6 The flowchart of obtaining a data source vector according to another embodiment of the present disclosure is schematically shown.

[0106] like Figure 6 As shown, the information of the data source includes the data source identifier of the data source, M data sources correspond to M data source identifiers, and obtaining the data source vector in this embodiment includes operations S610 to S620.

[0107] In operation S610, vector encoding is performed on each of the M data source identifiers.

[0108] For example, the data source identifier can be the data source name. Vector encoding can use methods such as one-hot encoding, word embedding, and binary encoding. Taking one-hot encoding as an example, it is also called single-hot vector encoding. For example, there are four data sources, named A Financial News, B Company, C Blog, and D Government Website. Their respective vector encodings are [1, 0, 0, 0], [0, 1, 0, 0], [0, 0, 1, 0], and [0, 0, 0, 1].

[0109] In operation S620, a data source vector is obtained according to the result of encoding each data source identifier.

[0110] In some embodiments, if the data source information only includes a data source identifier, the vector encoding can be directly used as the data source vector. In other embodiments, if the data source information includes both a data source identifier and risk event distribution information, the data source vector can be a concatenation of the two. For example, using the financial news item A above, the concatenation yields [0.4, 0, 0.1, 0.3, 0, 0.2, 1, 0, 0, 0]. Matrix transformation can also be performed on this basis, which is not limited here.

[0111] According to the embodiments of the present disclosure, using data source identifiers as input to train a risk warning model allows the model to learn the characteristics of each dimension of the data source. After each training text is concatenated with the data source identifier, the model updates the weights of each layer of the neural network through gradient descent after processing N training texts. Therefore, during the learning process, the format characteristics, risk event distribution, and stylistic characteristics of each data source can be obtained based on the correspondence between the data source identifier and the training text.

[0112] According to the embodiments of the present disclosure, if the data source identifier and risk event distribution information are input together, then for the model, the risk event distribution information is equivalent to certain prior information, eliminating the learning process. Other characteristics such as format characteristics and stylistic characteristics can be learned based on the data source identifier.

[0113] According to an embodiment of the present disclosure, author information, writing style, or text format data may also be vectorized and input into the risk warning model to provide certain prior information.

[0114] Figure 7 The flowchart of merging a text vector and a data source vector according to an embodiment of the present disclosure is schematically shown.

[0115] like Figure 7 As shown, the merging of the text vector and the data source vector in this embodiment includes operations S710 to S730.

[0116] In operation S710, a vector conversion model is constructed according to the coding layer in the pre-trained model.

[0117] In some embodiments, the BERT model includes a Token Embedding layer (i.e., encoding layer). This layer first converts the data representation of each word in a sentence into a one-hot representation, then performs a matrix transformation on the encoded data to obtain a text vector. Based on the calculation method used by the Embedding layer to obtain text vectors, a vector conversion model is constructed to implement the same vector acquisition process as the Embedding layer.

[0118] In operation S720 , the data source vector is input into a vector conversion model for conversion.

[0119] In operation S730 , the converted data source vector and the text vector are added or concatenated.

[0120] Exemplarily, the role of the vector conversion model is that the converted data source vector has the same dimension as the text vector, which facilitates addition or concatenation operations, thereby facilitating subsequent neural network processing.

[0121] According to an embodiment of the present disclosure, training a risk warning model includes training the risk warning model X times, where X is an integer greater than or equal to 2. The method also includes obtaining a new second training sample after each training of the risk warning model until X training times are completed, specifically including: performing data enhancement and / or replacing risk entities before each training of the risk warning model.

[0122] For example, a single training run of a risk prediction model can be called an epoch. An epoch completes one forward propagation and one backward propagation for all training samples in deep learning. Before each epoch, data augmentation is performed based on the N first training samples. Each epoch uses the N first training samples and then obtains new second training samples for training. This allows the model to see new samples with each round of training, allowing it to learn richer semantic information, avoid overfitting, and ultimately improve robustness.

[0123] According to the embodiments of the present disclosure, dynamic negative sampling can also be performed during each training of the risk warning model. Specifically, negative samples can be randomly extracted at each epoch, and the number of negative samples is also consistent with the number of first risk event labels. For example, 30 items can be randomly extracted from 100 texts without entities.

[0124] Figure 8 The flowchart of the risk warning method according to the embodiment of the present disclosure is schematically shown.

[0125] like Figure 8 As shown, the risk warning method of this embodiment includes operations S810 to S840.

[0126] In operation S810 , a text to be recognized and information about a data source to which the text to be recognized belongs are obtained.

[0127] In operation S820 , a first vector to be recognized is obtained according to the text to be recognized, and a second vector to be recognized is obtained according to information of the data source.

[0128] Exemplarily, the method for obtaining the first vector to be identified and the second vector to be identified may refer to the method for obtaining the text vector and the data source vector described in the above embodiment, and will not be described in detail here.

[0129] In operation S830 , the first vector to be identified and the second vector to be identified are combined to obtain a third vector to be identified.

[0130] Exemplarily, the method for obtaining the third vector to be identified may refer to the method for obtaining the training vector described in the above embodiment, and will not be described in detail here.

[0131] In operation S840, the third vector to be identified is input into the risk warning model to obtain a prediction result, wherein the prediction result includes a predicted risk entity and a predicted risk event. The risk warning model is composed of Figures 2 to 7 The method introduced is obtained by training.

[0132] The embodiments of the present disclosure provide a risk warning method. The risk warning model obtained by the training method of the embodiments of the present disclosure can take into account the text to be identified and the data source information to which it belongs, and output predicted risk entities and predicted risk events more accurately, thereby improving the risk warning effect.

[0133] Based on the above training method and risk warning method, the present disclosure also provides a training device and a risk identification device. Figure 9 and Figure 10 The above-mentioned device is described in detail.

[0134] Figure 9 The structural block diagram of the training device according to an embodiment of the present disclosure is schematically shown.

[0135] like Figure 9 As shown, the training device 900 of this embodiment includes a first acquisition module 910 , a first vector module 920 , a first merging module 930 , a first input module 940 and a model training module 950 .

[0136] The first acquisition module 910 can execute operation S210 to obtain N first training samples from M data sources, wherein each first training sample includes information about the training text and its data source, and label information of the training text, the label information includes a risk entity label and a risk event label, and N and M are integers greater than or equal to 1 respectively.

[0137] The first vector module 920 may perform operation S220 to obtain a text vector according to the training text, and obtain a data source vector according to information of the corresponding data source.

[0138] According to an embodiment of the present disclosure, the first vector module 920 may further perform operations S410 to S430, S510 to S520, and S510 to S520, which are not described in detail here.

[0139] The first merging module 930 may perform operation S230 to merge the text vector and the data source vector corresponding to each first training sample to obtain N training vectors corresponding to the N first training samples.

[0140] According to an embodiment of the present disclosure, the first vector module 920 may further perform operations S710 to S730, which will not be described in detail herein.

[0141] The first input module 940 may perform operation S240 to input each training vector into the risk warning model to obtain a prediction result, wherein the prediction result includes a predicted risk entity and a predicted risk event.

[0142] The model training module 950 may perform operation S250 to train a risk warning model based on an error between a prediction result and label information.

[0143] According to an embodiment of the present disclosure, the training device 900 may further include a data enhancement module. The data enhancement module may be used to perform operations S310 to S330, which will not be described in detail herein.

[0144] Figure 10 The structural block diagram of the risk warning device according to an embodiment of the present disclosure is schematically shown.

[0145] like Figure 10 As shown, the risk warning device 1000 of this embodiment includes a second acquisition module 1010 , a second vector module 1020 , a second merging module 1030 , and a second input module 1040 .

[0146] The second acquisition module 1010 may perform operation S810 to acquire the text to be recognized and information about the data source to which the text to be recognized belongs.

[0147] The second vector module 1020 may perform operation S820 to obtain a first vector to be recognized according to the text to be recognized, and obtain a second vector to be recognized according to information of a corresponding data source.

[0148] The second merging module 1030 may perform operation S830 to merge the first vector to be identified and the second vector to be identified to obtain a third vector to be identified.

[0149] The second input module 1040 can perform operation S840 to input the third vector to be identified into the risk warning model to obtain a prediction result, wherein the prediction result includes a predicted risk entity and a predicted risk event, and the risk warning model is obtained by training the device of claim 12.

[0150] It should be noted that the implementation methods, technical problems solved, functions implemented, and technical effects achieved of each module / unit / sub-unit in the device part embodiment are the same or similar to the implementation methods, technical problems solved, functions implemented, and technical effects achieved of each corresponding step in the method part embodiment, and will not be repeated here.

[0151] According to an embodiment of the present disclosure, any multiple modules in the training device 900 or the risk warning device 1000 can be combined into a single module, or any one of the modules can be split into multiple modules. Alternatively, at least part of the functionality of one or more of these modules can be combined with at least part of the functionality of other modules and implemented in a single module.

[0152] According to an embodiment of the present disclosure, at least one module in the training device 900 or the risk warning device 1000 can be at least partially implemented as a hardware circuit, such as a field programmable gate array (FPGA), a programmable logic array (PLA), a system on a chip, a system on a substrate, a system on a package, an application-specific integrated circuit (ASIC), or can be implemented by hardware or firmware such as any other reasonable way of integrating or packaging the circuit, or implemented in any one of the three implementation methods of software, hardware and firmware or in an appropriate combination of any of them. Alternatively, at least one module in the training device 900 or the risk warning device 1000 can be at least partially implemented as a computer program module, which can perform corresponding functions when the computer program module is run.

[0153] Figure 11 A block diagram of an electronic device suitable for implementing a training method or a risk warning method according to an embodiment of the present disclosure is schematically shown.

[0154] like Figure 11 As shown, the electronic device 1100 according to an embodiment of the present disclosure includes a processor 1101, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 1102 or a program loaded from a storage part 1108 into a random access memory (RAM) 1103. The processor 1101 may, for example, include a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or a related chipset and / or a dedicated microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 1101 may also include an onboard memory for caching purposes. The processor 1101 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of the present disclosure.

[0155] Various programs and data required for the operation of the electronic device 1100 are stored in the RAM 1103. The processor 1101, the ROM 1102, and the RAM 1103 are connected to each other via a bus 1104. The processor 1101 performs various operations of the method flow according to the embodiment of the present disclosure by executing the programs in the ROM 1102 and / or the RAM 1103. It should be noted that the programs may also be stored in one or more memories other than the ROM 1102 and the RAM 1103. The processor 1101 may also perform various operations of the method flow according to the embodiment of the present disclosure by executing the programs stored in the one or more memories.

[0156] According to an embodiment of the present disclosure, electronic device 1100 may further include an input / output (I / O) interface 1105, which is also connected to bus 1104. Electronic device 1100 may also include one or more of the following components connected to I / O interface 1105: an input section 1106 including a keyboard, mouse, etc.; an output section 1107 including devices such as a cathode ray tube (CRT), liquid crystal display (LCD), and speakers; a storage section 1108 including a hard disk; and a communication section 1109 including a network interface card such as a LAN card or modem. Communication section 1109 performs communication processing via a network such as the Internet. A drive 1110 is also connected to I / O interface 1105 as needed. Removable media 1111, such as a magnetic disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed in drive 1110 as needed, so that computer programs read from the removable media can be installed into storage section 1108 as needed.

[0157] The present disclosure also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments, or may exist independently and not be incorporated into the device / apparatus / system. The computer-readable storage medium carries one or more programs, and when executed, implements the method according to the embodiments of the present disclosure.

[0158] According to an embodiment of the present disclosure, a computer-readable storage medium may be a non-volatile computer-readable storage medium, such as but not limited to: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, apparatus, or device. For example, according to an embodiment of the present disclosure, a computer-readable storage medium may include the ROM 1102 and / or RAM 1103 described above and / or one or more memories other than ROM 1102 and RAM 1103.

[0159] The embodiments of the present disclosure also include a computer program product, which includes a computer program containing program code for executing the method shown in the flowchart. When the computer program product is run in a computer system, the program code is used to enable the computer system to implement the method provided by the embodiments of the present disclosure.

[0160] The computer program executes the above functions defined in the system / device of the embodiment of the present disclosure when the computer program is executed by the processor 1101. According to the embodiment of the present disclosure, the system, device, module, unit, etc. described above can be implemented by a computer program module.

[0161] In one embodiment, the computer program may be stored on a tangible storage medium such as an optical storage device or a magnetic storage device. In another embodiment, the computer program may be transmitted and distributed in the form of a signal on a network medium, downloaded and installed via the communication portion 1109, and / or installed from removable media 1111. The program code contained in the computer program may be transmitted using any appropriate network medium, including but not limited to wireless, wired, or any suitable combination thereof.

[0162] In such an embodiment, the computer program can be downloaded and installed from a network via the communication section 1109 and / or installed from the removable medium 1111. When the computer program is executed by the processor 1101, the above-described functions defined in the system of the embodiment of the present disclosure are performed. According to the embodiment of the present disclosure, the systems, devices, means, modules, units, etc. described above can be implemented by computer program modules.

[0163] According to an embodiment of the present disclosure, the program code for executing the computer program provided by the embodiment of the present disclosure can be written in any combination of one or more programming languages. Specifically, these computer programs can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages ​​include, but are not limited to, Java, C++, Python, "C" language or similar programming languages. The program code can be executed entirely on the user computing device, partially on the user device, partially on a remote computing device, or entirely on a remote computing device or server. In cases involving a remote computing device, the remote computing device can be connected to the user computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computing device (for example, using an Internet service provider to connect via the Internet).

[0164] The flowcharts and block diagrams in the accompanying drawings illustrate the possible implementation architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present disclosure. In this regard, each box in the flowchart or block diagram can represent a module, program segment, or a part of code, and the above-mentioned module, program segment, or a part of code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in an order different from that marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flowchart, and the combination of boxes in the block diagram or flowchart, can be implemented with a dedicated hardware-based system that performs the specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0165] Those skilled in the art will appreciate that the features described in the various embodiments and / or claims of this disclosure may be combined and / or coupled in various ways, even if such combinations and / or couplings are not explicitly described in this disclosure. In particular, the features described in the various embodiments and / or claims of this disclosure may be combined and / or coupled in various ways without departing from the spirit and teachings of this disclosure. All such combinations and / or couplings are intended to fall within the scope of this disclosure.

[0166] The embodiments of the present disclosure are described above. However, these embodiments are for illustrative purposes only and are not intended to limit the scope of the present disclosure. Although each embodiment has been described separately above, this does not mean that the measures in each embodiment cannot be used in combination to advantage. The scope of the present disclosure is defined by the appended claims and their equivalents. Without departing from the scope of the present disclosure, those skilled in the art may make various substitutions and modifications, which should all fall within the scope of the present disclosure.

Claims

1. A training method for a risk warning model, comprising: Obtain N first training samples from M data sources, where each first training sample includes a training text and information about its data source, and label information of the training text, the data source information includes a data source identifier and risk event distribution information of the data source, and the label information includes a risk entity label and a risk event label, where N and M are integers greater than or equal to 1; Obtaining a text vector according to the training text, and obtaining a data source vector by concatenating the data source identifier and the risk event distribution information; Merging the text vector and the data source vector corresponding to each of the first training samples to obtain N training vectors corresponding to the N first training samples; Inputting each of the training vectors into a risk warning model to obtain a prediction result, wherein the prediction result includes a predicted risk entity and a predicted risk event; The risk warning model is trained based on the error between the prediction result and the label information.

2. The method according to claim 1, further comprising obtaining the risk event distribution information, specifically comprising: Obtaining S first training samples from the data source from the N first training samples, where S is an integer greater than or equal to 1 and S is less than or equal to N; Determine a risk event label corresponding to each of the S first training samples; The risk event distribution information is obtained according to the proportion of each risk event label in the S first training samples.

3. The method according to claim 2, wherein: The obtaining of the data source vector according to the information of the data source includes: Determining a first dimension of the data source vector according to the number of types of risk event labels in the N first training samples, wherein each type of risk event label corresponds to a position in the data source vector; and The proportion of each risk event label in the S first training samples is used as the vector value of the corresponding position in the data source vector.

4. The method according to any one of claims 1 or 3, wherein The M data sources correspond to M data source identifiers, and obtaining a data source vector according to information of the corresponding data sources includes: Performing vector encoding on each of the M data source identifiers; The data source vector is obtained according to the result of encoding each data source identifier.

5. The method according to claim 4, wherein Obtaining a text vector according to the training text includes: The training text is input into a pre-training model to obtain the text vector.

6. The method according to claim 5, wherein: The pre-training model includes an encoding layer, and the encoding layer is used to obtain the text vector according to the training text. The method further includes: constructing a vector conversion model according to the coding layer; Inputting the data source vector into the vector conversion model for conversion; The merging of the text vector and the data source vector corresponding to each of the first training samples includes: The converted data source vector is added or concatenated with the text vector.

7. The method according to claim 1, wherein After obtaining the N first training samples, the method further includes: Determining the number of each risk event label in the N first training samples, wherein each first training sample corresponds to a risk event label; Determining a first risk event label based on the number of each risk event label, wherein the first risk event label is the risk event label with the largest number; Data enhancement is performed on the P first training samples corresponding to the second risk event label to obtain Q second training samples, wherein the Q second training samples are used to train the risk warning model, the second risk event label is any label other than the first risk event label, and P and Q are integers greater than or equal to 1 respectively.

8. The method according to claim 7, wherein: The data enhancement of the P first training samples corresponding to the second risk event label includes: performing the following operations on the training text of each first training sample in the P first training samples: According to the number of characters in the training text, a first preset number of punctuation marks is inserted into the training text, wherein the first preset number is less than or equal to the number of characters.

9. The method according to any one of claims 1 or 7, wherein The method further includes replacing risk entities in the first training sample and / or the second training sample.

10. The method according to claim 9, wherein: The training of the risk warning model includes training the risk warning model X times, where X is an integer greater than or equal to 2. The method further includes obtaining a new second training sample after each training of the risk warning model until the X training times are completed, specifically including: Before each training of the risk warning model, the operations of data enhancement and / or replacement of risk entities are performed.

11. A risk early warning method, comprising: Obtaining a text to be recognized and information about a data source to which the text to be recognized belongs; Obtaining a first vector to be recognized based on the text to be recognized, and obtaining a second vector to be recognized based on information of the data source; Merging the first vector to be identified and the second vector to be identified to obtain a third vector to be identified; The third vector to be identified is input into a risk warning model to obtain a prediction result, wherein the prediction result includes a predicted risk entity and a predicted risk event, and the risk warning model is trained by the method according to any one of claims 1 to 10.

12. A training device for a risk warning model, comprising: A first acquisition module is configured to acquire N first training samples from M data sources, wherein each first training sample includes a training text and information about its data source, and label information of the training text, wherein the information about the data source includes a data source identifier and risk event distribution information of the data source, and the label information includes a risk entity label and a risk event label, and N and M are integers greater than or equal to 1; A first vector module is configured to obtain a text vector based on the training text, and to obtain a data source vector by concatenating the data source identifier and the risk event distribution information; A first merging module, configured to merge the text vector and the data source vector corresponding to each of the first training samples to obtain N training vectors corresponding to the N first training samples; A first input module is configured to input each of the training vectors into a risk warning model to obtain a prediction result, wherein the prediction result includes a predicted risk entity and a predicted risk event; A model training module is used to train the risk warning model based on the error between the prediction result and the label information.

13. A risk warning device comprising: A second acquisition module is used to acquire the text to be recognized and information about the data source to which the text to be recognized belongs; A second vector module is configured to obtain a first vector to be recognized based on the text to be recognized, and to obtain a second vector to be recognized based on information of the data source; a second merging module, configured to merge the first vector to be identified and the second vector to be identified to obtain a third vector to be identified; The second input module is used to input the third vector to be identified into the risk warning model to obtain a prediction result, wherein the prediction result includes a predicted risk entity and a predicted risk event, and the risk warning model is obtained by training the device according to claim 12.

14. An electronic device comprising: one or more processors; a storage device for storing one or more programs, When the one or more programs are executed by the one or more processors, the one or more processors are enabled to execute the method according to any one of claims 1 to 11.

15. A computer-readable storage medium having executable instructions stored thereon, which, when executed by a processor, causes the processor to perform the method according to any one of claims 1 to 11.

16. A computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the method according to any one of claims 1 to 11 is implemented.

Images