Detailed Description
The following description of the embodiments of the present application will be made more apparent and fully hereinafter with reference to the accompanying drawings, in which some, but not all embodiments of the application are shown. All other embodiments, which can be made by those skilled in the art based on the embodiments of the application without making any inventive effort, are intended to be within the scope of the application.
Although a logical order of illustration is depicted in the flowchart, in some cases the operations shown and described may be performed in an order different than shown in the figures.
FIG. 1 illustrates a flow diagram of a fraud prediction method provided by some embodiments of the present application. The fraud prediction method may be applied to mobile devices including, but not limited to, smart phones, tablet computers, smart watches, and gaming machines or similar devices. The specific flow of the fraud prediction method may be as follows:
S101, obtaining content to be detected in the mobile equipment.
Wherein the content to be detected may be an object for determining a fraud probability. The specific form of the content to be detected may be various, for example, image, text or voice.
In some embodiments, the content to be detected may be an image. For example, the fraud prediction method may acquire an image displayed by a display interface of the mobile device. For example, the fraud method may obtain a multimedia message or an image in a rich media communication received by the mobile device.
In some embodiments, the content to be detected may be text. For example, the fraud prediction method may obtain text of a short message received by the mobile device. For example, the fraud prediction method may obtain chat text in instant messaging software or social media software in the mobile device.
In some embodiments, the content to be detected may be speech. For example, the fraud prediction method may obtain the speech of the mobile device during the telephone communication. For example, the fraud prediction method may obtain the voice of the instant messaging software or the social media software of the mobile device in real-time voice communication or voice message.
S102, determining the fraud probability of the content to be detected based on the fraud prediction model.
The fraud prediction model may be a trained machine learning model or a deep learning model, among others. The fraud prediction method may further comprise fine tuning the pre-training model to obtain a trained fraud prediction model, prior to operation S102. The fraud prediction model may comprise one or more models. For example, different models or the same model may be used for different types of content to be detected to derive the fraud probability. For example, a model or models may be used for the same type of content to be detected to derive the fraud probability. The training method of the fraud prediction model will be described later in connection with specific embodiments.
Wherein the probability of fraud can be used to evaluate the likelihood of including fraud information in the content to be detected, the greater the probability of fraud, the higher the likelihood of including fraud information, the less the probability of fraud, and the lower the likelihood of including fraud information. The fraud probability may be a continuously distributed probability value, for example, the fraud probability may be any probability value within an interval of 0% to 100%. The fraud probability may also be a discrete distribution of probability ratings, e.g. the fraud probability may be divided into three ratings, high, medium and low. The fraud probability may also be a judgment as to whether it is fraud, for example, the fraud probability may be classified as high, considered as fraud, and low, considered as not fraud.
In some embodiments, after deriving the fraud probability, a corresponding operation may be performed according to the fraud probability. For example, the corresponding operation may be to generate reminders of different risk levels according to the fraud probability. When the fraud probability is in the first probability interval, the risk level is lower, and only the popup window is used for reminding. When the fraud probability is within the second probability interval, a confirmation dialog box is popped up, and the dialog box can be closed after the user needs to confirm. When the fraud probability is in the third probability interval, the risk level is higher, and the user is required to release the use restriction through the safe question and answer. Of course, the probability interval or the reminding mode can be preset by the fraud prediction method or set by the user. For example, the corresponding operation may be to mask the message according to the fraud probability. As another example, the corresponding operation may be to limit the behavior of the user, such as a transfer transaction, according to the probability of fraud.
The fraud prediction method provided by some embodiments of the present application can be applied to a mobile device, acquire content to be detected in the mobile device, and determine fraud probability of the content to be detected based on a fraud prediction model. According to the fraud prediction method provided by some embodiments of the application, the fraud probability is predicted through the fraud prediction model, so that the intelligent level of fraud prediction is improved.
As previously mentioned, the content to be detected may be in the form of an image. In some embodiments, acquiring the content to be detected in the mobile device in operation S101 may include:
And acquiring a display interface of the mobile equipment according to the preset frequency, and acquiring the content to be detected in the display interface.
The acquiring the display interface of the mobile device may be acquiring the display interface according to a preset frequency when the mobile device is in a bright screen state. The acquiring of the content to be detected in the display interface may be acquiring an image of the whole display interface or acquiring an image of a partial area of the display interface. For example, fraud prediction methods in some embodiments may determine the location of a message bullet window based on user settings or the model of the mobile device, acquiring an image of the message bullet window area.
The preset frequency may be set in the fraud prediction method or set by the user himself. The preset frequency may be a fixed frequency or an adjustable frequency. For example, the fraud prediction method in some embodiments may adjust the preset frequency according to the mobile phone power or the power consumption situation.
In some embodiments, the fraud prediction model that fraud predicts the content to be detected in the form of an image may include a word recognition model as well as a content recognition model. FIG. 2 illustrates a flow chart for determining fraud probabilities provided by some embodiments of the present application. Determining the fraud probability of the content to be detected based on the fraud prediction model may comprise:
S201, text recognition is carried out on the content to be detected based on the text recognition model, and the text to be detected is obtained.
The content to be detected may be an image to be detected, and the text recognition model may be used to recognize text in the image to be detected. The word recognition model may be one or more models.
In some embodiments, the word recognition model may be a model of a concatenation of a text detection model and a character recognition model. The text detection model can be used for detecting a text region in an image to be detected, then inputting the text region image into the character recognition model, and outputting characters in the text region image, namely the text to be detected. The text detection model may use a general purpose object detection model, such as a YOLO series model, an SSD model, a fast R-CNN model, or a modified model based on these models. The text detection model may also be a dedicated detection model for the text field, e.g., EAST detection model (EFFICIENT AND Accurate Scene Text Detector), CTPN model (Column Proposal Networks).
In some embodiments, the text recognition model may be an end-to-end text recognition model, and the text to be detected may be directly output according to the input image to be detected.
S202, carrying out content recognition on the text to be detected based on the content recognition model to obtain fraud probability.
The content recognition model can be used for analyzing and understanding the text to be detected to obtain fraud probability.
In some embodiments, the content recognition model may be a natural language processing model. The natural language processing model may be a Bag of Words model (Bag of Words, boW), an N-gram model, a hidden markov model (Hidden Markov Model, HMM), a recurrent neural network (Recurrent Neural Network, RNN), or a Transformer model.
In order to obtain a better prediction effect in the task of fraud prediction, the content recognition pre-training model can be fine-tuned using data of the fraud prediction domain. For this purpose, a training sample set and a test sample set may be constructed, each sample being a piece of text content, one sample may correspond to a tag, which is used to indicate whether fraud information is present in the sample. For example, when fraud information exists in the sample, the corresponding tag is 1, no fraud content exists in the sample, and the corresponding tag is 0.
In some embodiments, it may be desirable to convert text in character form to feature vectors prior to entering the content recognition pre-training model. And continuously adjusting model parameters according to the output result of the training sample after the content recognition pre-training model and the difference degree between the labels so as to obtain a trained content recognition model.
The training process of the content recognition model may be performed in a server. The trained content recognition model may be stored in the mobile device, or in the cloud device. Based on the trained content recognition model, the text to be detected or the feature vector of the text to be detected can be input into the content recognition model, and the fraud probability is output.
According to the fraud prediction method provided by some embodiments of the application, when the fraud probability of the content to be detected in the image form is predicted, the text in the image to be detected is extracted first, and then the content identification is performed on the extracted text, so that the fraud probability is determined.
To increase the speed of fraud prediction, in some embodiments, the prediction of fraud probability may be performed using a multi-level recognition model. For example, the content recognition model may include a first recognition model and a second recognition model, wherein the first recognition model is less computationally intensive than the second recognition model. For the same input data and the same computing device, the time taken to make an inference using the first recognition model is less than the time taken to make an inference using the second recognition model. And, at least over the set of test samples, the accuracy or recall of the first recognition model is lower than the second recognition model.
FIG. 3 illustrates a flow diagram of multi-level content identification provided by some embodiments of the application. The method for identifying the content of the text to be detected based on the content identification model, to obtain the fraud probability, specifically comprises the following steps:
s301, obtaining a first fraud probability of the text to be detected based on the first recognition model.
The first fraud probability can be used for initially judging whether the text to be detected contains fraud information or not, and determining whether further prediction is needed by using the second recognition model or not according to the initial judgment result.
S302, when the first fraud probability is smaller than the first threshold value or larger than the second threshold value, the first fraud probability is taken as the fraud probability.
When the first fraud probability is smaller or larger, the result obtained by the first recognition model can be considered to be more reliable, namely the first fraud probability can be adopted as the fraud probability.
S303, obtaining the fraud probability of the text to be detected based on the second recognition model when the first fraud probability is greater than or equal to the first threshold value and less than or equal to the second threshold value.
When the first fraud probability is in the middle zone, the first recognition model can be considered to be difficult to judge whether the text to be detected contains fraud information, and at the moment, the second recognition model can be used for further prediction to obtain the fraud probability.
It should be noted that the first recognition model and the second recognition model may be stored together in the mobile device, or the first recognition model and the second recognition model may be stored together in the cloud device, or the first recognition model may be stored in the mobile device, and the second recognition model may be stored in the cloud device.
Some embodiments of the present application provide a fraud prediction method that can increase the prediction speed of fraud probability by using a multi-level content recognition model. Under the condition that the result obtained by using the lightweight first recognition model is more reliable, the second recognition model is needed to be used for secondary judgment under the condition that the result is less.
In other embodiments, the fraud prediction model that fraud predicts the content to be detected in the form of an image may be an end-to-end image semantic understanding model. The image semantic understanding model may receive input in the form of an image and analyze whether fraud information exists in the input image to be detected.
Also, to obtain better prediction effect, a training sample set and a test sample set may be constructed, each sample is an image, and one sample may correspond to one label, where the label is used to indicate whether fraud information exists in the sample. For example, when fraud information exists in the sample, the corresponding tag is 1, no fraud content exists in the sample, and the corresponding tag is 0. For another example, when fraud information is present in the sample, the corresponding tag may be what type of fraud is occurring in the image, and when fraud content is not present in the sample, the corresponding tag may be unoccupied. And continuously adjusting model parameters according to the difference degree between the output result of the training sample after passing through the image semantic understanding model and the label so as to obtain a trained image semantic understanding model.
Likewise, the training process of the image semantic understanding model may be performed in a server. The trained image semantic understanding model can be stored in the mobile device or the cloud device. Based on the trained image semantic understanding model, the image to be detected can be input into the image semantic understanding model, and the fraud probability is output.
It will be appreciated that for other forms of content to be detected, the fraud prediction model may be other models that match the input form. For example, for content to be detected in text form, the prediction of fraud probability may be performed using the content recognition model described above as a fraud prediction model. For example, for content to be detected in the form of speech, a speech recognition model may be used as a fraud prediction model for prediction of fraud probability.
Similar to the two-stage (word recognition+content recognition) fraud prediction, the single-stage fraud prediction may also use a multi-level fraud prediction model for the prediction of fraud probability. In some embodiments, the fraud prediction model may include a first prediction model and a second prediction model, wherein the first prediction model is less computationally intensive than the second prediction model. For the same input data and the same computing device, the time taken to make an inference using the first predictive model is less than the time taken to make an inference using the second predictive model. And, at least over the set of test samples, the accuracy or recall of the first predictive model is lower than the second predictive model.
FIG. 4 illustrates a flow diagram of multi-level fraud prediction provided by some embodiments of the present application. The determining the fraud probability of the content to be detected based on the fraud prediction model may specifically include:
S401, obtaining a second fraud probability of the content to be detected based on the first prediction model.
The first prediction model may be a model for performing fraud probability prediction on the content to be detected in the form of an image, text or voice. The second fraud probability can be used for initially judging whether the content to be detected contains fraud information or not, and determining whether further prediction is needed by using a second prediction model or not according to the initial judgment result.
S402, when the second fraud probability is smaller than the first threshold value or larger than the second threshold value, taking the second fraud probability as the fraud probability.
When the second fraud probability is smaller or larger, the result obtained by the first prediction model can be considered to be more reliable, i.e. the second fraud probability can be adopted as the fraud probability.
S403, obtaining the fraud probability of the content to be detected based on the second prediction model when the second fraud probability is greater than or equal to the first threshold value and less than or equal to the second threshold value.
When the second fraud probability is in the middle zone, the first prediction model can be considered to be difficult to judge whether the content to be detected contains fraud information, and at this time, the second prediction model can be used for further prediction to obtain the fraud probability.
It should be noted that the first prediction model and the second prediction model may be stored together in the mobile device, or the first prediction model and the second prediction model may be stored together in the cloud device, or the first prediction model may be stored in the mobile device, and the second prediction model may be stored in the cloud device.
Some embodiments of the present application provide a fraud prediction method that can increase the prediction speed of fraud probability by using a multi-level fraud prediction model. Under the condition that the result obtained by using the lightweight first prediction model is more reliable, the second prediction model is needed to be used for secondary judgment under the condition that fewer conditions exist.
Fig. 5 is a schematic structural diagram of a prediction apparatus according to some embodiments of the present application. The prediction apparatus 500 may be applied to a mobile device, and may specifically include:
the acquiring module 501 may be configured to acquire content to be detected in a mobile device.
The determination module 502 may be used for determining the fraud probability of the content to be detected based on the fraud prediction model.
Some embodiments of the present application may first acquire content to be detected in a mobile device by the acquisition module 501. The determination module 502 may then determine a fraud probability for the content to be detected based on the fraud prediction model. The prediction apparatus 500 provided by some embodiments of the present application can predict fraud probability through a fraud prediction model, improving the level of intellectualization of fraud prediction.
In some embodiments, the determination acquisition module 501 may include a first acquisition unit and a second acquisition unit. The first obtaining unit may be configured to obtain a display interface of the mobile device according to a preset frequency. The second obtaining unit may be configured to obtain the content to be detected in the display interface.
In some embodiments, the determination module 502 may include a text recognition unit and a content recognition unit. The text recognition unit can be used for recognizing the text of the content to be detected based on the text recognition model to obtain the text to be detected. The content recognition unit can be used for carrying out content recognition on the text to be detected based on the content recognition model to obtain fraud probability.
In some embodiments, the content recognition unit may include a first recognition subunit and a second recognition subunit. Wherein the first recognition subunit may be configured to obtain a first fraud probability for the text to be detected based on the first recognition model. The second recognition subunit may be configured to obtain, based on the second recognition model, a fraud probability of the text to be detected with the first fraud probability as the fraud probability if the first fraud probability is smaller than the first threshold or larger than the second threshold, or if the first fraud probability is larger than or equal to the first threshold and smaller than or equal to the second threshold.
In some embodiments, the determination module 502 may include a first prediction unit and a second prediction unit. The first prediction unit may be configured to obtain a second fraud probability of the content to be detected based on the first prediction model. The second prediction unit may be configured to obtain the fraud probability of the content to be detected based on the second prediction model, with the first fraud probability as the fraud probability when the second fraud probability is smaller than the first threshold or larger than the second threshold, or when the second fraud probability is larger than or equal to the first threshold and smaller than or equal to the second threshold.
In addition, the application also provides a mobile device, and fig. 6 shows a schematic structural diagram of the mobile device according to some embodiments of the application. The mobile device 600 includes, but is not limited to, a smart phone, tablet, smart watch, and gaming machine or the like. Specifically, the present application relates to a method for manufacturing a semiconductor device.
The mobile device 600 may include components such as a processor 601 of one or more processing cores, a memory 602 of one or more computer readable storage media, and the like. Those skilled in the art will appreciate that the mobile device 600 structure shown in fig. 6 is not limiting of the mobile device 600 and may include more or fewer components than shown, or may combine certain components, or a different arrangement of components. Wherein:
The processor 601 is a control center of the mobile device 600, connects various parts of the entire mobile device 600 using various interfaces and lines, and performs various functions of the mobile device 600 and processes data by running or executing software programs and/or modules stored in the memory 602, and invoking data stored in the memory 602, thereby performing overall monitoring of the mobile device 600. Alternatively, the processor 601 may include one or more processing cores, and preferably the processor 601 may integrate an application processor that primarily processes operating systems, user interfaces, applications, etc., and a modem processor that primarily processes wireless communications. It will be appreciated that the modem processor described above may not be integrated into the processor 601.
The memory 602 may be used to store software programs and modules, and the processor 601 may execute various functional applications and data processing by executing the software programs and modules stored in the memory 602. The memory 602 may mainly include a storage program area that may store an operating system, application programs required for at least one function, and the like, and a storage data area that may store data created according to the use of the mobile device 600, and the like. In addition, the memory 602 may include high-speed random access memory, and may also include non-volatile memory, such as at least one magnetic disk storage device, flash memory device, or other volatile solid-state storage device. Accordingly, the memory 602 may also include a memory controller to provide access to the memory 602 by the processor 601.
In particular, in this embodiment, the processor 601 in the mobile device 600 loads executable files corresponding to the processes of one or more application programs into the memory 602 according to the following instructions, and the processor 601 runs the application programs stored in the memory 602, so as to implement the operations in any fraud prediction method provided by the embodiment of the present application.
Specific processes of the mobile device 600 performing the fraud prediction method may refer to descriptions in fig. 2 to 5, and the present application is not described herein.
Those of ordinary skill in the art will appreciate that all or a portion of the operations in the various methods of the above embodiments may be performed by instructions, or by instructions controlling associated hardware, which may be stored in a computer-readable storage medium and loaded and executed by a processor.
To this end, the present application provides a computer readable storage medium having stored thereon a computer program that is capable of being loaded by a processor to perform the operations of any of the fraud prediction methods provided by the present application.
The specific implementation of each operation above may be referred to the previous embodiments, and will not be described herein.
The computer readable storage medium may include, among others, read Only Memory (ROM), random access Memory (RAM, random Access Memory), magnetic or optical disks, and the like.
Because the instructions stored in the computer readable storage medium can execute the operations in any of the fraud prediction methods provided by the present application, the beneficial effects that any of the fraud prediction methods provided by the present application can achieve can be achieved, which are detailed in the foregoing embodiments and are not described herein.
It should be noted that, in the fraud prediction method provided by the embodiment of the present application, the principle of protecting user privacy is strictly followed in the process of data acquisition and prediction. Before any personal information is collected, a clear, detailed description of the information is provided to the user, including but not limited to data type, collection purpose, range of use, shelf life, etc. The user must indicate that he has read and understood the information and voluntarily give authorization for data collection and specific use purposes by an explicit and affirmative action, such as clicking a confirmation button or turning on a function switch.
The above detailed description of a fraud prediction method, a prediction device, a mobile device and a storage medium provided by the present application has been provided, and specific examples are provided herein to illustrate the principles and embodiments of the present application, and the above description of the examples is only for aiding in understanding the method and core concept of the present application, and meanwhile, for those skilled in the art, according to the concept of the present application, there are variations in the specific embodiments and application scope, and in summary, the present disclosure should not be construed as limiting the present application.