Terminal adaptability determination method, device, electronic device and storage medium
By comparing the use of learning frameworks and feature encoders, the problem of low adaptability judgment between user terminals and soft terminals is solved, fast and accurate adaptability judgment is achieved, and the efficiency of soft terminal promotion is improved.
Patent Information
- Application Number
- CN202211385809.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-11-07
- Publication Date
- 2024-11-12
- Estimated Expiration
- 2042-11-07
AI Technical Summary
The prior art is difficult to quickly and accurately determine whether the user terminal is adapted to a soft terminal application based on the DPI information of the user terminal, especially in the case of insufficient sample data and incomplete known adaptation information, resulting in low judgment efficiency.
By obtaining the deep message detection information of the terminal to be detected, input it to the pre-trained feature encoder for feature abstraction processing, calculate the similarity between the feature vector to be detected and the target feature vector, use the comparison learning framework to perform self-supervised learning, and extract High-quality feature vectors to judge terminal adaptability.
It realizes the rapid and accurate identification of the adaptability of the terminal to be detected and the soft terminal based on the deep message detection information, reduces labor cost waste, improves the efficiency of soft terminal promotion, and provides a reliable basis for soft terminal promotion.
Smart Images

Figure CN115734029B_ABST
Abstract
Description
Technical Field
[0001] The present disclosure relates to the field of data processing technology, and in particular to a terminal adaptability determination method, device, electronic device and storage medium. Background Art
[0002] Soft terminal technology is to realize multiple terminal functions on one terminal by installing application software programs. For example, users only need a smart TV and can obtain the operator's existing set-top box functions and realize home TV intelligent services by installing the application provided by the operator.
[0003] It can be understood that only when the user terminal is compatible with the soft terminal application can the user terminal realize the services provided by the soft terminal. Usually, most types of terminal devices can be adapted to the soft terminal application, but there are still a small number of terminal models or a small number of brand terminals that cannot be adapted. Therefore, in order to avoid business acceptance disputes caused by terminal incompatibility and reduce user experience perception in advance, a mining model can be established to determine whether the user terminal is adapted to the soft terminal application based on the DPI (Deep Packet Inspection) data of the user terminal.
[0004] However, the current DPI analysis technology requires the establishment of different mining models for different terminal brands and models for identification, and the continuous input of large amounts of sample data for full identification. In addition, the mining model needs to be iteratively updated according to the upgrade of the soft terminal application. In actual situations, when sample data is insufficient and the information on known compatible terminals is incomplete, the efficiency of determining whether the user terminal is compatible with the soft terminal application is difficult to guarantee. Summary of the invention
[0005] The present disclosure provides a terminal adaptability judgment system, method, device, electronic device and storage medium to at least solve the problem in the related art that it is difficult to judge whether a user terminal is adapted to an application program based on the DPI information of the user terminal. The technical solution of the present disclosure is as follows:
[0006] According to a first aspect of an embodiment of the present disclosure, a method for determining terminal adaptability is provided, including:
[0007] Obtain deep packet inspection information of the terminal to be detected;
[0008] Inputting the deep packet detection information into a pre-trained feature encoder for feature abstraction processing to obtain a feature vector to be detected, wherein the feature encoder is obtained by comparative learning of sample deep packet detection information;
[0009] Calculating the target similarity between the feature vector to be detected and the target feature vector, wherein the target feature vector is obtained by subjecting the target deep packet detection information to feature abstraction processing and data enhancement processing by the feature encoder, and the target deep packet detection information is deep packet detection information of a user terminal adapted to the soft terminal;
[0010] When the target similarity is less than a preset threshold, it is determined that the terminal to be detected is adapted to the soft terminal.
[0011] Optionally, the step of inputting the deep packet detection information into a pre-trained encoder for feature abstraction processing to obtain a feature vector to be detected of the deep packet detection information includes:
[0012] Performing encoding conversion on the deep packet inspection information to obtain character information;
[0013] Using a text recognition algorithm, the character information is subjected to rule matching and information matching processing to obtain information to be processed;
[0014] The information to be processed is input into a pre-trained encoder for feature abstraction processing to obtain a feature vector to be detected.
[0015] Optionally, before inputting the deep packet inspection information into a pre-trained feature encoder for feature abstraction processing to obtain a feature vector to be detected, the method further includes:
[0016] Obtain sample deep packet inspection information;
[0017] Inputting the sample deep message detection information into a preset encoder for feature abstraction processing to obtain a sample feature vector;
[0018] For each sample feature vector, data enhancement is performed on the sample feature vector to obtain at least one positive sample corresponding to the sample feature vector, a first similarity between the sample feature vector and the positive sample is calculated, and other sample feature vectors except the sample feature vector are used as negative samples to calculate a second similarity between the sample feature vector and the negative sample;
[0019] Determine a loss value according to the first similarity and the second similarity, and adjust parameters in the preset encoder according to the loss value, wherein the loss value is proportional to the first similarity and inversely proportional to the second similarity;
[0020] Inputting the sample deep message detection information into a preset encoder after parameter adjustment to perform feature abstraction processing to obtain a new sample feature vector;
[0021] When the quality evaluation result of the sample feature vector output by the preset encoder meets the preset condition, the preset encoder after parameter adjustment is used as the feature encoder.
[0022] Optionally, for each sample feature vector, data enhancement is performed on the sample feature vector to obtain two positive samples corresponding to the sample feature vector, including:
[0023] For each sample feature vector, the sample feature vector is input into the matrix perturbation model, noise is added to the sample feature vector, and at least one positive sample corresponding to the sample feature vector is obtained;
[0024] The determining a loss value according to the first similarity and the second similarity, and adjusting parameters in the preset encoder according to the loss value, includes:
[0025] A loss value is determined according to the first similarity and the second similarity, and parameters in the preset encoder and parameters in the matrix perturbation model are adjusted according to the loss value.
[0026] Optionally, the quality assessment result includes uniformity and alignment, and when the quality assessment result of the sample feature vector output by the preset encoder meets a preset condition, using the preset encoder after parameter adjustment as the feature encoder includes:
[0027] Calculating uniformity and alignment of sample feature vectors output by the preset encoder;
[0028] When the uniformity and the alignment meet the preset conditions, the preset encoder after parameter adjustment is used as the feature encoder;
[0029] When the uniformity and the alignment do not satisfy the preset conditions, returning to the step of performing data enhancement on each sample feature vector to obtain at least one positive sample corresponding to the sample feature vector.
[0030] Optionally, the calculating the target similarity between the feature vector to be detected and the target feature vector includes:
[0031] Calculating a third similarity between the feature vector to be detected and the target feature vector;
[0032] Calculating a fourth similarity between the feature vector to be detected and a reference feature vector, where the reference feature vector is deep packet inspection information of a user terminal that is not compatible with a soft terminal;
[0033] According to the third similarity and the fourth similarity, a target similarity between the feature vector to be detected and the target feature vector is determined, and the target similarity is proportional to the third similarity and inversely proportional to the fourth similarity.
[0034] According to a second aspect of an embodiment of the present disclosure, a terminal adaptability determination device is provided, including:
[0035] An acquisition module is used to obtain deep packet inspection information of the terminal to be detected;
[0036] A feature abstraction module is used to input the deep packet detection information into a pre-trained feature encoder for feature abstraction processing to obtain a feature vector to be detected, wherein the feature encoder is obtained by comparative learning of sample deep packet detection information;
[0037] A matching module, used to calculate the target similarity between the feature vector to be detected and the target feature vector, wherein the target feature vector is obtained by subjecting the target deep packet detection information to feature abstraction processing and data enhancement processing by the feature encoder, and the target deep packet detection information is deep packet detection information of a user terminal adapted to the soft terminal;
[0038] The determination module is used to determine that the terminal to be detected is adapted to the soft terminal when the target similarity is less than a preset threshold.
[0039] Optionally, the feature abstraction module is specifically used to:
[0040] Performing encoding conversion on the deep packet inspection information to obtain character information;
[0041] Using a text recognition algorithm, the character information is subjected to rule matching and information matching processing to obtain information to be processed;
[0042] The information to be processed is input into a pre-trained encoder for feature abstraction processing to obtain a feature vector to be detected.
[0043] Optionally, the device further comprises a training module, which is used to:
[0044] Obtain sample deep packet inspection information;
[0045] Inputting the sample deep message detection information into a preset encoder for feature abstraction processing to obtain a sample feature vector;
[0046] For each sample feature vector, data enhancement is performed on the sample feature vector to obtain at least one positive sample corresponding to the sample feature vector, a first similarity between the sample feature vector and the positive sample is calculated, and other sample feature vectors except the sample feature vector are used as negative samples to calculate a second similarity between the sample feature vector and the negative sample;
[0047] Determine a loss value according to the first similarity and the second similarity, and adjust parameters in the preset encoder according to the loss value, wherein the loss value is proportional to the first similarity and inversely proportional to the second similarity;
[0048] Inputting the sample deep message detection information into a preset encoder after parameter adjustment to perform feature abstraction processing to obtain a new sample feature vector;
[0049] When the quality evaluation result of the sample feature vector output by the preset encoder meets the preset condition, the preset encoder after parameter adjustment is used as the feature encoder.
[0050] Optionally, the training module is specifically used to:
[0051] For each sample feature vector, the sample feature vector is input into the matrix perturbation model, noise is added to the sample feature vector, and at least one positive sample corresponding to the sample feature vector is obtained;
[0052] The determining a loss value according to the first similarity and the second similarity, and adjusting parameters in the preset encoder according to the loss value, includes:
[0053] A loss value is determined according to the first similarity and the second similarity, and parameters in the preset encoder and parameters in the matrix perturbation model are adjusted according to the loss value.
[0054] Optionally, the quality assessment result includes uniformity and alignment, and the training module is used to:
[0055] Calculating uniformity and alignment of sample feature vectors output by the preset encoder;
[0056] When the uniformity and the alignment meet the preset conditions, the preset encoder after parameter adjustment is used as the feature encoder;
[0057] When the uniformity and the alignment do not satisfy the preset conditions, returning to the step of performing data enhancement on each sample feature vector to obtain at least one positive sample corresponding to the sample feature vector.
[0058] Optionally, the matching module is specifically used to:
[0059] Calculating a third similarity between the feature vector to be detected and the target feature vector;
[0060] Calculating a fourth similarity between the feature vector to be detected and a reference feature vector, where the reference feature vector is deep packet inspection information of a user terminal that is not compatible with a soft terminal;
[0061] According to the third similarity and the fourth similarity, a target similarity between the feature vector to be detected and the target feature vector is determined, and the target similarity is proportional to the third similarity and inversely proportional to the fourth similarity.
[0062] According to a third aspect of an embodiment of the present disclosure, there is provided an electronic device for determining terminal adaptability, including:
[0063] processor;
[0064] a memory for storing instructions executable by the processor;
[0065] The processor is configured to execute the instructions to implement any one of the terminal adaptability determination methods described above.
[0066] According to a fourth aspect of an embodiment of the present disclosure, a computer-readable storage medium is provided. When instructions in the computer-readable storage medium are executed by a processor of a terminal adaptability judgment electronic device, the terminal adaptability judgment electronic device can execute any one of the terminal adaptability judgment methods described above.
[0067] According to a fifth aspect of an embodiment of the present disclosure, a computer program product is provided, including a computer program / instruction, wherein when the computer program / instruction is executed by a processor, the terminal adaptability determination method described in any one of the above items is implemented.
[0068] The technical solution provided by the embodiments of the present disclosure brings at least the following beneficial effects:
[0069] The deep packet detection information of the terminal to be detected is obtained; the deep packet detection information is input into a pre-trained feature encoder for feature abstraction processing to obtain a feature vector to be detected, where the feature encoder is obtained by comparative learning of the sample deep packet detection information; the target similarity between the feature vector to be detected and the target feature vector is calculated, where the target feature vector is obtained by subjecting the target deep packet detection information to feature abstraction processing and data enhancement processing by the feature encoder, where the target deep packet detection information is the deep packet detection information of the user terminal adapted to the soft terminal; when the target similarity is less than a preset threshold, it is determined that the terminal to be detected is adapted to the soft terminal.
[0070] In this way, through the comparative learning framework, self-supervised learning can be performed on the sample deep packet detection information. The learned feature encoder can better express the features of the deep packet detection information and extract higher quality feature vectors to be detected, which are used to classify the deep packet detection information and determine whether the corresponding terminal to be detected is suitable for the soft terminal. This can achieve rapid and accurate judgment of the compatibility of the terminal to be detected and the soft terminal based on the deep packet detection information of the terminal to be detected, providing a reliable basis for the promotion of soft terminals, while reducing the waste of manpower costs and improving the promotion efficiency of soft terminals.
[0071] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS
[0072] The drawings herein are incorporated into and constitute a part of the specification, illustrate embodiments consistent with the present disclosure, and together with the description are used to explain the principles of the present disclosure, and do not constitute improper limitations on the present disclosure.
[0073] Figure 1 The figure is a flow chart of a method for determining terminal adaptability according to an exemplary embodiment.
[0074] Figure 2 The present invention is a data cleaning flow chart according to an exemplary embodiment.
[0075] Figure 3 It is a flowchart showing a specific implementation method according to an exemplary embodiment.
[0076] Figure 4 The present invention is a block diagram of a terminal adaptability determination device according to an exemplary embodiment.
[0077] Figure 5 The invention is a block diagram of an electronic device for determining terminal adaptability according to an exemplary embodiment.
[0078] Figure 6 The invention is a block diagram showing a device for determining the adaptability of a terminal according to an exemplary embodiment. DETAILED DESCRIPTION
[0079] In order to enable ordinary persons in the art to better understand the technical solutions of the present disclosure, the technical solutions in the embodiments of the present disclosure will be clearly and completely described below in conjunction with the accompanying drawings.
[0080] It should be noted that the terms "first", "second", etc. in the specification and claims of the present disclosure and the drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present disclosure described herein can be implemented in an order other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present disclosure. Instead, they are merely examples of devices and methods consistent with some aspects of the present disclosure as detailed in the appended claims.
[0081] Figure 1 is a flow chart of a method for determining terminal adaptability according to an exemplary embodiment. Figure 1 As shown, the terminal adaptability determination method includes:
[0082] In step S11, deep packet inspection information of the terminal to be detected is obtained.
[0083] Soft terminal technology is a technology that implements multiple terminal functions on a terminal by installing soft terminal applications. Usually, soft terminal applications run on the Android platform. Only when the user terminal is compatible with the soft terminal can the user terminal realize the services provided by the soft terminal.
[0084] For example, users only need a smart TV and, by installing the application provided by the operator, they can obtain the operator's existing set-top box functions and realize intelligent home TV services. The technical means cover the compatible support of IPTV (Internet Protocol television, interactive network television) network and Ethernet network of mainstream smart terminals, and the software of a series of network application services including network video and network communication, realizing the Internet integration of soft terminals on mainstream smart terminals, thereby enhancing the application value of IPTV software terminals.
[0085] In this application, the deep message detection information of the terminal to be detected can be used to determine whether the terminal to be detected is compatible with the soft terminal. In this way, compared with the manual door-to-door identification method, the labor cost can be reduced. In addition, in addition to using deep message detection information, other software adaptability terminal log data can also be used for adaptability judgment to achieve the purpose of identifying terminal information with low cost and high efficiency, which is not specifically limited.
[0086] Among them, deep message detection information is to add application protocol identification, datagram content detection and deep decoding of application layer data on top of traditional IP (Internet Protocol) data packet detection technology through deep packet inspection, and collect information formed by user online behavior. By analyzing the deep message detection information, user information, location information, service perception, network quality, user behavior and other data can be obtained. In other words, different terminals to be detected have different deep message detection information. Therefore, the compatibility of the terminal to be detected and the soft terminal can be judged based on the deep message detection information of the terminal to be detected.
[0087] For example, deep packet inspection information includes but is not limited to the following data:
[0088] UA (UserAgent): mainly refers to the information obtained by the browser, which may include application information, terminal model, terminal type, operating system version, browser type, etc. The terminal type can be a mobile phone, computer, tablet, smart TV, box, etc.
[0089] URL (Uniform Resource Locator): mainly contains website address information, and may also contain user account information, terminal serial number, IMSI (International Mobile Subscriber Identity), etc.
[0090] Local data (Cookies): account information, including social software account, email, shopping software account, etc.; mobile phone number, which can be a number from another network or a number from this network; terminal serial code, etc.
[0091] Among them, the deep message detection information of the terminal to be detected can be obtained in real time, that is, whenever a terminal to be detected needs to be judged for compatibility with the soft terminal, the deep message detection information of the terminal to be detected is obtained in real time for analysis; or, it can also be obtained according to a preset period, that is, according to a preset period, the deep message detection information of the terminal to be detected that needs to be judged for compatibility with the soft terminal is collected, and then the collected deep message detection information is analyzed. It can be set according to user needs and business needs, and is not specifically limited.
[0092] In step S12, the deep packet detection information is input into a pre-trained feature encoder for feature abstraction processing to obtain a feature vector to be detected. The feature encoder is obtained by comparative learning of sample deep packet detection information.
[0093] In this step, the deep packet detection information is input into the feature encoder for feature abstraction processing, and the feature encoder extracts features from the deep packet detection information to obtain a feature vector of the deep packet detection information, that is, a feature vector to be detected.
[0094] Among them, the feature encoder is obtained by contrastive learning of the sample deep packet detection information. Contrastive learning is a self-supervised learning method. During the contrastive learning process, there is no need to manually label the sample deep packet detection information. The sample deep packet detection information itself can be directly used as supervision information to learn the feature expression of the sample deep packet detection information. The obtained feature vector to be detected can be used to complete downstream tasks. In this application, the downstream task is to judge the compatibility of the terminal to be detected and the soft terminal.
[0095] In this way, it is possible to reduce labor costs and achieve cost reduction and efficiency improvement for enterprises. Moreover, comparative learning has excellent text semantic similarity calculation capabilities, which is conducive to improving the accuracy of compatibility judgment between the terminal to be detected and the soft terminal, and enhancing user perception.
[0096] In one implementation, before the deep packet detection information is input into a pre-trained feature encoder for feature abstraction processing to obtain a feature vector to be detected, the deep packet detection information needs to be cleaned. Specifically, the step of cleaning the deep packet detection information includes:
[0097] The deep packet detection information is encoded and converted to obtain character information; the character information is matched with rules and information using a text recognition algorithm to obtain information to be processed; the information to be processed is input into a pre-trained encoder for feature abstraction processing to obtain a feature vector to be detected.
[0098] That is to say, after obtaining the deep message detection information of the terminal to be detected, first, the character information can be obtained by encoding and converting the deep message detection information, wherein the encoding conversion can adopt the Base64url (transcoded character set) algorithm; then, a rule library and an information library for text recognition are established, and the text recognition algorithm is used to perform rule matching and information matching processing on the character information, and the character information corresponding to the UA, URL, Cookie and other information is cleaned to obtain the information to be processed. The information to be processed can be used for subsequent compatibility judgment between the terminal to be detected and the soft terminal. At the same time, the rule library and the information library can be continuously updated; then, the information to be processed can be input into the pre-trained encoder for feature abstraction processing to obtain the feature vector to be detected.
[0099] In this way, irrelevant and redundant information in the deep packet inspection information can be filtered, which is beneficial to reducing the amount of calculation in subsequent steps, and can reduce the interference of irrelevant and redundant information on adaptability judgment, thereby improving the accuracy of adaptability judgment.
[0100] In the present application, the feature encoder is obtained by comparative learning of sample deep message detection information, that is, it is necessary to obtain the feature encoder by comparative learning of sample deep message detection information in advance. Specifically, before the deep message detection information is input into the pre-trained feature encoder for feature abstraction processing and the feature vector to be detected is obtained, it also includes:
[0101] Obtain sample deep message detection information; input the sample deep message detection information into a preset encoder for feature abstraction processing to obtain a sample feature vector; for each sample feature vector, perform data enhancement on the sample feature vector to obtain at least one positive sample corresponding to the sample feature vector, calculate a first similarity between the sample feature vector and the positive sample, and use other sample feature vectors except the sample feature vector as negative samples to calculate a second similarity between the sample feature vector and the negative sample; determine a loss value based on the first similarity and the second similarity, and adjust the parameters in the preset encoder based on the loss value, wherein the loss value is proportional to the first similarity and inversely proportional to the second similarity; input the sample deep message detection information into a preset encoder after parameter adjustment for feature abstraction processing to obtain a new sample feature vector; when a quality assessment result of the sample feature vector output by the preset encoder meets a preset condition, use the preset encoder after parameter adjustment as the feature encoder.
[0102] That is to say, for the sample feature vector corresponding to each sample deep message detection information, a positive sample of the sample feature vector is obtained through data enhancement, thereby solving the problem of insufficient sample quantity. Then, comparative learning is performed based on the positive and negative samples of the sample feature vector. It can be understood that high-quality feature expression can better classify the sample deep message detection information. In other words, it can make the sample feature vector more similar to the positive sample, that is, the first similarity is greater, and at the same time make the sample feature vector farther away from the negative sample, that is, the second similarity is smaller. Therefore, the parameters in the preset encoder are adjusted according to the loss value. Finally, after quality evaluation, a feature encoder can be obtained. The feature vector extracted by the feature encoder can better perform feature analysis and extract higher quality feature vectors, thereby realizing the judgment of the adaptability of the terminal to be detected and the soft terminal according to the deep message detection information.
[0103] Among them, for each sample feature vector, data enhancement is performed on the sample feature vector to obtain at least one positive sample corresponding to the sample feature vector, including: for each sample feature vector, the sample feature vector is input into a matrix perturbation model, noise is added to the sample feature vector, and at least one positive sample corresponding to the sample feature vector is obtained.
[0104] That is to say, the NoisyTune method can be introduced to perform data enhancement on the sample feature vector, which adds noise to the original feature vector through matrix-wise perturbing, thereby generating at least one positive sample and solving the problem of insufficient positive samples of the sample feature vector.
[0105] Then, correspondingly, the loss value is determined according to the first similarity and the second similarity, and the parameters in the preset encoder are adjusted according to the loss value, including: determining the loss value according to the first similarity and the second similarity, and adjusting the parameters in the preset encoder and the parameters in the matrix perturbation model according to the loss value. In other words, in the process of contrastive learning, the parameters in the preset encoder and the parameters in the matrix perturbation model can be adjusted at the same time, so that the positive samples obtained subsequently can have a better migration effect, further improving the feature expression effect of the feature encoder.
[0106] In addition, data enhancement for sample feature vectors may also be performed in other ways, such as the dropout algorithm, etc., which are not specifically limited. Dropout is an algorithm for optimizing the training process of deep neural networks. Using the dropout algorithm during the training process can prevent the model from overfitting.
[0107] In one implementation, the quality assessment results include uniformity and alignment, where alignment is defined as the expected distance between feature vectors, which can indicate the closeness between the sample feature vector and the positive sample, and uniformity is defined as the uniformity of the feature vector in the overall feature distribution. It can be understood that high-quality feature expression should keep the sample feature vector close to the positive sample, and the positive and negative samples should be distributed as evenly as possible.
[0108] Then, when the quality assessment result of the sample feature vector output by the preset encoder meets the preset conditions, the preset encoder after parameter adjustment is used as the feature encoder, including: calculating the uniformity and alignment of the sample feature vector output by the preset encoder; when the uniformity and alignment meet the preset conditions, the preset encoder after parameter adjustment is used as the feature encoder; when the uniformity and alignment do not meet the preset conditions, returning to each sample feature vector, performing data enhancement on the sample feature vector, and obtaining the step of at least one positive sample corresponding to the sample feature vector.
[0109] The preset condition may refer to that both uniformity and alignment meet preset thresholds, or at least one of them meets the preset threshold, and there is no specific limitation.
[0110] It can be seen from the above steps that before extracting the feature vector, the deep message detection information needs to be cleaned first. Then, correspondingly, before the sample deep message detection information is input into the preset encoder for feature abstraction processing and the sample feature vector is obtained, the sample deep message detection information also needs to be cleaned. Specifically, it can include: encoding and converting the sample deep message detection information to obtain sample character information; using the text recognition algorithm to perform rule matching and information matching processing on the sample character information to obtain sample information to be processed; inputting the sample information to be processed into the preset encoder for feature abstraction processing to obtain a sample feature vector. It can be understood that the process of data cleaning the sample deep message detection information is the same as the process of data cleaning the deep message detection information, and will not be repeated here.
[0111] In step S13, the target similarity between the feature vector to be detected and the target feature vector is calculated. The target feature vector is obtained by performing feature abstraction processing and data enhancement processing on the target deep packet detection information through a feature encoder. The target deep packet detection information is deep packet detection information of a user terminal adapted to the soft terminal.
[0112] In this step, after obtaining the feature vector to be detected, the target similarity between the feature vector to be detected and the target feature vector can be calculated, wherein the target deep message detection information is the deep message detection information of the user terminal adapted to the soft terminal, obtained through feature abstraction processing and data enhancement processing. Therefore, the target similarity can be used as the similarity between the deep message detection information of the terminal to be detected and the deep message detection information of the user terminal adapted to the soft terminal, that is, the similarity between the terminal to be detected and the user terminal adapted to the soft terminal, and then, it can be used to judge the compatibility of the terminal to be detected and the soft terminal. Among them, the target similarity between the feature vector to be detected and the target feature vector can be calculated by using algorithms such as cosine similarity algorithm, Jaccard similarity coefficient, Pearson correlation coefficient, etc., without specific limitation.
[0113] In one implementation, the similarity between the feature vector to be detected and the target feature vector can be calculated in a manner similar to determining the loss value in the contrastive learning process. Specifically, calculating the target similarity between the feature vector to be detected and the target feature vector includes:
[0114] Calculate the third similarity between the feature vector to be detected and the target feature vector; calculate the fourth similarity between the feature vector to be detected and the reference feature vector, where the reference feature vector is deep packet detection information of a user terminal that is not compatible with the soft terminal; and determine the target similarity between the feature vector to be detected and the target feature vector based on the third similarity and the fourth similarity.
[0115] Among them, the target similarity is proportional to the third similarity and inversely proportional to the fourth similarity. In this way, the similarity between the feature vector to be detected and the target feature vector can be better reflected, and it can be judged whether the feature vector to be detected is closer to the target feature vector or the reference feature vector.
[0116] In step S14, when the target similarity is less than a preset threshold, it is determined that the terminal to be detected is adapted to a soft terminal.
[0117] In this step, if the target similarity is less than the preset threshold, then it can be considered that the feature vector to be detected is relatively close to the target feature vector, that is, it can be considered that the user terminal corresponding to the target feature vector is relatively close to the terminal to be detected. Therefore, the adaptability of the user terminal corresponding to the target feature vector and the soft terminal can be used as the adaptability of the terminal to be detected and the soft terminal, that is, it is determined that the terminal to be detected is adapted to the soft terminal.
[0118] On the contrary, if the target similarity is not less than the preset threshold, then it can be considered that the feature vector to be detected is not similar enough to the target feature vector, that is, it can be considered that the user terminal corresponding to the target feature vector is not close to the terminal to be detected. Therefore, it can be determined that the terminal to be detected is not suitable for the soft terminal.
[0119] From the above, it can be seen that the technical solution provided by the embodiment of the present disclosure performs self-supervised learning on the sample deep packet detection information through a comparative learning framework. The learned feature encoder can better express the features of the deep packet detection information and extract higher quality feature vectors to be detected, which are used to classify the deep packet detection information and determine whether the corresponding terminal to be detected is suitable for the soft terminal. This can achieve rapid and accurate judgment of the compatibility of the terminal to be detected and the soft terminal based on the deep packet detection information of the terminal to be detected, providing a reliable basis for the promotion of soft terminals, while reducing the waste of manpower costs and improving the promotion efficiency of soft terminals.
[0120] The method in the embodiment of the present invention is described below with a specific embodiment. The flowchart of the specific embodiment is as follows: Figure 2 shown.
[0121] First, input data is obtained, including deep packet detection information of the terminal to be detected and sample deep packet detection information.
[0122] Then, the input data is cleaned. The process of this step is as follows: Figure 2 As shown, the method includes two parts: model training and model application, which respectively perform data cleaning on the sample deep packet detection information and the deep packet detection information of the terminal to be detected.
[0123] During the model training process, sample deep message detection information can be obtained, and then the sample deep message detection information is encoded and converted using the Base64url algorithm, processed using the text content recognition algorithm, and a rule base and information base for text recognition are established. The newly generated sample deep message detection information is trained at regular intervals to update the rule base and information base, and to evaluate the effect of the text content recognition algorithm and iteratively optimize it. Then, the sample deep message detection information is marked using the user data of the soft terminal platform to obtain the sample information to be processed, which is represented as the sample data set P i :
[0124] P i ={A,L,C,S}
[0125] Among them, A is the UA data in the sample deep message detection information, which is mainly the information obtained by the browser, and may hide user account, terminal model, terminal type, tablet, smart TV, box, operating system version, browser type and other information; L is the URL information in the sample deep message detection information, which is mainly the website information; C is the Cookie data in the sample deep message detection information, which is mainly the user cache information; S is the soft terminal platform user registration information, which is used to identify whether the sample deep message detection information is compatible with the soft terminal.
[0126] In addition, during the model application process, the newly acquired deep packet detection information of the terminal to be detected can be input every day, and the deep packet detection information of the terminal to be detected can be similarly encoded, converted, rule matched and information matched, and the UA, URL and Cookie can be cleaned to obtain the information to be detected.
[0127] Then, the sample data set is subjected to feature abstraction processing through encoder encoding. For example, the preset encoder may adopt a transformer encoder to perform semantic abstraction (sentence embedding) on the data in the sample data set to obtain a word vector.
[0128] Specifically, the data in the cleaned sample data set can be input into the encoder, and the self-attention mechanism of the transformer can be used to automatically select information that is strongly related to the current scene for abstraction, and the sample feature vector X corresponding to each data sample in the sample data set can be obtained. i , expressed as:
[0129] X i =TR_encoder(P i )
[0130] Then, we use NoisyTune to perform data enhancement. By adding matrix-level perturbations to the pre-trained model parameters before fine-tuning, we add noise to the sample data set to obtain positive samples with better transferability, solve the problem of insufficient positive samples, and improve the learning effect of contrastive learning. For example, the obtained positive samples can be expressed as follows:
[0131]
[0132] in, Indicates from arrive The noise is uniformly distributed in the range; λ represents the noise intensity hyperparameter; std represents the standard deviation. In this application, the noise intensity hyperparameter can be set to 0.1-0.25. If the noise intensity hyperparameter is less than 0.1 or greater than 0.25, the positive sample effect is poor.
[0133] Then, the positive sample set corresponding to the sample feature vector can be expressed as:
[0134]
[0135] Where z is the NoisyTune noise superposition.
[0136] Then, the semantic similarity is calculated by contrastive learning. Perform comparative learning and make similarity judgment based on the objective function, where the objective function can be expressed as:
[0137]
[0138] τ is a hyperparameter that controls the normalized distribution of loss values. The larger the τ, the smoother the distribution. The smaller the τ, the greater the gap between positive and negative samples, which can speed up convergence. N is the amount of data of the sample feature vector for each iteration. The sim function represents the cosine similarity calculation. is any sample feature vector, for The positive sample is Obtained through NoisyTune, for Negative samples, that is, except In this application, the adjustment hyperparameter τ can be set to 0-0.5, and N can be set to 500.
[0139] Then, alignment and uniformity are used for learning evaluation. If the sample feature vector is highly close to its positive sample and the distribution is more uniform, that is, the smaller the alignment and uniformity evaluation data is, the higher the quality of the positive sample is, that is, the encoder's feature abstraction processing effect and data enhancement effect are better.
[0140] Among them, the alignment can be expressed as:
[0141]
[0142] is a pair of sample feature vector and any sample data in its positive sample set, p pos Represents a positional parameter.
[0143] Uniformity can be expressed as:
[0144]
[0145] Among them, p data Represents data distribution.
[0146] Then, the adaptability of the terminal to be detected is judged. In this step, the deep message detection information P′ of the terminal to be detected can be i After cleaning, the information to be detected is input into the trained encoder for feature abstraction processing to obtain the feature vector to be detected. Then, the objective function is used to confirm whether the feature vector to be detected is a positive sample of a target feature vector. If so, the feature vector to be detected is similar to the target feature vector, that is, the adaptability of the user terminal and the soft terminal corresponding to the terminal to be detected and the target feature vector is the same.
[0147] Therefore, the sample data P corresponding to the target feature vector can be i The S is migrated to the deep packet inspection information P′ of the terminal to be detected i N, and then the sample data P corresponding to the target feature vector i The user account information in the deep packet inspection information P′ of the terminal to be detected i The final output data can be expressed as:
[0148] P′ i ={De,N}
[0149] Among them, De represents the user account information, and N is the compatibility result between the terminal to be detected and the soft terminal.
[0150] Figure 4 The present invention is a block diagram of a terminal adaptability determination device according to an exemplary embodiment, including:
[0151] The acquisition module 201 is used to acquire the deep packet inspection information of the terminal to be detected;
[0152] A feature abstraction module 202 is used to input the deep packet detection information into a pre-trained feature encoder for feature abstraction processing to obtain a feature vector to be detected, wherein the feature encoder is obtained by comparative learning of sample deep packet detection information;
[0153] A matching module 203 is used to calculate the target similarity between the feature vector to be detected and the target feature vector, wherein the target feature vector is obtained by subjecting the target deep packet detection information to feature abstraction processing and data enhancement processing by the feature encoder, and the target deep packet detection information is deep packet detection information of a user terminal adapted to the soft terminal;
[0154] The determination module 204 is configured to determine that the terminal to be detected is adapted to the soft terminal when the target similarity is less than a preset threshold.
[0155] Optionally, the feature abstraction module 202 is specifically used to:
[0156] Performing encoding conversion on the deep packet inspection information to obtain character information;
[0157] Using a text recognition algorithm, the character information is subjected to rule matching and information matching processing to obtain information to be processed;
[0158] The information to be processed is input into a pre-trained encoder for feature abstraction processing to obtain a feature vector to be detected.
[0159] Optionally, the device further comprises a training module, which is used to:
[0160] Obtain sample deep packet inspection information;
[0161] Inputting the sample deep message detection information into a preset encoder for feature abstraction processing to obtain a sample feature vector;
[0162] For each sample feature vector, data enhancement is performed on the sample feature vector to obtain at least one positive sample corresponding to the sample feature vector, a first similarity between the sample feature vector and the positive sample is calculated, and other sample feature vectors except the sample feature vector are used as negative samples to calculate a second similarity between the sample feature vector and the negative sample;
[0163] Determine a loss value according to the first similarity and the second similarity, and adjust parameters in the preset encoder according to the loss value, wherein the loss value is proportional to the first similarity and inversely proportional to the second similarity;
[0164] Inputting the sample deep message detection information into a preset encoder after parameter adjustment to perform feature abstraction processing to obtain a new sample feature vector;
[0165] When the quality evaluation result of the sample feature vector output by the preset encoder meets the preset condition, the preset encoder after parameter adjustment is used as the feature encoder.
[0166] Optionally, the training module is specifically used to:
[0167] For each sample feature vector, the sample feature vector is input into the matrix perturbation model, noise is added to the sample feature vector, and at least one positive sample corresponding to the sample feature vector is obtained;
[0168] The determining a loss value according to the first similarity and the second similarity, and adjusting parameters in the preset encoder according to the loss value, includes:
[0169] A loss value is determined according to the first similarity and the second similarity, and parameters in the preset encoder and parameters in the matrix perturbation model are adjusted according to the loss value.
[0170] Optionally, the quality assessment result includes uniformity and alignment, and the training module is used to:
[0171] Calculating uniformity and alignment of sample feature vectors output by the preset encoder;
[0172] When the uniformity and the alignment meet the preset conditions, the preset encoder after parameter adjustment is used as the feature encoder;
[0173] When the uniformity and the alignment do not satisfy the preset conditions, returning to the step of performing data enhancement on each sample feature vector to obtain at least one positive sample corresponding to the sample feature vector.
[0174] Optionally, the matching module 203 is specifically configured to:
[0175] Calculating a third similarity between the feature vector to be detected and the target feature vector;
[0176] Calculating a fourth similarity between the feature vector to be detected and a reference feature vector, where the reference feature vector is deep packet inspection information of a user terminal that is not compatible with a soft terminal;
[0177] According to the third similarity and the fourth similarity, a target similarity between the feature vector to be detected and the target feature vector is determined, and the target similarity is proportional to the third similarity and inversely proportional to the fourth similarity.
[0178] From the above, it can be seen that the technical solution provided by the embodiment of the present disclosure performs self-supervised learning on the sample deep packet detection information through a comparative learning framework. The learned feature encoder can better express the features of the deep packet detection information and extract higher quality feature vectors to be detected, which are used to classify the deep packet detection information and determine whether the corresponding terminal to be detected is suitable for the soft terminal. This can achieve rapid and accurate judgment of the compatibility of the terminal to be detected and the soft terminal based on the deep packet detection information of the terminal to be detected, providing a reliable basis for the promotion of soft terminals, while reducing the waste of manpower costs and improving the promotion efficiency of soft terminals.
[0179] Figure 5 The invention is a block diagram of an electronic device for determining terminal adaptability according to an exemplary embodiment.
[0180] In an exemplary embodiment, a computer-readable storage medium including instructions is also provided, such as a memory including instructions, and the instructions can be executed by a processor of an electronic device to complete the method. Optionally, the computer-readable storage medium can be a ROM, a random access memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, an optical data storage device, etc.
[0181] In an exemplary embodiment, a computer program product is also provided. When the computer program product is executed on a computer, the computer implements the terminal adaptability determination method.
[0182] From the above, it can be seen that the technical solution provided by the embodiment of the present disclosure performs self-supervised learning on the sample deep packet detection information through a comparative learning framework. The learned feature encoder can better express the features of the deep packet detection information and extract higher quality feature vectors to be detected, which are used to classify the deep packet detection information and determine whether the corresponding terminal to be detected is suitable for the soft terminal. This can achieve rapid and accurate judgment of the compatibility of the terminal to be detected and the soft terminal based on the deep packet detection information of the terminal to be detected, providing a reliable basis for the promotion of soft terminals, while reducing the waste of manpower costs and improving the promotion efficiency of soft terminals.
[0183] Figure 6 It is a block diagram of a device 800 for determining terminal adaptability according to an exemplary embodiment.
[0184] For example, apparatus 800 may be a mobile phone, a computer, a digital broadcast electronic device, a messaging device, a game console, a tablet device, a medical device, a fitness device, a personal digital assistant, etc.
[0185] Reference Figure 6 , the device 800 may include one or more of the following components: a processing component 802 , a memory 804 , a power component 806 , a multimedia component 808 , an audio component 810 , an input / output (I / O) interface 812 , a sensor component 814 , and a communication component 816 .
[0186] The processing component 802 generally controls the overall operation of the device 800, such as operations associated with display, phone calls, data communications, camera operations, and recording operations. The processing component 802 may include one or more processors 820 to execute instructions to complete all or part of the steps of the described method. In addition, the processing component 802 may include one or more modules to facilitate the interaction between the processing component 802 and other components. For example, the processing component 802 may include a multimedia module to facilitate the interaction between the multimedia component 808 and the processing component 802.
[0187] The memory 804 is configured to store various types of data to support operations on the device 800. Examples of such data include instructions for any application or method operating on the device 800, contact data, phone book data, messages, pictures, videos, etc. The memory 804 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.
[0188] The power supply component 807 provides power to the various components of the device 800. The power supply component 807 may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power for the device 800.
[0189] The multimedia component 808 includes a screen that provides an output interface between the device 800 and the account. In some embodiments, the screen may include a liquid crystal display (LCD) and a touch panel (TP). If the screen includes a touch panel, the screen may be implemented as a touch screen to receive input signals from the account. The touch panel includes one or more touch sensors to sense touch, slide, and gestures on the touch panel. The touch sensor may not only sense the boundaries of the touch or slide action, but also detect the duration and pressure associated with the touch or slide operation. In some embodiments, the multimedia component 808 includes a front camera and / or a rear camera. When the device 800 is in an operating mode, such as a shooting mode or a video mode, the front camera and / or the rear camera may receive external multimedia data. Each front camera and rear camera may be a fixed optical lens system or have a focal length and optical zoom capability.
[0190] The audio component 810 is configured to output and / or input audio signals. For example, the audio component 810 includes a microphone (MIC), and when the device 800 is in an operating mode, such as a call mode, a recording mode, and a speech recognition mode, the microphone is configured to receive an external audio signal. The received audio signal can be further stored in the memory 804 or sent via the communication component 816. In some embodiments, the audio component 810 also includes a speaker for outputting audio signals.
[0191] I / O interface 812 provides an interface between processing component 802 and peripheral interface modules, which may be keyboards, click wheels, buttons, etc. These buttons may include but are not limited to: home button, volume button, start button, and lock button.
[0192] The sensor assembly 814 includes one or more sensors for providing various aspects of the status assessment of the device 800. For example, the sensor assembly 814 can detect the open / closed state of the device 800, the relative positioning of components, such as the display and keypad of the device 800, and the sensor assembly 814 can also detect the position change of the device 800 or a component of the device 800, the presence or absence of contact with the device 800, the orientation or acceleration / deceleration of the device 800, and the temperature change of the device 800. The sensor assembly 814 may include a proximity sensor configured to detect the presence of a nearby object without any physical contact. The sensor assembly 814 may also include an optical sensor, such as a CMOS or CCD image sensor, for use in imaging applications. In some embodiments, the sensor assembly 814 may also include an accelerometer, a gyroscope sensor, a magnetic sensor, a pressure sensor, or a temperature sensor.
[0193] The communication component 816 is configured to facilitate wired or wireless communication between the device 800 and other devices. The device 800 can access a wireless network based on a communication standard, such as WiFi, a carrier network (such as 2G, 3G, 4G or 5G), or a combination thereof. In an exemplary embodiment, the communication component 816 receives a broadcast signal or broadcast-related information from an external broadcast management system via a broadcast channel. In an exemplary embodiment, the communication component 816 also includes a near field communication (NFC) module to facilitate short-range communication. For example, the NFC module can be implemented based on radio frequency identification (RFID) technology, infrared data association (IrDA) technology, ultra-wideband (UWB) technology, Bluetooth (BT) technology and other technologies.
[0194] In an exemplary embodiment, the apparatus 800 may be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors or other electronic components to execute the methods described in the first and second aspects.
[0195] In an exemplary embodiment, a non-transitory computer-readable storage medium including instructions is also provided, such as a memory 804 including instructions, which can be executed by a processor 820 of the device 800 to complete the method. Alternatively, for example, the storage medium can be a non-transitory computer-readable storage medium, for example, the non-transitory computer-readable storage medium can be a ROM, a random access memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, an optical data storage device, etc.
[0196] In an exemplary embodiment, a computer program product including instructions is also provided. When the computer program product is run on a computer, the computer is enabled to execute the terminal adaptability determination method described in any one of the embodiments.
[0197] From the above, it can be seen that the technical solution provided by the embodiment of the present disclosure performs self-supervised learning on the sample deep packet detection information through a comparative learning framework. The learned feature encoder can better express the features of the deep packet detection information and extract higher quality feature vectors to be detected, which are used to classify the deep packet detection information and determine whether the corresponding terminal to be detected is suitable for the soft terminal. This can achieve rapid and accurate judgment of the compatibility of the terminal to be detected and the soft terminal based on the deep packet detection information of the terminal to be detected, providing a reliable basis for the promotion of soft terminals, while reducing the waste of manpower costs and improving the promotion efficiency of soft terminals.
[0198] Those skilled in the art will readily appreciate other embodiments of the present disclosure after considering the specification and practicing the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of the present disclosure that follow the general principles of the present disclosure and include common knowledge or customary techniques in the art that are not disclosed in the present disclosure. The specification and examples are intended to be exemplary only, and the true scope and spirit of the present disclosure are indicated by the following claims.
[0199] It should be understood that the present disclosure is not limited to the exact structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.
Claims
1. A terminal adaptability determination method, characterized in that: include: Obtain deep packet inspection information of the terminal to be detected; Inputting the deep packet detection information into a pre-trained feature encoder for feature abstraction processing to obtain a feature vector to be detected, wherein the feature encoder is obtained by comparative learning of sample deep packet detection information; Calculating the target similarity between the feature vector to be detected and the target feature vector, wherein the target feature vector is obtained by subjecting the target deep packet detection information to feature abstraction processing and data enhancement processing by the feature encoder, and the target deep packet detection information is deep packet detection information of a user terminal adapted to the soft terminal; When the target similarity is less than a preset threshold, it is determined that the terminal to be detected is adapted to the soft terminal.
2. The terminal adaptability determination method according to claim 1, characterized in that: The step of inputting the deep packet detection information into a pre-trained encoder for feature abstraction processing to obtain a feature vector to be detected of the deep packet detection information includes: Performing encoding conversion on the deep packet inspection information to obtain character information; Using a text recognition algorithm, the character information is subjected to rule matching and information matching processing to obtain information to be processed; The information to be processed is input into a pre-trained encoder for feature abstraction processing to obtain a feature vector to be detected.
3. The terminal adaptability determination method according to claim 1, characterized in that: Before inputting the deep packet detection information into a pre-trained feature encoder for feature abstraction processing to obtain a feature vector to be detected, the method further includes: Obtain sample deep packet inspection information; Inputting the sample deep message detection information into a preset encoder for feature abstraction processing to obtain a sample feature vector; For each sample feature vector, data enhancement is performed on the sample feature vector to obtain at least one positive sample corresponding to the sample feature vector, a first similarity between the sample feature vector and the positive sample is calculated, and other sample feature vectors except the sample feature vector are used as negative samples to calculate a second similarity between the sample feature vector and the negative sample; Determine a loss value according to the first similarity and the second similarity, and adjust parameters in the preset encoder according to the loss value, wherein the loss value is proportional to the first similarity and inversely proportional to the second similarity; Inputting the sample deep message detection information into a preset encoder after parameter adjustment to perform feature abstraction processing to obtain a new sample feature vector; When the quality evaluation result of the sample feature vector output by the preset encoder meets the preset condition, the preset encoder after parameter adjustment is used as the feature encoder.
4. The terminal adaptability determination method according to claim 3, characterized in that: The step of performing data enhancement on each sample feature vector to obtain at least one positive sample corresponding to the sample feature vector includes: For each sample feature vector, the sample feature vector is input into the matrix perturbation model, noise is added to the sample feature vector, and at least one positive sample corresponding to the sample feature vector is obtained; The determining a loss value according to the first similarity and the second similarity, and adjusting parameters in the preset encoder according to the loss value, includes: A loss value is determined according to the first similarity and the second similarity, and parameters in the preset encoder and parameters in the matrix perturbation model are adjusted according to the loss value.
5. The terminal adaptability determination method according to claim 3, characterized in that: The quality evaluation result includes uniformity and alignment, and when the quality evaluation result of the sample feature vector output by the preset encoder meets a preset condition, the preset encoder after parameter adjustment is used as the feature encoder, including: Calculating uniformity and alignment of sample feature vectors output by the preset encoder; When the uniformity and the alignment meet the preset conditions, the preset encoder after parameter adjustment is used as the feature encoder; When the uniformity and the alignment do not satisfy the preset conditions, returning to the step of performing data enhancement on each sample feature vector to obtain at least one positive sample corresponding to the sample feature vector.
6. The terminal adaptability determination method according to claim 1, characterized in that: The calculating the target similarity between the feature vector to be detected and the target feature vector comprises: Calculating a third similarity between the feature vector to be detected and the target feature vector; Calculating a fourth similarity between the feature vector to be detected and a reference feature vector, where the reference feature vector is deep packet inspection information of a user terminal that is not compatible with a soft terminal; According to the third similarity and the fourth similarity, a target similarity between the feature vector to be detected and the target feature vector is determined, and the target similarity is proportional to the third similarity and inversely proportional to the fourth similarity.
7. A terminal adaptability determination device, characterized in that: include: An acquisition module is used to obtain deep packet inspection information of the terminal to be detected; A feature abstraction module is used to input the deep packet detection information into a pre-trained feature encoder for feature abstraction processing to obtain a feature vector to be detected, wherein the feature encoder is obtained by comparative learning of sample deep packet detection information; A matching module, used to calculate the target similarity between the feature vector to be detected and the target feature vector, wherein the target feature vector is obtained by subjecting the target deep packet detection information to feature abstraction processing and data enhancement processing by the feature encoder, and the target deep packet detection information is deep packet detection information of a user terminal adapted to the soft terminal; The determination module is used to determine that the terminal to be detected is adapted to the soft terminal when the target similarity is less than a preset threshold.
8. The terminal adaptability determination device according to claim 7, characterized in that: The feature abstraction module is specifically used for: Performing encoding conversion on the deep packet inspection information to obtain character information; Using a text recognition algorithm, the character information is subjected to rule matching and information matching processing to obtain information to be processed; The information to be processed is input into a pre-trained encoder for feature abstraction processing to obtain a feature vector to be detected.
9. The terminal adaptability determination device according to claim 7, characterized in that: The device also includes a training module for: Obtain sample deep packet inspection information; Inputting the sample deep message detection information into a preset encoder for feature abstraction processing to obtain a sample feature vector; For each sample feature vector, data enhancement is performed on the sample feature vector to obtain at least one positive sample corresponding to the sample feature vector, a first similarity between the sample feature vector and the positive sample is calculated, and other sample feature vectors except the sample feature vector are used as negative samples to calculate a second similarity between the sample feature vector and the negative sample; Determine a loss value according to the first similarity and the second similarity, and adjust parameters in the preset encoder according to the loss value, wherein the loss value is proportional to the first similarity and inversely proportional to the second similarity; Inputting the sample deep message detection information into a preset encoder after parameter adjustment to perform feature abstraction processing to obtain a new sample feature vector; When the quality evaluation result of the sample feature vector output by the preset encoder meets the preset condition, the preset encoder after parameter adjustment is used as the feature encoder.
10. The terminal adaptability determination device according to claim 9, characterized in that: The training module is specifically used for: For each sample feature vector, the sample feature vector is input into the matrix perturbation model, noise is added to the sample feature vector, and at least one positive sample corresponding to the sample feature vector is obtained; The determining a loss value according to the first similarity and the second similarity, and adjusting parameters in the preset encoder according to the loss value, includes: A loss value is determined according to the first similarity and the second similarity, and parameters in the preset encoder and parameters in the matrix perturbation model are adjusted according to the loss value.
11. The terminal adaptability determination device according to claim 9, characterized in that: The quality assessment results include uniformity and alignment, and the training module is used to: Calculating uniformity and alignment of sample feature vectors output by the preset encoder; When the uniformity and the alignment meet the preset conditions, the preset encoder after parameter adjustment is used as the feature encoder; When the uniformity and the alignment do not satisfy the preset conditions, returning to the step of performing data enhancement on each sample feature vector to obtain at least one positive sample corresponding to the sample feature vector.
12. The terminal adaptability determination device according to claim 7, characterized in that: The matching module is specifically used for: Calculating a third similarity between the feature vector to be detected and the target feature vector; Calculating a fourth similarity between the feature vector to be detected and a reference feature vector, where the reference feature vector is deep packet inspection information of a user terminal that is not compatible with a soft terminal; According to the third similarity and the fourth similarity, a target similarity between the feature vector to be detected and the target feature vector is determined, and the target similarity is proportional to the third similarity and inversely proportional to the fourth similarity.
13. An electronic device, characterized in that: include: processor; a memory for storing instructions executable by the processor; The processor is configured to execute the instruction to implement the terminal adaptability determination method according to any one of claims 1 to 6.
14. A computer-readable storage medium, characterized in that: When the instructions in the computer-readable storage medium are executed by a processor of the terminal adaptability determination electronic device, the terminal adaptability determination electronic device is enabled to execute the terminal adaptability determination method according to any one of claims 1 to 6.
15. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the terminal adaptability determination method according to any one of claims 1 to 6 is implemented.
