Network information analysis method based on privacy grouping and emotion recognition
A technology of emotion recognition and network information, applied in biological neural network models, digital data information retrieval, special data processing applications, etc., can solve problems such as loss, poor sensitivity of similarity, and unconsidered differences
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
Method used
Image
Examples
Embodiment 1
[0118] combine figure 1 , taking CSI data image processing release as an example.
[0119] First, the original CSI data of the wireless signal is collected from the receiver, followed by data preprocessing, feature extraction and recognition. In the stage of collecting big data, arrange the transmitter and receiver at any position, and the user performs action collection between the transmitter and the receiver. The transmitter uses a wireless router, the receiver uses a network card and is connected to an external computer, and the sampling frequency is set to 1000Hz. Use the receiver to obtain channel state information. If the obtained local outlier factor LOF value is greater than 1, it will be regarded as outlier removal. Then perform feature extraction. If information on 3 links is received, each link contains 30 subcarriers, and a 30*3 matrix is obtained each time. The sampling rate of the signal is set to 1000 data packets per second, and the contents of the data p...
Embodiment 2
[0121] Described Encoder-Decoder model comprises encoding model and decoding model, and encoding part is the data that length is n, and hidden layer output matrix is
[0122] H=[h 1 ,h 2 ,..., h n ]
[0123] Generate data as v s :
[0124]
[0125] The decoding model is composed of aspect attention modules, and the number of modules N is the same as the total number of aspects in the data set, that is, one aspect attention module corresponds to a specific aspect; when the input data contains multiple aspects, the encoded output H is They are respectively sent to the corresponding aspect attention modules. In each aspect attention module, there is a corresponding aspect information, that is, the aspect vector v ai , aspect vector v ai Splicing with each hidden state in the input matrix H, then performing attention calculations, and finally obtaining data about specific aspects through weighted averaging:
[0126] e tif =tanh(W aif [h t ,v aif ]+b aif )
[0127] ...
Embodiment 3
[0131] Similar to Examples 1 and 2, the difference is that the emotional polarity classification includes: the v s Send it to the full connection layer, and output the probability of the sample to be classified on each emotion category through the softmax function, and get the emotion polarity corresponding to different aspects:
[0132]
[0133] where P tif is the weight matrix of the fully connected layer, b pif is the bias item of the full link layer, and C is the number of categories.
[0134] In order to verify the results of the present invention, the F1 value index is selected to carry out comparative experiments. The F1 value is the harmonic mean of the precision rate and the recall rate, namely:
[0135]
[0136] The present invention selects 6 classic models for three-category comparison, and the verification results are shown in Table 1. It can be seen from the comparison results that the present invention improves the F2 values of negative classificatio...
PUM
Login to View More Abstract
Description
Claims
Application Information
Login to View More 


