To improve the efficiency of recognition while determining the emotional tendencies of goods evaluation accurately, this paper proposed a text classification approach based on Matrix Projection (MP) and Normalized Vector (NLV) to realize sentiment analysis for goods evaluation. Firstly, this approach extracted feature words of goods evaluation by utilizing matrix projection, and then computed the average Feature Frequency (FF) of feature words in each category, and obtained normalized vector through normalized processing to feature frequency of each category by using Normalized Function (NLF). Finally, it predicted the sentiment tendency by comparing similarity between feature vector of goods evaluation and normalized vector of each category. Compared with the k-Nearest Neighbor (kNN), Naive Bayesian (NB) and Support Vector Machine (SVM) algorithm, the experimental results show that the proposed approach has higher prediction accuracy and speed of classification. Especially compared with the kNN the approach has obvious advantages, its macro average F1 value is more than 12% higher than the kNN and classification time is reduced by 11/12〖BP(〗reduce to或reduce by〖BP)〗. Compared with the SVM its speed is greatly improved.

For malware analysis and cipher application security validating, identification and filtration of cryptographic function from binary code has great significance. The memory operation and basic block loop characters were analyzed from cryptographic functions. According to the theory of binary data's information entropy, the characteristic of high-entropy of cryptographic algorithms was verified, a cryptographic functions filtration model was constructed based on dynamic loop entropy, and the hybrid (dynamic and static) method was adopted to reconstruct dynamic memory data in basic block loop. The experimental result shows that the filtration model has reliability and veracity.

A recommendation-based centralized trust model was proposed to deal with three kinds of common attacks including self-promoting attack, slandering attack and whitewashing attack. The model calculated the similarity of consumption custom between buyers and the timeliness of the buyers' evaluation opinions to determine the recommendation trust value, and obtained the direct trust value on the basis of the buyer's own trading experience. According to the direct and recommendation trust value, composite trust value could be calculated. Simulations were given by injecting the three kinds of attacks into real transaction data. The results prove that, compared with the existing reputation management models, the proposed model can effectively prevent attacks, and provides more accurate reputation information of sellers and goods to consumers.