Feature Selection and Classification of Electronic Product Fault Short Text by Integrating TF-IDF and Wor D2vec

Driven by Internet technology and social media, news text data shows explosive growth, and traditional manual classification has been difficult to meet the needs of efficient retrieval. In response to the problem of news short text classification, this study reviews existing technologies through literature review and proposes two improved models, standardized TF-IDF and 1+log (TF) - IDF. Nonlinear scaling is used to optimize word frequency weight allocation, and a voting mechanism is combined to fuse three classifiers: support vector machine, random forest, and gradient boosting tree, to construct a more robust ensemble framework. The experimental results show that the improved model improves the F1 score on the news headline dataset by 4.2% -5.7% compared to traditional methods, especially in the recognition of low frequency professional terms. The voting fusion strategy further improves the classification accuracy by 3.8%. However, research is still limited by domain dependencies in datasets, IDF calculations ignoring differences in word distribution between categories, and insufficient capture of contextual semantics. In the future, we can explore directions such as cross domain data expansion, introduction of BERT pre training models, dynamic weight adjustment mechanisms, and knowledge graph integration to optimize the generalization and semantic parsing capabilities of news short text classification.