Exploration of BERT distillation in garbage public opinion identification

This paper attempts BERT distillation technology in garbage public opinion recognition task to improve the performance of textCNN classifier, and uses its small and fast advantages to successfully land.

Practice of Active Learning Method: Make the Model "Active"

This paper shares an active learning method that allows the model to participate in the selection of "learning samples", and distinguishes all samples in the sample pool according to different strategies to improve the model performance.

Random talk on visual object tracking: from principle to application

This paper aims to make a comprehensive, detailed and timely overview of tracking, and study the definition, application, architecture, algorithm and evaluation of tracking.

How does Magic Horse Search improve the timeliness of search?

This article shares the practice and exploration of Whatsmart Search on the timeliness of search ranking, starting from the optimization of basic features, optimizing the sorting and recall model through labeled data, and the recall system and collection system of timeliness sorting.

What are the progress and trends of natural language processing technology?

This article shares the summary of NLP technologies, platforms and products in the past year by experts in various fields of the Natural Language Processing Technology Group of DAMO Academy, and looks forward to the trend of the new year, depicting the new decade of Ali's natural language processing technology.