Topic modeling aims to discover hidden topics or themes across documents that capture semantic information beyond individual words. It aims to address a key challenge in building a machine learning algorithm that learns from text data by going beyond the lexical level of what has been written to the semantic level of what was intended. The resulting topics can be used to annotate documents based on their association with various topics.

In other words, topic modeling aims to automatically summarize large collections of documents to facilitate organization and management, as well as search and recommendations. At the same time, it can enable the understanding of documents to the extent that humans can interpret the descriptions of topics.

Topic models aim to address the curse of dimensionality that can plague the bag-of-words model. The document representation based on high-dimensional sparse vectors can make similarity measures noisy, leading to inaccurate distance measurement and overfitting of text classification models.

Moreover, the bag of words model ignores word order and loses context as well as semantic information because it is not able to capture synonymy (several words have the same meaning) and polysemy (one word has several meanings). As a result, document retrieval or similarity search may miss the point when the documents are not indexed by the terms used to search or compare.

These shortcoming prompt this question: how do we model and learn meaning topics that facilitate a more productive interaction with text data?