Textual information processing

The previous chapter discussed processing image or video data and interpreting them to develop new input (gaze-controlled interface, gesture recognition system) or output (augmented or mixed reality) interactive systems. This chapter will discuss processing textual data and interpreting it for developing intelligent interactive systems. In the late 1960s and early 1970s, expert systems were one of the most popular AI systems predicting chemical composition (Dendral; Lindsay et al. 1993 ) or even prescription medicine (Mycin; Shortliffe & Buchanan 1975 ) from textual responses from users. With the emergence of the World Wide Web in the 1990s, text and hyperlink processing gained momentum not only in terms of giving access to information but also creating information for the audience. This creation part took a new turn with the emergence of social networking and microblogging sites. The development of machine learning models often worked vis-à-vis with image and textual data analysis – for example, LSTM (long short-term memory) and transformer models were originally developed for textual content analysis but later used for computer vision systems as well in the form of vision transformers. Presently, a major research area of intelligent interaction design is centred on using large language models (LLMs) and combining them with computer vision models to understand users’ instructions and context. This chapter will start from basic content analysis using stemming, stop words, parts of speech tagging and the concept of a vector space model. Next, we will discuss expert systems in the context of online recommendation systems and chatbots. Finally, we will seek to understand the basic concept and structure of LLMs.