Holes in the Outline: Subject-dependent Abstract Quality and its Implications for Scientific Literature Search

Scientific literature search engines typically index abstracts instead of the full-text of publications. The expectation is that the abstract provides a comprehensive summary of the article, enumerating key points for the reader to assess whether their information needs could be satisfied by reading the full-text. Furthermore, from a practical standpoint, obtaining the full-text is more complicated due to licensing issues, in the case of commercial publishers, and resource limitations of public repositories and pre-print servers. In this article, we use topic modelling to represent content in abstracts and full-text articles. Using Computer Science as a case study, we demonstrate that how well the abstract summarises the full-text is subfield-dependent. Indeed, we show that abstract representativeness has a direct impact on retrieval performance, with poorer abstracts leading to degraded performance. Finally, we present evidence that how well an abstract represents the full-text of an article is not random, but is a consequence of style and writing conventions in different subdisciplines and can be used to infer an "evolutionary" tree of subfields within Computer Science.