Annotations are the most powerful tool in RWR; as is often the case, the more powerful a tool, the more difficult it is to use it for maximum effectiveness. If it is desired, simple text can be entered into the term fields, and used as a basic aid in locating a reference. Greater functionality can be gained however if the term and discussion fields are used in the following manner to create the RWR style of annotation.
Underlying this method of annotation is the desire to capture concepts, claims, interpretations, discussions, and other salient meaning from an article. There are significant areas of research into the use of machine learning for the purpose of extracting or inferring meaning or content from textual material. Rather than wait for the development of production ready machine learning tools, RWR builds on the idea that when reading, the reader must reflect and interpret the information being presented. In the paper based environment, the reader might make marginal notes, highlight passages, or make notes on index cards. All of these techniques work well, except for one missing functionality, you can’t search across a library for that note you made, which you know is somewhere in that pile of paper. At a basic level, RWR provides the electronic equivalent of making marginal notes, highlights, and index cards, so you can then search using them. RWR goes further and introduces the concept of using simple sentences to capture a concept or claim that is the central idea of an annotation.
A simple sentence has a variety of definitions, with a common one giving it the basic structure of subject verb object. If we consider how language is learnt in children, it typically progresses from single words to simple sentences. We can use that progression as a metaphor for how retrieval of meaning from text has progressed from the use of keywords to simple sentence annotations. A keyword is typically a noun, and therefore limited in the information content it can convey, where a simple sentence can convey a far richer meaning.
This leads to the recommended method for using annotations in RWR. A common article writing style uses a paragraph as a unit of information, i.e. only one thing should be discussed in a paragraph. The first sentence of a paragraph often states what that one thing is (a claim, assertion, idea …), with the remainder of the paragraph elaborating, providing evidence for or against, etc. there may be a conclusion or restatement at the end of the paragraph. The reader when encountering a paragraph of interest needs to interpret that paragraph, with the first task being to create a simple sentence that captures what that paragraph is about. Then in the discussion, provide some discussion. Which might include an assessment of validity, initial impressions, further investigation required. The criteria for evaluation of whether the annotation is complete is whether on subsequent reading you have nothing further to add. In other words you should never have to read a paragraph in depth twice. You may skim a paragraph and make a brief note, and subsequently come back some time later and re-read in depth when you have a particular need, but not read twice in depth. RWR can tag the paragraph, and show a highlight which references the annotation.
A comparison of the functionality gained with RWR annotations over keywords and plain text search will help illustrate how much more powerful this annotation method is and why it is worth the effort of composing the simple sentences and discussion in an annotation.
Keywords on their own are quite limited within a library, based on the assumption that large groups of the articles in the library cluster around a few keywords. Keywords are useful in a heterogeneous context however, but that is not likely the case here. Keywords are also assigned to a document, they say that this document says something pertaining to these keywords. But can’t give any indication of what is said. Text searches can improve on keywords in that they can show the context in which the search term is used, and if there is a relationship between documents, some inference of relative importance can be gained. In both cases however, search terms and keywords are typically nouns, and are unable to provide much in the way of meaning of information content.
If you were performing a meta-analysis, or systematic review, keywords and search terms help you find articles that are relevant to your topic, but provide little value when you are writing the analysis and you need to reference the claims and results from a variety of articles. A simple sentence annotation on the other hand is a very useful tool for finding specific information and your notes when you read the article earlier.