The Introduction to a paper is a place for you to tell the story of the research that is presented. That story is not what you did to complete the research, but rather why the work is interesting. And while the research you are writing about in a paper might be part of a larger story (e.g., your thesis), the paper’s story is also not necessarily that larger story. Instead, it is the story that frames just the current work and its contributions as clearly as possible. The goal is to capture the reader’s attention, provide context for the included research, and set expectations for what is to come.
A simple, reliable Introduction outline is:
- General problem area
- Specific problem being addressed
- Thesis statement
I will illustrate this structure using the introduction from a paper that Lydia Chilton and I wrote for WWW 2011 titled Addressing People’s Information Needs Directly in a Web Search Result Page. Other papers that follow this structure fairly closely include Understanding and Predicting Personal Navigation and #TwitterSearch: A Comparison of Microblog Search and Web Search.
General Problem Area
Set the context for the paper by introducing the general problem space. This should not involve defining an entire field. If you are writing a paper for SIGIR, for example, you do not need to describe why information retrieval is important. Instead, assume the reader is a member of the community the paper is being published in, and use the first paragraph to sell them on the importance of the area of focus.
In our example paper, Lydia and I were writing for the World Wide Web community, which is familiar with the web and web search, but (at least at the time we were publishing) less familiar with the concept of inline answers:
Most Web search engine users have discovered (perhaps without realizing it) a search feature that we call in this paper Answers, where relevant content is provided directly within the search result page... Answers are a step towards the long-standing goal of Web search engines to directly address their searchers’ needs, versus merely linking to relevant content.
Specific Problem Being Addressed
Once the problem area is defined, frame the specific problem that the paper will address. Give the background necessary to show that there is a hole in the literature that needs to be filled. This is often a good place to a quick sketch of relevant existing research.
The presence of an Answer on a search result page changes the value of the interaction metrics that have traditionally been used to evaluate and improve Web search result quality. When people’s needs are met by an Answer, they may not interact with the search result page at all, a signal that is traditionally interpreted as a negative experience. More sophisticated interaction metrics, such as ones based on interaction with other elements on the page or repeat engagement, must be used instead. However, because there are many different ways Answers can address people’s needs, there are also many different ways they can influence user interaction.
A common challenge for Introductions is that they often come across as somewhat abstract. While a good Introduction should allow the reader to see the research in a paper in a broad context, it can be helpful for the reader to provide a concrete example for them to hold in mind as they make their way through the paper. An illustration or image is also useful, and when possible I try to include one on the first page.
In our WWW paper, we provided a specific example of inline answer in the very first paragraph.
For example, the query “weather” no longer returns a link to http://www.weather.com as the first result. Instead, major search engines like Bing, Google, and Yahoo! use the top result space to answer the user’s query directly within the result page, providing a pictorial weather forecast of the user’s local weather. Figure 1 shows a snapshot of the Weather Answer returned by Bing for the query “weather Beijing”.Other examples of Introduction examples include:
- The query WSDM, in Understanding and Predicting Personal Navigation.
- A question about cell phone charges in Hawaii, in SearchBuddies: Bringing Search Engines into the Conversation.
- The normal body temperature for a dog, in Direct Answers for Search Queries in the Long Tail.
- Rachel's quotation, in The Perfect Search Engine: A Study of Orienteering Behavior in Directed Search.
Once the problem is clearly defined, you should describe how you addressed it in the paper. The thesis statement often begins with something along the lines of, “In this paper, we show that…” The sentence should set expectations for what can be learned by reading the paper.
In this paper, we explore the important factors that impact Answer use, and suggest several new ways to interpret query log data in the presence of Answers.
In addition to setting the reader’s expectations for what they will learn from the paper, the Introduction also needs to set their expectations for how you will support your thesis. Provide a quick sketch of what methodology is used (e.g., log analysis, lab study, longitudinal observations) and the scale (e.g., tens of people, millions of queries).
The more typical your approach is for the community you are publishing in, the less it will need to be foreshadowed in the Introduction. In the Answers paper, we used query log analysis. This approach is common in the WWW community, so merely mentioning it in our thesis statement was sufficient. Atypical approaches can very valuable to bring to a community, but require more up-front context setting. If you think there is an aspect of your approach that you will need to convince your readers to buy into (such a small study size or a controversial method), mention it as early as possible.
Towards the end of the Introduction you should highlight the paper’s contributions. Typically I do this with a few sentences summarizing key findings, but in the Answer paper we explicitly call out the contributions in bullet form:
- Answers that provide content inline can reduce engagement with the search result page, thus cannibalizing interaction.
- Repeat usage gives us insight into the relevance of Answers, even when clicks are cannibalized.
- People who consistently use the same Answer type over time are often monitoring Answer content. Repeat Answer usage within a session indicates task-based reuse.
- Answers that are triggered with identical queries are often being monitored for new content, while those that are triggered with different queries are exploratory.
End the Introduction with a simple outline to help the reader navigate the paper.
After a discussion of related work, we describe the query log data we analyzed and provide details of the specific Answer types we studied. We then present our findings… These findings suggest rich opportunities for search engines as they attempt to directly meet their user’s needs. We conclude with a discussion of these promising directions.
All of the above should be done within the first page of your paper. (Although if you include an image on the front page, you may spill over by a paragraph.) While it is nice to set the stage for your research, you also want to get your reader to the meat of the paper quickly.
Once you have mastered the basic Introduction structure, it can fun to play with the format to pique the reader’s interest. I particularly enjoy papers that draw the reader in from the first paragraph, and have written papers that lead with a particularly surprising finding:
Imagine that you are busy. Now imagine that the phone rings. It seems reasonable to assume you would be less likely to answer the call at that point than if you were available... Surprisingly, however, when analyzing over a hundred thousand enterprise phone calls we observed that people were more likely to answer the phone when busy than at other times. [From: Understanding How the Projection of Availability State Impacts the Reception of Incoming Communication]Or interesting quotation:
The proverb, “A journey of a thousand miles begins with a single step,” is attributed to the sixth century BCE philosopher Laozi. People have attempted to accomplish large tasks by decomposing them into manageable parts for millennia, and modern approaches to time management continue to take Laozi’s words to heart. [From: Selfsourcing Personal Tasks]While playing around, however, there are a few things to avoid:
- Do not define the entire problem area. Know what your reader already knows.
- Do not oversell your research. Set accurate expectations.
- Do not describe your personal journey through the research.
- Do not tell your larger research story unless it is relevant to this paper.
- Do not repeat content from the abstract.
- Do not use subsections.
- Do not take up more than a page and a half.
Many people have opinions about when to write a paper’s framing content (e.g., Abstract, Title, Introduction). I have colleagues who like to write the Introduction at the onset of their research, to help them structure their entire research agenda. Others write the Introduction when they actually sit down to write the paper, and still others advocate for writing it last, after the paper’s contributions have been distilled. My personal approach tends to do a little of each – I have a story in mind at the onset of a research project that I tell and refine as I conduct the research, I put that story down on paper when I actually start writing, and I revisit that story when I am done.
The key with respect to timing is that at different times the purpose of writing the Introduction is different. Avoid getting hung up on the wrong purpose at the wrong time. For example, people sometimes waste a lot of time agonizing over the way the Introduction frames the rest of the paper before the rest of the paper is actually written. I think it is a good idea to write a version of the story down early, but to avoid spending a lot of time doing so or getting attached to it. Think of the first draft as just filler text, and revisit it regularly throughout the writing process.
This post is part of a series of posts about the formula for academic papers. The components being discussed are: