Ch. 5: Presentation of Search Results
This chapter describes interfaces for the search results presentation portion of the information seeking process, focusing for the most part on ideas that are currently in use.
5.1: Document Surrogates
The most common way that search results are displayed is as a vertical list of information summarizing the retrieved documents. (These search results listings are often known as “search engine results pages,” or SERPs, in industry.) Typically, an item in the results list consists of the document's title and a set of important metadata, such as date, author, source (URL), and length of the article, along with a brief summary of a relevant portion of the document (see Figure 5.1). The representation for a document within a results listing is often called a search hit. This collection of information is sometimes referred to as the document surrogate. Marchionini and White, 2008 note that document surrogates are summary information intended to help the user understand the primary object, as opposed to metadata more broadly construed, which can also serve this purpose but is often more tailored towards use by computer programs.
The quality of the document surrogate has a strong effect on the ability of the searcher to judge the relevance of the document. Even the most relevant document is unlikely to be selected if the title is uninformative or misleading. (Some Web search algorithms try to capture the quality of the title description as part of the ranking score.) The descriptiveness of the summary is also very important and is discussed in detail below.
To determine which of the many aspects of a document surrogate lead to the best usability for Web search engine results, Clarke et al., 2007 tested which features of surrogates (consisting of query-biased summaries along with a title and URL) were associated with receiving significantly more clicks in a query log. They used the evaluation trick of clickthrough inversion, in which the features of the clicked-on surrogate are compared to those of the surrogate directly above it in the search results listings. The motivation is that because users are known to be biased towards clicking documents higher up in the rankings (Joachims et al., 2005), if they buck the trend and click on a lower-ranked document, the features of that document's surrogate must include compelling information that prompted the click. Evaluating on 10,000 pairs of summaries, where hit A appears above B, but B receives more clicks, Clarke et al., 2007 found significant effects for the following features:
- Summary is present in hit B but is missing in A.
- Summary is long in hit B (> 100 characters) but short (< 25 characters) in hit A.
- Title of hit B contains more query term matches than A's title.
- Title of hit B starts with a phrase contained in the query, but A's title does not.
- Title, summary, and URL together for hit B contain the query as a phrase match, but do not for A.
- Summary for hit B contains one match for every query term; for hit A there are more matches for some terms but some are missing.
- URL for B is of the form www. query .com but is not in this format for A.
- URL for B is shorter, in terms of slashes, than A's.
- URL for B is shorter, in characters, than A's.
- Summary B, but not A, passes a simple readability test.
Clarke et al., 2007 concluded that these and other results supported the following heuristics:
- Where possible, all the query terms should appear in the surrogate, reflecting their relationship to the corresponding Web page.
- When the query terms are present in the title for the hit, they need not appear in the summary.
- Length and complexity of URLs should be reduced, and URLs should be selected and displayed in a manner that emphasizes their relationship to the query.
They also found effects for the appearance of particular words. Among others, official, and, tourism, attractions, sexy, and information had positive influence on clickthrough, whereas encyclopedia, wikipedia, and free had negative influence.
5.2: KWIC, or Query-Oriented Summaries
As mentioned above, most search results listings today show an extract from a retrieved document that summarizes its contents. This extract is referred to with several different names, including summary, snippet, and abstract.
An important property of modern Web search surrogates is the display of a summary that takes the searcher's query terms into account. This is referred to as keyword-in-context (KWIC) extractions for use in display of retrieval results. In KWIC views (also referred to as query-biased, query-dependent, query-oriented or user-directed summaries (Tombros and Sanderson, 1998) ), sentence fragments, full sentences, or groups of sentences that contain query terms are extracted from the full text and presented for viewing along with other kinds of surrogate information (such as document title and abstract). Early versions of this idea were developed in the Snippet Search tool by Pedersen et al. (Pedersen et al., 1991, Rao et al., 1994) and the SuperBook tool (Landauer et al., 1993) (see Figure 8.4 in Chapter 8).
A KWIC, or query-oriented extract, is different than a standard abstract, whose goal is to summarize the main topics of the document but might not contain references to the terms within the query. A query-oriented extract shows sentences that summarize the ways the query terms are used within the document. In addition to showing which subsets of query terms occur in a retrieved documents, this display also exposes the context in which the query terms appear with respect to one another.
Research on document summarization indicates that the most generally applicable heuristic for making a good short summary is to show the first few sentences of a document (Kupiec et al., 1995). By contrast, research suggests that query-biased summaries are superior to showing the first few sentences in retrieval results. Tombros and Sanderson, 1998, in a study with 20 participants using TREC ad hoc data, found higher precision and recall and higher subjective preferences for query-biased summaries over summaries showing the first few sentences of retrieval results. Those using query-biased summaries also invoked significantly fewer views of the full text articles, effectively avoiding many of the non-relevant documents. Similar results for timing and subjective measurements were found by White et al., 2003a in a study with 24 participants.
With a query-biased summary, in many cases, an information need can be satisfied by viewing the document surrogate alone. For example, for a search on how to prevent cheese from molding on Google, some of the query extracts contain answer suggestions alongside the query terms themselves, while others are cut off and require a visit to the full page (see Figure 5.2).
Although KWIC and query term highlighting has been thought to be an effective technique for decades (Luhn, 1959), the prevalence of query-biased summaries is relatively recent. Hearst, 1999b wrote:
“The KWIC facility is usually not shown in Web search result display, most likely because the system must have a copy of the original document available from which to extract the sentences containing the search terms.”
At that time, costs of storage and concerns about intellectual property rights prevented search engines from storing entire copies of the crawled data. Instead, they stored and displayed only the first few sentences of text from each document. Subsequent to that writing, Google began storing full text of documents, making them visible in their cache and using their content for query-biased summaries. Keyword-in-context summaries have become the de facto standard for web search engine result displays.
5.2.1: Sentence Selection for Query-Oriented Summaries
There are significant design questions about how best to formulate and display query-biased summaries. As with standard document summarization and extraction, tradeoff decisions must be made between how many lines of text to show and which lines to display.
Several researchers have experimented with models in which sentences are scored according to attributes such as position in the document, the words they contain, and the proportion of query terms they contain. The highest scoring sentences are then included in the summary. White et al., 2003a experimented with different sentence selection mechanisms, including giving more weight to sentences that contained query words along with text formatting (e.g., boldface or italics). Goldstein et al., 1999 augmented this kind of model with linguistic cues, finding that summary sentences tended to begin with articles more than non-summary sentences, and included indefinite articles more frequently than definite articles (e.g., a versus the). Proper nouns and other named entities tended to appear at a higher percentage in summary vs. non-summary sentences. Goldstein et al., 1999 found negative evidence for inclusion of anaphoric references (e.g., this, these) negations (e.g., not, no), evaluative or vague words (e.g., often, about, several), along with a number of other features.
Those approaches ignore relationships between sentences. Varadarajan and Hristidis, 2006 presented a method to create query specific summaries by identifying the most query-relevant fragments and then combining them using graphs representing the document structure. In a small comparison study, 15 participants assigned higher ratings to the resulting summaries than to those produced by two commercial desktop search systems.
5.2.2: Summary Length for Query-Oriented Summaries
For determining how many words or sentences to show, there is an inherent tradeoff between showing long, informative summaries and minimizing the screen space required by each search hit. There is also a tension between showing short snippets that contain all or most of the query terms and showing coherent stretches of text. If the query terms do not co-occur near one another in the same sentences, then the extract has to become very long if full sentences are to be shown. Some Web search engine snippets compromise by showing fragments of sentences instead.
Paek et al., 2004 experimented with showing differing amounts of summary information in results listings, where only one result in each list of 10 was relevant. For half of the test questions, the results were visible in the original snippet, and for the other half, the participant needed to view more information from the relevant search result. They compared three interface conditions:
- (i) A standard search results listing, in which a mouse click on the title brings up the full text of the Web page,
- (ii) “Instant” view, which upon a mouseclick, expanded the document summary to show additional sentences from the document, where those sentences contained query terms and the answer to the search task, and
- (iii) A “dynamic” view that responded to a mouse hover, and dynamically expanded the summary with a few words at a time.
Eleven out of 18 participants preferred the instant view over the other two views, and on average all participants produced faster and more accurate results with this view. Seven participants preferred dynamic view over the others, but many participants found it disruptive. The dynamic view suffered from the problem that, as the text expanded, the mouse no longer covered the selected results, and so an unintended, different search result sometimes started to expand. Notably, none of the participants preferred the standard results listing view.
Cutrell and Guan, 2007 compared search summaries of varying length: short (1 line of text), medium (2-3 lines), and long (6--7 lines), using search engine-produced snippets (it is unclear if the summary text was contiguous or included ellipses). In a study with 22 participants, they found that adding more information to the summary significantly improved performance for information tasks (e.g., “find when the Titanic set sail for its only voyage and what port it left from”) but degraded performance for navigational tasks (e.g., “find the home page of World Cup 2006 soccer games”). They postulated that this effect resulted from the extra text distracting searchers from the URL. Using eye tracking, they found that participants spent a larger proportion of time looking at information other than the URL for the navigational queries with long contexts, thus suggesting that the less relevant information was distracting them. They did not report on subjective responses to the different summary lengths.
Lin et al., 2003 found that a short paragraph was preferred over a single sentence and an entire document for a question-answering system. Kaisser et al., 2008 asked judges to categorize a large set of long (question-like) queries according to the expected answer type (person, place, product, advice, general information, etc.) and preferred response length (word or phrase, sentence, one or more paragraphs, full document). They then developed high-quality answer passages of different lengths for a subset of these queries, and asked judges to rate the quality of these answers. They found that different query types are best served with different response lengths, and that for a subset of especially clear queries, human judges can predict the preferred result length. Their results furthermore suggest that standard summaries are too short in many cases, assuming that a longer summary shows information that is relevant for the query.
Thus, the evidence suggests that for queries that are more exploratory in nature, a paragraph-length excerpt may be preferable to a short, elided snippet, despite the extra scrolling it requires to see more results, so long as that paragraph is not too long.
5.2.3: Sentence Fragments vs. Full Sentences for Query-Biased Summaries
It is unclear if sentence fragments are preferable to full sentences, despite the fact that sentences take up more space on the page. Aula, 2004 performed a controlled experiment comparing three different layouts for query-biased summaries, as shown in Figure 2.3 in Chapter 2, with the aim of determining if showing a series of sentence fragments separated by ellipses is desirable. In the “Bold” case, the summary was left unchanged from what was returned by the Google search engine, with query terms bolded and fragments separated by ellipses, while in the “Plain” style, the summary was the same as in “Bold” but with the boldface weighting removed. In the “List” condition, the fragmented paragraph was replaced with a bulleted list; every time an ellipsis appeared in the summary, a new list item was created, preceded by a small arrow. Incomplete sentences were marked with ellipses at the start and/or end of the list item. The 27 participants each performed 30 tasks, doing 10 tasks in each display condition. They were asked to find the right answer for a query from the results list as quickly as possible. The list style view produced significantly better results than the standard bolded style or the standard style without bolding. In fact, in the bolded view, participants were significantly slower than with the standard view without boldface. However, the task of finding the one answer out of a list of 10 incorrect answers is different than that of determining if a document is relevant to a query or not, and the author cautions against drawing conclusions against this kind of highlighting based on this study alone.
Rose et al., 2007 varied search results summaries along several dimensions, finding that text choppiness and sentence truncation had negative effects, and genre cues had positive effects. They did not find effects for varying summary length, but they only compared relatively similar summary lengths (2 vs. 3 vs. 4 lines long). White et al., 2003b performed an experiment with 18 participants that found that showing high-ranking sentences alone might be better than showing snippets.
Kanungo and Orr, 2009 obtained hand-labeled readability scores and then used a machine learning algorithm to determine which of a small set of features predicted readability. They found that the following features negatively influenced readability scores: a large percentage of capital letters, a large percentage of punctuation, a large percentage of stopwords (as it can signal spam), and a large number of characters per word.
5.3: Highlighting Query Terms
Highlighting of query terms has been found time and again to be a useful feature for information access interfaces (Landauer et al., 1993, Lesk, 1997, Marchionini, 1995). Term highlighting refers to altering the appearance of portions of text in order to make them more visually salient, or “eye-catching.” Highlighting can be done in boldface, reverse video, or more commonly today, by displaying a colored background behind each occurrence of a query term, assigning a different color to each term. This display helps draw the searcher's attention to the parts of the document most likely to be relevant to the query, and to show how closely the query terms appear to one another in the text. As discussed in Chapter 4, query term proximity is a strong indicator of relevance.
Highlighting can occur both in retrieval results listings and in the retrieved documents themselves. In systems in which the user can view the full text of a retrieved document, it is often useful to highlight the occurrences of the terms or descriptors that match those of the user's query. The Firefox Web browser and Google toolbar allow users to search for words within the currently viewed document, and then display the hits with color highlighting. Color highlighting has also been found to be useful for scanning lists of citation records (Baldonado and Winograd, 1998).
If the text is long, then showing an overview of where the highlighted terms occur throughout the document can be useful. This can be done in several ways. One way is to use the document scrollbar to show the location of term hits. The Read Wear/Edit Wear system (Hill et al., 1992) used the scrollbar to indicate information such as the amount of time a reader has spent looking at a particular location in the document, or locations of query term hits. Byrd, 1999 suggested applying a different color to each query term, and showing the corresponding colors to the appropriate locations within a scrollbar-like widget (see Figure 10.13). The Thumbar system (Graham, 1999) used a similar scrollbar widget on the left hand side and a visualization of hits for the important terms of the document on the right hand side. This idea has been applied in the Chrome Web browser, which uses such a visualization to show where search hits occur within a searched web page (see Figure 5.3).
Another way to highlight query term hits in a long document is to use an overview display. Baudisch et al., 2004 asked 13 participants to compare three different methods of viewing long web pages retrieved for pre-defined queries. In the first case, the web page was shown as usual, but each query term was highlighted with a different color. In the second (see Figure 5.4), an overview screen on the side showed a miniature version of the entire document with query terms highlighted, as well as providing highlighting on the main page as in the first case. The third design was similar to the first but also showed a fisheye view with term highlighting at the bottom of the page. The participants were asked to perform 4 tasks. The time taken depended on the task and the interface, but 10 out of 13 participants preferred the highlighted overview version over the other two. Chapter 10 discusses other variations on these ideas.
5.4: Additional Features of Results Listings
In addition to the standard metadata of title, author, date, etc., and search result summaries, this section discusses several other features that have been found useful (or not) for search results listings.
Number of Hits Per Page:: Web search engines typically show ten results, or “hits,” per page, with hyperlinks to additional pages of results. In the earlier days of the Web, this was not standardized and as many as 30 hits per page were often shown (Reiterer et al., 2005). A Google VP reported that despite the fact that users said they wanted more hits per page, an experiment in which the number of hits was increased to 30 hits per page showed a 20% reduction in traffic (Linden, 2006). The reason turned out to be that while the page with 10 results took 0.4 seconds to generate, the page with 30 results took 0.9 seconds on average. Linden, 2006 found similar user sensitively to half-second delays at Amazon.com.
Graphical Displays of Relevance Score:: For many years, systems that use statistical ranking showed a numerical score or an icon alongside the title, indicating the computed degree of match or probability of relevance to the query. Icons included partially filled horizontal bars or a line of graphical stars, as in movie ratings, indicating degree of match or relevance.
Thus graphical or numerical displays of relevance scores have fallen out of favor and are now rarely shown; there are several possible explanations for this. First, in order to understand the score one must have some knowledge of the complex underlying ranking algorithm, which is of course not to be expected for general users. Second, often the top scores are close together numerically, and so showing the score does not add information beyond the rank ordering provided by the vertical results list. Third, other surrogates such as query terms in context are often more informative than a general relevance score. Fourth, showing the score gives information to a search engine's competitors and spammers who might try to reverse-engineer the ranking algorithm. And finally, usability studies that compare interfaces with and without graphical bars tend to find that users do not prefer them, nor do they affect the timing scores (Tanin et al., 2007, White et al., 2007) (although these studies did not use standard search results listings). More elaborate graphical representations of search results matches are discussed in Chapter 10.
Previews of Document Content:: In most graphical search interfaces, clicking on the document's title or an iconic representation of the document shown beside the title will bring up a view of the document itself, either in another window, or replacing the listing of search results. Some systems have experimented with making the document content more immediately available without the need to leave the Web page; Figure 5.5 shows content prefetching when viewing a hyperlink using Snap's Snap Shot system.
Indicators of Search Result Diversity:: Some Web search engines attempt to support a notion of “diversity” in the first few results displayed. This is especially important for ambiguous queries for which there are several common interpretations or meanings for a given word. For example, a search on the term labs at Google at the time of writing shows three sets of results on the first page, separated by horizontal rules (see Figure 5.1). The first set shows hits on two research laboratories, the second shows the message See results for labrador retrievers along with hits on this topic, and the third section is general search results ordering.
Indicators of Additional/Related Hits:: Some Web search engines group related hits from one Web site using an indented link, along with a link to more hits from that site; an example can be seen under the link Adobe Labs - Homepage in Figure 5.1.
Sitelinks:: More recently, Web search engines have adopted a feature that shows an indented list of important pages from the top-scoring Web site for a query, along with a link to more pages from that Web site. This feature is referred to in the industry as sitelinks or deep links, and informal reports suggest they are frequently clicked on. Presumably, these links are chosen based on clickthrough popularity for other queries as well as descriptive titles on the links. This feature exposes some of the content that is buried one or more levels within the Web site, thus potentially saving the user time and effort in scanning the site's home page to find the next link, and also eliminating the need to load the home page. (This interface is a simpler version of the idea of exposing the structure of a site's hits espoused by the Cha-Cha (Chen et al., 1999) and AMIT (Wittenburg and Sigman, 1997) projects.) In Figure 5.1, sitelinks for the Google Labs site point to Trends, Code Search, and other pages on the site.
Shortcuts:: Search engines are also attempting to provide “shortcuts” for directed or focused information needs directly on the search results page, becoming in effect “answer engines” for certain queries (Nielsen, 2004b). Figure 5.6 shows the extensive, contextually relevant information provided by Yahoo in response to the query weather in Berkeley.
Blended Results and Media Types:: Web search engines are increasingly blending search results from multiple information sources, not just Web pages (iProspect, 2008). Figure 5.7 shows results for a very general query on kittens at Hakia, which the search engine converts into a query on Cats. Recognizing that this is a very general query, the system provides general resources about the topic, separated by tabs, including news headlines, general pet care sites, general sites for finding a pet, and a table of contents for these different types of information.
Figure 5.8 shows results for the general query jets at Microsoft's search engine, which blends sports scores, news, and Web search results. Note also the diversity of topics in the first few results, including a link to the New York Jets football team site, a link to a site selling private jets, and a link to an engineering society.
Informal reports suggest that in most cases, these kinds of multimedia results best placed a few positions down in the search results list; when they replace the first hit they can cause people to leave the site. When placed just above the “fold” (above where scrolling is needed) they can increase clickthrough. Eye-tracking studies suggest that even when placed lower down, an image often attracts the eye first (Hotchkiss et al., 2007). It is unclear if information-rich layouts such as those used by Ask.com and Hakia galleries are desirable or if this much information is too overwhelming for users on a daily basis. It may be best to show this richness only for certain types of queries, such as the general ones shown here.
Organic Results vs. Advertisements: In most Web search engines, the search hits are shown in order of computed relevance to the query. In some cases, however, paid advertisements are shown at the top and/or to the side of search results. These are usually visually distinguished to differentiate the ads from the “organic” results, as those hits based solely on relevance as called. The discussion of search ads is outside the scope of this book.
5.5: The Effects of Search Results Ordering
Search results are often listed in an order specified by a relevance metric. Alternatively, results are ordered according to a metadata attribute, such as reverse chronological order for news search and email search, or number of citing papers for journal article search. It is also common to group results by well-defined metadata fields, such as grouping email by sender name or journal article by author name.
Studies and query logs show that searchers rarely look beyond the first page of search results. If the searcher does not find what they want in the first page, they usually either give up or reformulate their query (Chapter 6 discusses query reformulation in detail). Furthermore, several studies suggest that web searchers expect the best answer to be among the top one or two hits in the results listing, and this expectation influences whether or not they will click on a result. Several of these studies are discussed below.
An eye-tracking study by Granka et al., 2004 on 26 participants and 397 queries found that on average, participants took 7.78 seconds to select a document, but the time varied significantly among the 10 pre-defined search tasks, from 5-6 seconds up to 11 seconds for the most difficult questions. The first result was selected approximately 85% of the time, and the second link about 10% of the time. Furthermore, the first and second results were by far the most viewed, with a sharp dropoff starting with the third result. A followup eye-tracking study by Joachims et al., 2005 with 29 participants showed that the percentage of times a search result in the top ten listings was looked at dropped off sharply from first search result to sixth, and then flattened at around 5% of the time for results 7--10. Likelihood of clicking on the result, however, dropped much more dramatically, falling from 43% of the time for the first hit to 15% of the time for the second hit, 10% of the time for the third hit, and 5% or less for the rest. This result held despite the fact that 5 of the 22 participants were shown the results in reverse order of their original ranking, and another 5 of the participants were shown the top two hits in swapped position.
Joachims et al., 2005 also found that participants tended to view the first and second-ranked results right away, with a large gap before viewing the third-ranked abstract. They also found that while participants did not necessarily view all abstracts above a click, they view substantially more abstracts above than below the click. More surprisingly, they also found that the abstract right below a click is viewed roughly 50% of the time.
Joachims et al., 2005 also found bias in relevance judgements based on placement location. They did a followup experiment focusing on the top two results, since these are scanned equally frequently. They compared how often a participant clicked on either result 1 or result 2 depending on the manually judged relevance of the abstract. They found that participants were influenced in their relevance assessment by the order of presentation, since the number of clicks on link 1 was significantly higher than its relevance would merit.
Despite these results, Joachims et al., 2005 found that participants were not blindly following link order. In a condition in which they complete reversed the rank order of the top 10 results, they found that in the reversed condition, participants viewed lower ranked links more frequently, scanning significantly more results than in the normal condition. Those who saw the reversed condition were also much less likely to click on the first link and were more likely to click on a lower-ranked link. The average rank of a click in the normal condition was 2.66 and 4.03 in the reversed condition. However, the average relevance of the selected documents in the reversed condition was lower than in the normal condition.
Guan and Cutrell, 2007 performed an eye-tracking study in which they pre-determined the queries and results, and controlled which of the top 10 positions the one relevant result appeared in. They also contrasted navigational and informational query types. In a study with 18 participants, they found a significant main effect of target position on total task time and on query type. Participants spent more time when the target was farther down the result list, but this extra time did not result in more success at making the correct choice. The click accuracy rates dropped from about 84% when the target result was in position 1 or 2, to about 11% when the correct response was in position 8. For navigational queries, when participants did not find the result below position 3, they either selected the first hit (40% of the time) or reformulated their query. For informational search, participants rarely reformulated the query without first trying to click on the first hit (about 50% of the time) or clicking on the other links at random. One might infer from this that participants are more confident in the search engine ranking for the relatively easier navigational queries than for the more general informational ones. Guan and Cutrell, 2007 examined the results of the eye-tracking data and found that participants did look at the lower ranked results. They concluded that the participants' behavior was caused by their expectation that the relevant results would be at or near the top.
A somewhat different kind of behavior was seen in an eye-tracking study by Aula et al., 2005b. They analyzed the eye movements of 42 students on 10 pre-defined queries, and found two distinct styles of scanning results list. 46% of the participants were “economic,” scanning at most half of the 6-7 visible search results in 50% of the tasks. The remaining 54% were “exhaustive” evaluators, who for most queries viewed more than half of the visible results, and in some cases scrolled down to see the full list of 10 hits. The difference in time before first action was significantly shorter for the economic searchers, especially when good results were available. The authors find a marginal difference between the evaluation style and computer experience, with more experienced searchers tending to use the economic style, and speculate that the Granka et al., 2004 study may have employed only expert searchers, thus explaining the different results.
(a)
(b)
5.6: Visualization of Search Results
The bulk of the information visualization ideas that are tried for search apply to the display of retrieval results. Most of these ideas do not survive in mainstream search interfaces, but some ideas are currently getting some play. One frequently suggested idea is to show search results as thumbnail images rather than as textual surrogates (Czerwinski et al., 1999, Dziadosz and Chandrasekar, 2002, Woodruff et al., 2001), but none have shown a proven advantage for search results viewing. Nonetheless, a startup company called SearchMe presents a search engine using a “cover flow” interface, which seems to be influenced by the Web Forager/Web Book interface of Card et al., 1996, as shown in Figure 5.9. It remains to be seen if people will use this kind of interface on a regular basis. The extreme sensitivity of searchers to delays of even 0.5 seconds suggests that such highly interactive and visual displays need to have a clear use-case advantage over simple text results before they will succeed. Other approaches to using information visualization for search results display are described in detail in Chapter 10.
5.7: Conclusions
Search results presentation is a critical component of the search cycle. This chapter has summarized empirical research showing which aspects of a document are best shown in retrieval results, along with the characteristics of the results listing itself. Although the basic look of Web search results listings is similar to what was seen ten years ago, a number of subtle innovations have been introduced and found useful for helping searchers make decisions about which links to select for further investigation. These include showing query terms in the search results surrogate, striving for proximity of the query terms to one another where possible, balancing a tradeoff between length and informativeness in document summaries, providing information “scent” about the top-ranked Web site, where appropriate, and differentiating the presentation for ambiguous queries.
Additional important aspects in the design of search results listings that were not covered in this chapter but were discussed in Chapter 1 include providing highly responsive (fast) results, paying attention to small details in the layout, font, color and spacing of the results, and being attuned to aesthetic issues in design. In addition, presentation of search results has been a major focus of the work on employing visualization techniques to search interfaces; these efforts are discussed in detail elsewhere.