Negative Relevance Feedback for Exploratory Search with Visual Interactive Intent Modeling

Abstract

In difficult information seeking tasks, the majority of topranked documents for an initial query may be non-relevant, and negative relevance feedback may then help find relevant documents. Traditional negative relevance feedback has been studied on document results; we introduce a system and interface for negative feedback in a novel exploratory search setting, where continuous-valued feedback is directly given to keyword features of an inferred probabilistic user intent model. The introduced system allows both positive and negative feedback directly on an interactive visual interface, by letting the user manipulate keywords on an optimized visualization of modeled user intent. Feedback on the interactive intent model lets the user direct the search: Relevance of keywords is estimated from feedback by Bayesian inference, influence of feedback is increased by a novel propagation step, documents are retrieved by likelihoods of relevant versus non-relevant intents, and the most relevant keywords (having the highest upper confidence bounds of relevance) and the most non-relevant ones (having the smallest lower confidence bounds of relevance) are shown as options for further feedback. We carry out task-based information seeking experiments with real users on difficult real tasks; we compare the system to the nearest state of the art baseline allowing positive feedback only, and show negative feedback significantly improves the quality of retrieved information and user satisfaction for difficult tasks.

A study of visually linked keywords to support exploratory browsing in academic search

“…The colour map and associated hexadecimal codes used for keyword highlighting, created using ColorBrewer (https:// colorbrewer2.org/ #type=qualitative&scheme=Set1&n=9 ). Peltonen, Belorustceva, et al, 2017;Peltonen, Strahl, et al, 2017), they generally operate as a container for saved search results or snippets, providing limited interactivity beyond adding and deleting items from the workspace, and in some cases supporting the sharing of work with others (Yue et al, 2014). While we could have done the same, we hypothesized that the visually linked keywords could also help searchers when they are re-evaluating the search results saved in the workspace.…”

Conversational Product Search Based on Negative Feedback

“…Later, Karimzadehgan and Zhai [11] further improved the performance of negative feedback by building a more general negative topic model. Peltonen et al [19] introduced a novel search interface, where keyword features of the non-relevant results are provided to users, and they are asked for feedback on the keywords. Then a probabilistic user intent model is estimated to refine reranking.…”

“…Most previous work on negative feedback only uses result-level non-relevant information except [19], which further acquires keywordlevel feedback on non-relevant results. Although we also ask users for feedback on finer-grained information, we leverage aspect-value pairs of non-relevant results and focus on product search.…”

Topic-Relevance Map

“…Interactive interfaces that enable transparent control on user models have recently become popular [35,36,2,4,42,32]. The idea behind these approaches is that, as opposite to visualizing results, the user model is visualized and the user can interactively provide feedback on the search intentions using the visualization.…”