search history

search history

Exploring the Evolution and Impact of Search History

1. Introduction to Search History

Search histories contain more user sessions with the user submitting multiple queries on multiple topics, the order of these queries being an important aspect that can be essential for understanding user behaviors. For instance, a user may enter the query “history” before entering a list of “World War II” related queries, indicating a progression of information needs as the user refines or redirects his or her search. Such information may not be evident in a single session format but is valuable for understanding how users conduct research and for providing additional context to the queries being submitted. The unique characteristics and substantial amount of search history information contain potential inferences for advancing our understanding of users of search engines that may be incorporated into existing IR models to improve the performance of search engines. Despite the valuable data contained within search histories, only a handful of studies have been conducted to explore what proportion of search history also exhibit their inherent privacy issues.

Searches that a user conducts through a search engine are automatically associated with an individual search history that provides a detailed record of the user’s information needs. This search history, or “web history” as it is commonly referred to by industry, consistently ranks amongst the most valuable assets for search engines as it has been recognized to have potential for increasing the accuracy of search results, assisting analysis of web-surfing trends to help advertisers, and improving the user experience by personally customizing search interfaces in a process known as “search personalization”. Accompanying its numerous advantages, research within the field of Information Retrieval (IR) has also shown that search history has inherent privacy issues, seriously affecting a user’s right to keep search information without personally identifying information private. Retaining and analyzing search history can potentially reveal the personal characteristics of a user including health problems, financial situations, or personal sexual preferences.

2. Technologies and Algorithms Behind Search History

In this section, we discuss possible technologies and algorithms that are able to create and apply search history support mechanisms. Thus, this section provides a more detailed look into how it is possible to support this content-based matching mechanism of search systems using information from different sources of information about the user and his interaction with the search engine. The following sections present a first specific mechanism and demonstrate how it can be used. Then topics such as security and privacy will be addressed, in order to take a broader look at how search systems can deal with and how users can benefit from search history.

When a user sends a query to a search system, that query reaches a cloud computing-based platform, which may contain a vast collection of web documents. For processing the query, the search system identifies a set of web documents that relate to the user query in a way, ranks these documents, and returns the top-ranked documents as a result set. Typically, the processing of a query is divided into two main steps: a candidate document selection (CDS) phase, followed by a result ranking (RR) phase. In the CDS stage, the search system needs to identify an initial set of documents that relate to the query, while in the RR stage, the documents of the candidate set are ranked according to the expected level of interest that the user has on them. Different algorithms can be used for either one or both of the CDS or the RR steps.

3. Privacy and Ethical Considerations in Search History

There are currently no U.S. laws and regulations providing protection against the search corporation employees who view queries when attempting to help customers or those designs that use the content to aid marketers in honing precisely targeted ads. Proponents of search engine privacy and data retention policies and other related information technology (IT) security mechanisms hope to improve privacy rights by employing persuasive argument and adopting information retrieval (IR) algorithms, automated user interfaces and back end information systems to disclose specific information for appropriate business uses. The results benefit business and private lives in unexpected ways.

Search engines are a great resource for finding specific information. Users generally trust the search engine companies, as well as the companies running the various open content wikis or congresses creating and promoting technical standards. Users feel comfortable that their anonymous queries will be kept private from prying eyes. Trust in the system is so high that users are comfortable inputting questions that they may not wish to share with close friends or family. Unfortunately, the data becomes traceable when these same users log in to take advantage of additional search features, use search services engineered to work with other value-added services, and custom search engine solutions that automatically “guess” what makes the most relevant results without filtering the user IP address from the search query.

4. Applications and Future Developments of Search History

With just single datasets of earlier searches, this dissertation has illustrated how past searches can yield nuanced user profiles. By triangulating these search profiles with different data sources, user modeling can both enrich classical information retrieval tasks such as movie recommendations with long-term user context or guide intimately customized information retrieval such as providing detailed hurricane information tailored to an individual’s decision-making process. And of course the same user modeling that informs machine-learned predictions can provide a nuanced analysis of user’s underlying health and preferences.

We open this chapter by discussing applications of search history. In the same way that user searches are aggregated to predict TV programs, user queries can be used to inform directed speech recognition; the Google Android mobile phone operating system already exploits repeated searches to accelerate verbal command processing. The same search prediction models that underpin increasing productivity across a range of human-computer dialog systems may suggest natural next steps in dialogue. In the domain of intelligent tutoring systems, instead of relying on user responses during testing or on identifying comprehension mismatches, selecting the sequence of hints or even adaptive assessment items can be made more efficient by temporal search patterns.

Just as many applications draw from natural language processing, automatic speech recognition, and machine translation technologies, the outcome of this dissertation, i.e., the inference from search history, has enormous potential to inform downstream machine learning tasks, particularly those that are user-centric. I focus on several such applications in human-computer dialog systems, user profiling in information retrieval systems, and intelligent tutoring systems and conclude by forecasting the ways search history can be exploited in the future.