Search Query Logs Research Articles

Long-Term Regional Influenza-Like-Illness Forecasting Using Exogenous Data.

Disease forecasting is a longstanding problem for the research community, which aims at informing and improving decisions with the best available evidence. Specifically, the interest in respiratory disease forecasting has dramatically increased since the beginning of the coronavirus pandemic, rendering the accurate prediction of influenza-like-illness (ILI) a critical task. Although methods for short-term ILI forecasting and nowcasting have achieved good accuracy, their performance worsens at long-term ILI forecasts. Machine learning models have outperformed conventional forecasting approaches enabling to utilize diverse exogenous data sources, such as social media, internet users' search query logs, and climate data. However, the most recent deep learning ILI forecasting models use only historical occurrence data achieving state-of-the-art results. Inspired by recent deep neural network architectures in time series forecasting, this work proposes the Regional Influenza-Like-Illness Forecasting (ReILIF) method for regional long-term ILI prediction. The proposed architecture takes advantage of diverse exogenous data, that are, meteorological and population data, introducing an efficient intermediate fusion mechanism to combine the different types of information with the aim to capture the variations of ILI from various views. The efficacy of the proposed approach compared to state-of-the-art ILI forecasting methods is confirmed by an extensive experimental study following standard evaluation measures.

Online Search Behavior Related to COVID-19 Vaccines: Infodemiology Study.

BackgroundVaccination against COVID-19 is an important public health strategy to address the ongoing pandemic. Examination of online search behavior related to COVID-19 vaccines can provide insights into the public's awareness, concerns, and interest regarding COVID-19 vaccination.ObjectiveThe aim of this study is to describe online search behavior related to COVID-19 vaccines during the start of public vaccination efforts in the United States.MethodsWe examined Google Trends data from January 1, 2021, through March 16, 2021, to determine the relative search volume for vaccine-related searches on the internet. We also examined search query log data for COVID-19 vaccine-related searches and identified 5 categories of searches: (1) general or other information, (2) vaccine availability, (3) vaccine manufacturer, (4) vaccine side-effects and safety, and (5) vaccine myths and conspiracy beliefs. In this paper, we report on the proportion and trends for these different categories of vaccine-related searches.ResultsIn the first quarter of 2021, the proportion of all web-based search queries related to COVID-19 vaccines increased from approximately 10% to nearly 50% of all COVID-19–related queries (P<.001). A majority of COVID-19 vaccine queries addressed vaccine availability, and there was a particularly notable increase in the proportion of queries that included the name of a specific pharmacy (from 6% to 27%; P=.01). Queries related to vaccine safety and side-effects (<5% of total queries) or specific vaccine-related myths (<1% of total queries) were uncommon, and the relative frequency of both types of searches decreased during the study period.ConclusionsThis study demonstrates an increase in online search behavior related to COVID-19 vaccination in early 2021 along with an increase in the proportion of searches related to vaccine availability at pharmacies. These findings are consistent with an increase in public interest and intention to get vaccinated during the initial phase of public COVID-19 vaccination efforts.

The use of OPAC query logs to support evidence-informed information literacy (IL) training at The University of Dar es Salaam

This paper reports the results of the study which analysed OPAC users’ searching behaviour through the use of search queries. The study analysed how OPAC search queries can be used to assess the effectiveness of the information literacy programme offered by the University of Dar es Salaam. Data were collected unobtrusively and stored in the OPAC web server’s log file. OPAC search logs were qualitatively and quantitatively analysed to determine search patterns and search query formulation. Information literacy course contents were also qualitatively analysed and compared with users’ search activities. The results suggest that search query logs are essential and high quality evidence that can be used to inform information literacy pedagogical methods and course contents improvement. A comparison between OPAC users’ search behaviours and information lteracy course content identified some areas for readjusting the way information literacy instructions are delivered and improving OPAC search functionalities. The study has contributed to the conceptual understanding and the use of evidence to support evidence-informed practices (EIPs) in delivering information literacy courses.

Data Augmentation for Abstractive Query-Focused Multi-Document Summarization

The progress in Query-focused Multi-Document Summarization (QMDS) has been limited by the lack of sufficient largescale high-quality training datasets. We present two QMDS training datasets, which we construct using two data augmentation methods: (1) transferring the commonly used single-document CNN/Daily Mail summarization dataset to create the QMDSCNN dataset, and (2) mining search-query logs to create the QMDSIR dataset. These two datasets have complementary properties, i.e., QMDSCNN has real summaries but queries are simulated, while QMDSIR has real queries but simulated summaries. To cover both these real summary and query aspects, we build abstractive end-to-end neural network models on the combined datasets that yield new state-of-the-art transfer results on DUC datasets. We also introduce new hierarchical encoders that enable a more efficient encoding of the query together with multiple documents. Empirical results demonstrate that our data augmentation and encoding methods outperform baseline models on automatic metrics, as well as on human evaluations along multiple attributes.

Spoken Conversational Context Improves Query Auto-completion in Web Search

Web searches often originate from conversations in which people engage before they perform a search. Therefore, conversations can be a valuable source of context with which to support the search process. We investigate whether spoken input from conversations can be used as a context to improve query auto-completion. We model the temporal dynamics of the spoken conversational context preceding queries and use these models to re-rank the query auto-completion suggestions. Data were collected from a controlled experiment and comprised conversations among 12 participant pairs conversing about movies or traveling. Search query logs during the conversations were recorded and temporally associated with the conversations. We compared the effects of spoken conversational input in four conditions: a control condition without contextualization; an experimental condition with the model using search query logs; an experimental condition with the model using spoken conversational input; and an experimental condition with the model using both search query logs and spoken conversational input. We show the advantage of combining the spoken conversational context with the Web-search context for improved retrieval performance. Our results suggest that spoken conversations provide a rich context for supporting information searches beyond current user-modeling approaches.

Burstiness-Aware Web Search Analysis on Different Levels of Evidences

Personalizing the analysis for web search potentially improves the search experience. A good analytical model for web search should leverage not only collective wisdom but also individual characteristics. Most of the existing analytical models, however, focus on how to utilize the collective wisdom, from a crowd, for instance. In this paper, we address the problem of user-specific web search analysis by considering the so-called burstiness in web search, which captures the behavior of rare words appearing many times in a single document. We go beyond click graph and propose two probabilistic topic models, Topic Independence Model and Topic Dependence Model. The former adopts the assumption that the generation of query terms and URLs are topically independent, and the latter captures the coupling between search queries and URLs. We also capture the temporal burstiness of topics by utilizing continuous Beta distribution. Through a large-scale analysis of a real-life search query log, we observe that each user's web search trail enjoys multiple kinds of user-based unique characteristics. On a massive search query log, the new models achieve a better held-out likelihood than existing baselines, and they can also effectively reveal the latent evolution of topics on the corpus level and user-based level.

Surveillance of early stage COVID-19 clusters using search query logs and mobile device-based location information

Two clusters of the coronavirus disease 2019 (COVID-19) were confirmed in Hokkaido, Japan, in February 2020. To identify these clusters, this study employed web search query logs of multiple devices and user location information from location-aware mobile devices. We anonymously identified users who used a web search engine (i.e., Yahoo! JAPAN) to search for COVID-19 or its symptoms. We regarded them as web searchers who were suspicious of their own COVID-19 infection (WSSCI). We extracted the location of WSSCI via a mobile operating system application and compared the spatio-temporal distribution of WSSCI with the actual location of the two known clusters. In the early stage of cluster development, we confirmed several WSSCI. Our approach was accurate in this stage and became biased after a public announcement of the cluster development. When other cluster-related resources, such as detailed population statistics, are not available, the proposed metric can capture hints of emerging clusters.

When Are Search Completion Suggestions Problematic?

Problematic web search query completion suggestions-perceived as biased, offensive, or in some other way harmful-can reinforce existing stereotypes and misbeliefs, and even nudge users towards undesirable patterns of behavior. Locating such suggestions is difficult, not only due to the long-tailed nature of web search, but also due to differences in how people assess potential harms. Grounding our study in web search query logs, we explore when system-provided suggestions might be perceived as problematic through a series of crowd-experiments where we systematically manipulate: the search query fragments provided by users, possible user search intents, and the list of query completion suggestions. To examine why query suggestions might be perceived as problematic, we contrast them to an inventory of known types of problematic suggestions. We report our observations around differences in the prevalence of a) suggestions that are problematic on their own versus b) suggestions that are problematic for the query fragment provided by a user, for both common informational needs and in the presence of web search voids-topics searched by few to no users. Our experiments surface a rich array of scenarios where suggestions are considered problematic, including due to the context in which they were surfaced. Compounded by the elusive nature of many such scenarios, the prevalence of suggestions perceived as problematic only for certain user inputs, raises concerns about blind spots due to data annotation practices that may lead to some types of problematic suggestions being overlooked.

PatternRank+NN

We propose a ranking framework, called PatternRank+NN, for expanding a set of seed entities of a particular class (i.e., entity set expansion) from Web search queries. PatternRank+NN consists of two parts: PatternRank and NN. Unlike the traditional methods, PatternRank brings user behaviors into entity set expansion from Web search queries. PatternRank is a Markov chain which simulates the Web search query process of users on the graph model for Web search query log, and ranks the features of the class. The features in the front rank are used to generate candidate entities of the class. NN, a ranking strategy called Nearest Neighbor, ranks these candidate entities such that the set of seed entities can be expanded from the candidate entities in the front rank. Our experiments demonstrate the superior performance of PatternRank+NN in comparison with the state-of-the-art methods.

Deep Learning for Adverse Event Detection from Web Search

Adverse event detection is critical for many real-world applications including timely identification of product defects, disasters, and major socio-political incidents. In the health context, adverse drug events account for countless hospitalizations and deaths annually. Since users often begin their information seeking and reporting with online searches, examination of search query logs has emerged as an important detection channel. However, search context - including query intent and heterogeneity in user behaviors - is extremely important for extracting information from search queries, and yet the challenge of measuring and analyzing these aspects has precluded their use in prior studies. We propose DeepSAVE, a novel deep learning framework for detecting adverse events based on user search query logs. DeepSAVE uses an enriched variational autoencoder encompassing a novel query embedding and user modeling module that work in concert to address the context challenge associated with search-based detection of adverse events. Evaluation results on three large real-world event datasets show that DeepSAVE outperforms existing detection methods as well as comparison deep learning auto encoders. Ablation analysis reveals that each component of DeepSAVE significantly contributes to its overall performance. Collectively, the results demonstrate the viability of the proposed architecture for detecting adverse events from search query logs.

Answering recreational web searches with relevant things to do results

Recreational queries from users searching for places to go and things to do or see are very common in web and mobile search. Users specify constraints for what they are looking for, like suitability for kids, romantic ambiance or budget. Queries like “restaurants in New York City” are currently served by static local results or the thumbnail carousel. More complex queries like “things to do in San Francisco with kids” or “romantic places to eat in Seattle” require the user to click on every element of the search engine result page to read articles from Yelp, TripAdvisor, or WikiTravel to satisfy their needs. Location data, which is an essential part of web search, is even more prevalent with location-based social networks and offers new opportunities for many ways of satisfying information seeking scenarios.In this paper, we address the problem of recreational queries in information retrieval and propose a solution that combines search query logs with LBSNs data to match user needs and possible options. At the core of our solution is a framework that combines social, geographical, and temporal information for a relevance model centered around the use of semantic annotations on Points of Interest with the goal of addressing these recreational queries. A central part of the framework is a taxonomy derived from behavioral data that drives the modeling and user experience. We also describe in detail the complexity of assessing and evaluating Point of Interest data, a topic that is usually not covered in related work, and propose task design alternatives that work well.We demonstrate the feasibility and scalability of our methods using a data set of 1B check-ins and a large sample of queries from the real-world. Finally, we describe the integration of our techniques in a commercial search engine.

Coverage pattern based framework to improve search engine advertising

Sponsored search has emerged as one of the most dominant forms for advertising on the Web. In sponsored search, advertisers create ad campaigns and bid on the keywords of potential search queries related to a given product or service. It has been observed that search queries follow a long-tail distribution of a small yet fat head of frequent queries and a long and thin tail of infrequent queries. Normally, the advertisers tend to bid on frequent keywords related to search queries. As a result, the ad space of the tail portion of search queries is harder to exploit. In this paper, we have proposed an improved allocation approach to utilize the ad space of the tail keywords related to search queries based on the knowledge of coverage patterns extracted from the transactions formed from search query logs. The advertisers bid on potential concepts represented by coverage patterns which consist of a combination of head and tail keywords. By facilitating the advertisers to bid on the concepts, the proposed approach improves the ad space utilization of tail queries. Experiments on the real-world dataset of search query logs demonstrate that the proposed approach indeed improves the performance of search engine advertising by improving ad space utilization of tail queries.

Modeling behaviors and lifestyle with online and social data for predicting and analyzing sleep and exercise quality

While recent data studies have focused on associations between sleep and exercise patterns as captured by digital fitness devices, it is known that sleep and exercise quality are affected by a much broader set of factors not captured by these devices, such as general lifestyle, eating, and stress. Here, we conduct a large-scale data study of exercise and sleep effects through an analysis of 8 months of exercise and sleep data for 20 k users, combined with search query logs, location information and aggregated social media data. We analyze factors correlated with better sleep and more effective exercise, and confirm these relationships through causal inference analysis. Further, we build linear models to predict individuals’ sleep and exercise quality. This analysis demonstrates the potential benefits of combining online and social data sources with data from health trackers, and is a potentially rich computational benchmark for health studies. We discuss the implications of our work for individuals, health practitioners and health systems.

The Characteristics of Voice Search

The growing popularity of mobile search and the advancement in voice recognition technologies have opened the door for web search users to speak their queries rather than type them. While this kind of voice search is still in its infancy, it is gradually becoming more widespread. In this article, we report a comprehensive voice search query log analysis of a commercial web search engine’s mobile application. We compare voice and text search by various aspects, with special focus on the semantic and syntactic characteristics of the queries. Our analysis suggests that voice queries focus more on audio-visual content and question answering and less on social networking and adult domains. In addition, voice queries are more commonly submitted on the go. We also conduct an empirical evaluation showing that the language of voice queries is closer to natural language than the language of text queries. Our analysis points out further differences between voice and text search. We discuss the implications of these differences for the design of future voice-enabled web search tools.

A framework for personalized adaptive user interest prediction based on topic model and forgetting mechanism

User interest is not static and changes dynamically. In the scenario of a search engine, this paper presents a personalized adaptive user interest prediction framework. It represents user interest as a topic distribution, captures every change of user interest in the history, and uses the changes to predict future individual user interest dynamically. More specifically, it first uses a personalized user interest representation model to infer user interest from queries in the user’s history data using a topic model; then it presents a personalized user interest prediction model to capture the dynamic changes of user interest and to predict future user interest by leveraging the query submission time in the history data. Compared with the Interest Degree Multi-Stage Quantization Model, experiment results on an AOL Search Query Log query log show that our framework is more stable and effective in user interest prediction.

Don't Mention It? Analyzing User-Generated Content Signals for Early Adverse Drug Event Warnings

Post-marketing surveillance entails monitoring, assessing, and detecting adverse events once drugs are approved and released into the market. User-generated content channels such as social media and search query logs are increasingly being leveraged as complementary data sources to traditional databases for post-marketing drug surveillance. However, the existing body of knowledge has leveraged diverse sets of channels, adverse event types, and modeling methods, resulting in varying results and diverging conclusions regarding the viability and efficacy of various online user-generated channels and accompanying modeling methods. The objective of this study is to examine the efficacy and impact of different online user-generated content channels, event characteristics, and event modeling strategies on early detection of adverse drug events. We incorporate a large test bed encompassing millions of tweets, forums postings, and search query logs pertaining to 143 adverse events. We also propose a novel heuristic-based event modeling method capable of improving precision, recall, and timeliness of alerts. Preliminary results shed light on the interplay between user-generated channels and event types, as well as the potential for more robust event modeling methods that go beyond basic mention models. Several current and future research directions are also discussed. The preliminary results reported have important implications for various stakeholder groups, including regulatory agencies, post-marketing monitoring teams, healthcare hedge fund managers, and third-party consumer advocacy groups.

Search and Breast Cancer

We seek to understand the evolving needs of people who are faced with a life-changing medical diagnosis based on analyses of queries extracted from an anonymized search query log. Focusing on breast cancer, we manually tag a set of Web searchers as showing patterns of search behavior consistent with someone grappling with the screening, diagnosis, and treatment of breast cancer. We build and apply probabilistic classifiers to detect these searchers from multiple sessions and to identify the timing of diagnosis using temporal and statistical features. We explore the changes in information seeking over time before and after an inferred diagnosis of breast cancer by aligning multiple searchers by the estimated time of diagnosis. We employ the classifier to automatically identify 1,700 candidate searchers with an estimated 90% precision, and we predict the day of diagnosis within 15 days with an 88% accuracy. We show that the geographic and demographic attributes of searchers identified with high probability are strongly correlated with ground truth of reported incidence rates. We then analyze the content of queries over time for inferred cancer patients, using a detailed ontology of cancer-related search terms. The analysis reveals the rich temporal structure of the evolving queries of people likely diagnosed with breast cancer. Finally, we focus on subtypes of illness based on inferred stages of cancer and show clinically relevant dynamics of information seeking based on the dominant stage expressed by searchers.

Dynamic Query Intent Prediction from a Search Log Stream

The users that used search engines are obligated to express their goals in few words (queries). Sometimes search queries are ambiguous. Moreover, the users' intents are dynamically evolving. This paper analyzes the user's query logs to classify the related queries, the related intent topic categories and the related intent types and use this classification to dynamically predict the users' future queries, its intent topic and its intent type. AOL Search Query Log is taken as an experimental data set. Then use evaluation metrics to evaluate the prediction results.

Query intent inference via search engine log

Mining the latent intents behind search queries is critical for contemporary search engines. Therefore, there has been lots of effort on studying how to infer the intents of search queries via search engine query log. However, the task of query log-based intent inference is not trivial, since it involves cross-disciplinary knowledge of data modeling and data mining. In this paper, we tackle the problem of query intent inference by integrating multiple information sources in a seamless manner. We first propose a comprehensive data model called Search Query Log Structure (SQLS) that represents the relation between search queries via the User dimension, the URL dimension, the Session dimension and the Term dimension. In order to explore the effective ways of using such multidimensional information modeled by SQLS, we survey and compare three frameworks, namely Result-Oriented Framework, Laplacian-Oriented Framework and Topic-Oriented Framework, to infer the intents of search queries. Experimental results show that the three frameworks significantly outperform the state-of-the-art approach and meet the diverse requirements arising from different application scenarios.

Lean Marketing: Know who not to advertise to!

Internet users search the web for more and more relevant information, such as where to find the best online deals, where to go for vacation, what to wear to a wedding, and many others. Online search advertisers target these users based on their search queries. Targeting the right set of users is important as each user clicking on an online search ad costs money to the advertiser. Savvy advertisers usually pick keywords that are niche and “non-intuitive” because such keywords are cheaper than more popular keywords due to lesser number of advertisers competing for them. In order to find such keywords, multiple information sources such as web ontologies open to the public, search engine results, and search query logs were mined in the past. In these studies, the universe of all relevant keywords was approximated as the relevant keywords of all techniques combined. A given keyword selection technique was then evaluated based on the relevance of the keywords it selected to the target page that is to be advertised. Since the universe of relevant keywords was approximate, it had implications for advertising efficiency. For the campaign dataset we used in this study, out of a total of $75K spent on the campaign, 10% of this total budget was spent on the irrelevant queries. Our primary goal in this work is to reduce this excess budget spent on the irrelevant queries. For this purpose, we learn the characteristics of the “irrelevant” queries using the campaign’s past performance data. Words and phrases that occur frequently in queries, which are not likely to convert, are used as a proxy to decide who not to advertise to. In this way, we plan to preserve more of the marketing budget for queries that are more likely to convert. In our empirical study, we used a logistic regressor for identifying negative words, which distinguished the non-converting queries from the converting ones, and pruned the non-converting queries using the negatives identified. Our approach decreased the average customer acquisition cost of a live search campaign by 25% and cut the overall campaign spend by 28.6%.