On the Study of Transformers for Query Suggestion

When conducting a search task, users may find it difficult to articulate their need, even more so when the task is complex. To help them complete their search, search engine usually provide query suggestions. A good query suggestion system requires to model user behavior during the search session. In this article, we study multiple Transformer architectures applied to the query suggestion task and compare them with recurrent neural network (RNN)-based models. We experiment Transformer models with different tokenizers, with different Encoders (large pretrained models or fully trained ones), and with two kinds of architectures (flat or hierarchic). We study the performance and the behaviors of these various models, and observe that Transformer-based models outperform RNN-based ones. We show that while the hierarchical architectures exhibit very good performances for query suggestion, the flat models are more suitable for complex and long search tasks. Finally, we investigate the flat models behavior and demonstrate that they indeed learn to recover the hierarchy of a search session.

Typo handling in searching of Quran verse based on phonetic similarities

The Quran search system is a search system that was built to make it easier for Indonesians to find a verse with text by Indonesian pronunciation, this is a solution for users who have difficulty writing or typing Arabic characters. Quran search system with phonetic similarity can make it easier for Indonesian Muslims to find a particular verse. Lafzi was one of the systems that developed the search, then Lafzi was further developed under the name Lafzi+. The Lafzi+ system can handle searches with typo queries but there are still fewer variations regarding typing error types. In this research Lafzi++, an improvement from previous development to handle typographical error types was carried out by applying typo correction using the autocomplete method to correct incorrect queries and Damerau Levenshtein distance to calculate the edit distance, so that the system can provide query suggestions when a user mistypes a search, either in the form of substitution, insertion, deletion, or transposition. Users can also search easily because they use Latin characters according to pronunciation in Indonesian. Based on the evaluation results it is known that the system can be better developed, this can be seen from the accuracy value in each query that is tested can surpass the accuracy of the previous system, by getting the highest recall of 96.20% and the highest Mean Average Precision (MAP) reaching 90.69%. The Lafzi++ system can improve the previous system.

An investigation of biases in web search engine query suggestions

Purpose Survey-based studies suggest that search engines are trusted more than social media or even traditional news, although cases of false information or defamation are known. The purpose of this paper is to analyze query suggestion features of three search engines to see if these features introduce some bias into the query and search process that might compromise this trust. The authors test the approach on person-related search suggestions by querying the names of politicians from the German Bundestag before the German federal election of 2017. Design/methodology/approach This study introduces a framework to systematically examine and automatically analyze the varieties in different query suggestions for person names offered by major search engines. To test the framework, the authors collected data from the Google, Bing and DuckDuckGo query suggestion APIs over a period of four months for 629 different names of German politicians. The suggestions were clustered and statistically analyzed with regards to different biases, like gender, party or age and with regards to the stability of the suggestions over time. Findings By using the framework, the authors located three semantic clusters within the data set: suggestions related to politics and economics, location information and personal and other miscellaneous topics. Among other effects, the results of the analysis show a small bias in the form that male politicians receive slightly fewer suggestions on “personal and misc” topics. The stability analysis of the suggested terms over time shows that some suggestions are prevalent most of the time, while other suggestions fluctuate more often. Originality/value This study proposes a novel framework to automatically identify biases in web search engine query suggestions for person-related searches. Applying this framework on a set of person-related query suggestions shows first insights into the influence search engines can have on the query process of users that seek out information on politicians.