A Topic Modelling Based Approach Towards Personalized and Health-Aware Food Recommendation
|Title:||A Topic Modelling Based Approach Towards Personalized and Health-Aware Food Recommendation||Authors:||Khan, Mansura||Permanent link:||http://hdl.handle.net/10197/12884||Date:||2022||Online since:||2022-05-16T10:32:21Z||Abstract:||In this thesis, we present our research addressing various food-domain specific challenges. The overall aim is to produce systems that support personalized, health-aware, and context-aware Food Recommendation (FR). Chapter 2 describes a systematic literature review that identifies the core challenges in FR research and summarizes the current state-of-the-art. To support our FR research, we created two large-scale recipe corpora with 230,876 recipes and 55,314 recipes, respectively. Chapter 3 summarizes the corpus generation process and describes various properties of each corpus. In chapter 4, we describe research on identifying significant food features, which are multi-domain attributes that have an impact on peoples' eating habits or food choices. We investigated Ensemble Topic Modeling (EnsTM), a natural language processing approach, to parse large recipe corpora and extract dominant food features. Chapter 4 also describes our proposed feature-vector based data representation format for food-items. The feature-vector based format reduces the data volume and computation complexity of recommenders. Finally, the chapter discusses an intelligent, interactive, and open user-modelling technique, which is built on the identified food features. To achieve a meaningful impact on a user's eating habit FRS needs to be useful (e.g., enable personalization) in both cold-start and on going use scenarios. Both are addressed in this thesis. In chapter 6, we describe research on achieving personalization in a cold-start scenario. Taking advantage of the identified food features, the feature-vector based data representation format, and the feature-based intelligent user-modelling, we designed three novel EnsTM based recommenders. To assess their effectiveness in a cold-start scenario the EnsTM based variations were evaluated through a user study, with 48 participants, comparing these against a conventional Content Based (CB) approach. The EnsTM based recommenders performed significantly better than the CB approach. Longer-term use is addressed in chapter 7. To ensure such use FRS need strategies to progressively learn users' preferences, identify changes in eating habit, and accommodate these changes in the future FR. In chapter 7, we discuss seven unique hybrid feature and/or topic based recommenders which progressively learn user-preference from users' interactions with the system. Through an offline study we compared the proposed hybrid recommenders with seminal baselines. Six of our proposed models outperformed the baseline models. Chapter 8 reports two complementary experiments that investigate the necessity and impact of smart-nudging in FR. The first experiment investigates how knowledgeable people are on the healthiness of different commonly consumed food-items?. During a user study with 52 participants we found that people typically have very poor judgement on the nutritional contents contained within various food-items. This inspired us to investigate tools that can better convey a recipe's healthiness to users. Our proposed solution is a combination of EnsTM based food-type identification and smart-nudging techniques to promote healthier food choices among users. We proposed three visual smart-health-nudging techniques to promote healthier food options over others. We conducted a user study with 72 participants comparing recommendation scenarios with each of the three novel nudging techniques and a baseline with no nudging present. Results demonstrated that users are more likely to visit and consume healthier recipes under the influence of visual nudging content. Overall the results in this thesis demonstrate that the proposed EnsTM based recommendation approach performs better than the previous state-of-the-art, and can support FR in both a cold-start and ongoing use scenarios. It also provided evidence that these approaches can be combined with visual smart-nudge techniques to support healthier food choices.||Funding Details:||Science Foundation Ireland||Type of material:||Doctoral Thesis||Publisher:||University College Dublin. School of Computer Science||Qualification Name:||Ph.D.||Copyright (published version):||2022 the Author||Keywords:||Recommender system; Machine learning; Health-aware food recommendation; Topic modeling||Language:||en||Status of Item:||Peer reviewed||This item is made available under a Creative Commons License:||https://creativecommons.org/licenses/by-nc-nd/3.0/ie/|
|Appears in Collections:||Computer Science Theses|
Show full item record
If you are a publisher or author and have copyright concerns for any item, please email email@example.com and the item will be withdrawn immediately. The author or person responsible for depositing the article will be contacted within one business day.