{"id":{"repo_id":"mcmaster","oai_identifier":"oai:macsphere.mcmaster.ca:11375/33296"},"canonical_url":"https://search.ndltd.org/etd/mcmaster/oai:macsphere.mcmaster.ca:11375/33296","repository":{"repo_id":"mcmaster","name":"McMaster University","base_url":"https://macsphere.mcmaster.ca/server/oai/request"},"display":{"title":"Evaluation of LLM-driven Data Preparation","abstract":"Large language models (LLMs) have increasingly been used in data preparation pipelines due to their strong natural language understanding and reasoning capabilities. Data preparation tasks such as entity matching, and missing value imputation often rely on identifying syntactic patterns in the data, and domain-specific, contextual interpretations, the latter being a differentiating strength of LLM-based systems. However, these systems are imperfect demonstrating biases and inaccuracies. Studies have shown that 42\\% of model-centric approaches lead to inaccuracies and hallucinations, and 40\\% of models suffer from biases due to incomplete data. Existing work has studied LLM-based data preparation under fixed prompts and parameter configurations. In this paper, we present a systematic, comparative evaluation of LLM-driven entity matching, missing value imputation, and semantic reasoning across four state-of-the-art models. We evaluate the impact of various parameters including training example quantity, class ratios, input record lengths, missing values proportions, and the utility of key attribute values. Our results show the sensitivities and performance of these models using a suite of real data collections.","abstract_html":"Large language models (LLMs) have increasingly been used in data preparation pipelines due to their strong natural language understanding and reasoning capabilities. Data preparation tasks such as entity matching, and missing value imputation often rely on identifying syntactic patterns in the data, and domain-specific, contextual interpretations, the latter being a differentiating strength of LLM-based systems. However, these systems are imperfect demonstrating biases and inaccuracies. Studies have shown that 42\\% of model-centric approaches lead to inaccuracies and hallucinations, and 40\\% of models suffer from biases due to incomplete data. Existing work has studied LLM-based data preparation under fixed prompts and parameter configurations. In this paper, we present a systematic, comparative evaluation of LLM-driven entity matching, missing value imputation, and semantic reasoning across four state-of-the-art models. We evaluate the impact of various parameters including training example quantity, class ratios, input record lengths, missing values proportions, and the utility of key attribute values. Our results show the sensitivities and performance of these models using a suite of real data collections.","abstract_has_math":false,"creators":["Yaxuan Chen"],"institution":null,"degree_name":null,"degree_level":null,"degree_discipline":null,"degree_department":"Computing and Software","school":null,"contributors":[],"advisors":["Fei, Chiang"],"committee_chairs":[],"committee_members":[],"year":2026,"date_issued":"2026","date_published":"2026","updated_at":"2026-11-18T19:32:28Z","subjects":["data preparation","data-centric AI","data cleaning"],"languages":["en"],"rights":["Attribution 4.0 International"],"rights_urls":["http://creativecommons.org/licenses/by/4.0/"],"identifier_entries":[{"key":"dc:identifier.uri","label":"Identifier URI","values":["https://doi.org/10.71548/158"],"render_values":[{"text":"https://doi.org/10.71548/158","href":"https://doi.org/10.71548/158","code":true}]}],"person_identifier_entries":[]},"links":{"outbound_url":"https://hdl.handle.net/11375/33296","outbound_label":"Handle","outbound_source":"dc:identifier.uri","record_url":"https://hdl.handle.net/11375/33296","record_label":"Repository record","record_kind":"handle","record_source":"dc:identifier.uri","download_url":null,"download_label":null,"download_source":null,"access_status":null,"embargo_until":null,"access_note":null},"source_record":{"url":"https://macsphere.mcmaster.ca/server/oai/request?verb=GetRecord&metadataPrefix=dim&identifier=oai%3Amacsphere.mcmaster.ca%3A11375%2F33296","prefix":"dim"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:contributor.advisor","label":"Advisor","values":["Fei, Chiang"]},{"key":"dc:contributor.department","label":"Department","values":["Computing and Software"]},{"key":"dc:creator","label":"Author","values":["Yaxuan Chen"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date.accessioned","label":"Dc Date Accessioned","values":["2026-06-25T19:36:49Z"]},{"key":"dc:date.issued","label":"Date","values":["2026"]},{"key":"dc:type","label":"Dc Type","values":["Thesis"]}]},{"id":"subjects_keywords","label":"Subjects and Keywords","entries":[{"key":"dc:subject","label":"Dc Subject","values":["data preparation","data-centric AI","data cleaning"]}]},{"id":"language_rights","label":"Language and Rights","entries":[{"key":"dc:language.iso","label":"Language (ISO)","values":["en"]},{"key":"dc:rights","label":"Dc Rights","values":["Attribution 4.0 International"]},{"key":"dc:rights.uri","label":"Rights URI","values":["http://creativecommons.org/licenses/by/4.0/"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier.uri","label":"Identifier URI","values":["https://hdl.handle.net/11375/33296","https://doi.org/10.71548/158"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description.abstract","label":"Abstract","values":["Large language models (LLMs) have increasingly been used in data preparation pipelines due to their strong natural language understanding and reasoning capabilities. Data preparation tasks such as entity matching, and missing value imputation often rely on identifying syntactic patterns in the data, and domain-specific, contextual interpretations, the latter being a differentiating strength of LLM-based systems. However, these systems are imperfect demonstrating biases and inaccuracies. Studies have shown that 42\\% of model-centric approaches lead to inaccuracies and hallucinations, and 40\\% of models suffer from biases due to incomplete data. Existing work has studied LLM-based data preparation under fixed prompts and parameter configurations. In this paper, we present a systematic, comparative evaluation of LLM-driven entity matching, missing value imputation, and semantic reasoning across four state-of-the-art models. We evaluate the impact of various parameters including training example quantity, class ratios, input record lengths, missing values proportions, and the utility of key attribute values. Our results show the sensitivities and performance of these models using a suite of real data collections."]},{"key":"dc:description.degree","label":"Dc Description Degree","values":["Master of Science (MSc)"]},{"key":"dc:title","label":"Title","values":["Evaluation of LLM-driven Data Preparation"]},{"key":"others:access_status","label":"Others Access Status","values":["open.access"]}]}],"canonical_facts":{"dc:contributor.advisor":["Fei, Chiang"],"dc:contributor.department":["Computing and Software"],"dc:creator":["Yaxuan Chen"],"dc:date.accessioned":["2026-06-25T19:36:49Z"],"dc:date.issued":["2026"],"dc:description.abstract":["Large language models (LLMs) have increasingly been used in data preparation pipelines due to their strong natural language understanding and reasoning capabilities. Data preparation tasks such as entity matching, and missing value imputation often rely on identifying syntactic patterns in the data, and domain-specific, contextual interpretations, the latter being a differentiating strength of LLM-based systems. However, these systems are imperfect demonstrating biases and inaccuracies. Studies have shown that 42\\% of model-centric approaches lead to inaccuracies and hallucinations, and 40\\% of models suffer from biases due to incomplete data. Existing work has studied LLM-based data preparation under fixed prompts and parameter configurations. In this paper, we present a systematic, comparative evaluation of LLM-driven entity matching, missing value imputation, and semantic reasoning across four state-of-the-art models. We evaluate the impact of various parameters including training example quantity, class ratios, input record lengths, missing values proportions, and the utility of key attribute values. Our results show the sensitivities and performance of these models using a suite of real data collections."],"dc:description.degree":["Master of Science (MSc)"],"dc:identifier.uri":["https://hdl.handle.net/11375/33296","https://doi.org/10.71548/158"],"dc:language.iso":["en"],"dc:rights":["Attribution 4.0 International"],"dc:rights.uri":["http://creativecommons.org/licenses/by/4.0/"],"dc:subject":["data preparation","data-centric AI","data cleaning"],"dc:title":["Evaluation of LLM-driven Data Preparation"],"dc:type":["Thesis"],"others:access_status":["open.access"]},"updated_at":"2026-11-18T19:32:28Z"}