{"id":{"repo_id":"ets-quebec","oai_identifier":"oai:espace.etsmtl.ca:2650"},"canonical_url":"https://search.ndltd.org/etd/ets-quebec/oai:espace.etsmtl.ca:2650","repository":{"repo_id":"ets-quebec","name":"ETS (Quebec)","base_url":"https://espace.etsmtl.ca/cgi/oai2"},"display":{"title":"Unsupervised abstractive summaries of controllable length","abstract":"The Canadian company Croesus offers a virtual dashboard for financial advisors. Their flagship product, a software called Croesus Advisor, brings together financial portfolio management and customer relationship management (CRM) functionalities. Croesus is leading a Research & Development (R&D) project whose aim is to automate the generation of reports about the performance of financial portfolios over a given period. The generated reports will have to measure and explain variations of portfolio performances with respect to benchmark market(s) using natural language. Such a project is in the domain of Natural Language Generation (NLG). This R&D project involves more than one research problem. First, the time series describing the different assets in a given portfolio must be analyzed in order to identify significant events. Then, the publications describing these events must be identified and retrieved. Finally, the sequence of publications describing significant events should be summarized in a coherent and informative narrative of controllable length. This work focused on a particular aspect of the R&D project, that is, the unsupervised training of automatic abstractive summarization models with the capacity to control their output’s length. Experiments were performed using Recurrent Neural Network and Transformers trained with and without supervision. A Transformer-based model trained without supervision yielded ROUGE-1, ROUGE-2 and ROUGE-L F1 scores of 21.29, 6.4 and 19.41, respectively. Furthermore, a similar model trained in a supervised fashion achieved ROUGE-1, ROUGE-2 and ROUGE-L F1 scores of 40.27, 18.46 and 37.59. To the best of this author’s knowledge, the model trained under supervision constitutes the current state-of-the-art in abstractive summarization of controllable length. These results, alongside research conducted in parallel to this work, mitigated the risk with respect to the automatic summarization aspect of the R&D project to a tolerable level. Work on the remaining research problems of the project is set to start soon.","abstract_html":"The Canadian company Croesus offers a virtual dashboard for financial advisors. Their flagship product, a software called Croesus Advisor, brings together financial portfolio management and customer relationship management (CRM) functionalities. Croesus is leading a Research &amp; Development (R&amp;D) project whose aim is to automate the generation of reports about the performance of financial portfolios over a given period. The generated reports will have to measure and explain variations of portfolio performances with respect to benchmark market(s) using natural language. Such a project is in the domain of Natural Language Generation (NLG). This R&amp;D project involves more than one research problem. First, the time series describing the different assets in a given portfolio must be analyzed in order to identify significant events. Then, the publications describing these events must be identified and retrieved. Finally, the sequence of publications describing significant events should be summarized in a coherent and informative narrative of controllable length. This work focused on a particular aspect of the R&amp;D project, that is, the unsupervised training of automatic abstractive summarization models with the capacity to control their output’s length. Experiments were performed using Recurrent Neural Network and Transformers trained with and without supervision. A Transformer-based model trained without supervision yielded ROUGE-1, ROUGE-2 and ROUGE-L F1 scores of 21.29, 6.4 and 19.41, respectively. Furthermore, a similar model trained in a supervised fashion achieved ROUGE-1, ROUGE-2 and ROUGE-L F1 scores of 40.27, 18.46 and 37.59. To the best of this author’s knowledge, the model trained under supervision constitutes the current state-of-the-art in abstractive summarization of controllable length. These results, alongside research conducted in parallel to this work, mitigated the risk with respect to the automatic summarization aspect of the R&amp;D project to a tolerable level. Work on the remaining research problems of the project is set to start soon.","abstract_has_math":false,"creators":["Gazaille, Stéphane"],"institution":"École de technologie supérieure","degree_name":"Mémoire de maîtrise électronique","degree_level":"masters","degree_discipline":null,"degree_department":null,"school":null,"contributors":[],"advisors":[],"committee_chairs":[],"committee_members":[],"year":2020,"date_issued":"2020-08-17","date_published":"2020-08-17","updated_at":"2026-11-18T22:08:38Z","subjects":["traitement automatique du langage naturel, génération automatique de résumés abstractifs, apprentissage non-supervisé"],"languages":["en"],"rights":[],"rights_urls":[],"identifier_entries":[{"key":"dc:identifier","label":"Identifier","values":["INSERT-YOUR-LIBRARY-ID-HERE2650"],"render_values":[{"text":"INSERT-YOUR-LIBRARY-ID-HERE2650","href":null,"code":true}]}],"person_identifier_entries":[]},"links":{"outbound_url":null,"outbound_label":null,"outbound_source":null,"record_url":null,"record_label":null,"record_kind":null,"record_source":null,"download_url":"https://espace.etsmtl.ca/id/eprint/2650/1/GAZAILLE_Stephane.pdf","download_label":"Download PDF","download_source":"dc:identifier","access_status":null,"embargo_until":null,"access_note":null},"source_record":{"url":"https://espace.etsmtl.ca/cgi/oai2?verb=GetRecord&metadataPrefix=oai_etdms&identifier=oai%3Aespace.etsmtl.ca%3A2650","prefix":"oai_etdms"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:creator","label":"Author","values":["Gazaille, Stéphane"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date","label":"Dc Date","values":["2020-08-17"]},{"key":"dc:publisher","label":"Institution","values":["École de technologie supérieure"]},{"key":"dc:type","label":"Dc Type","values":["Electronic Thesis or Dissertation"]},{"key":"thesis:degree_level","label":"Degree Level","values":["masters"]},{"key":"thesis:degree_name","label":"Degree Name","values":["Mémoire de maîtrise électronique"]},{"key":"thesis:institution_name","label":"Thesis Institution Name","values":["École de technologie supérieure"]}]},{"id":"subjects_keywords","label":"Subjects and Keywords","entries":[{"key":"dc:subject","label":"Dc Subject","values":["traitement automatique du langage naturel, génération automatique de résumés abstractifs, apprentissage non-supervisé"]}]},{"id":"language_rights","label":"Language and Rights","entries":[{"key":"dc:language","label":"Dc Language","values":["en"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier","label":"Identifier","values":["INSERT-YOUR-LIBRARY-ID-HERE2650","https://espace.etsmtl.ca/id/eprint/2650/1/GAZAILLE_Stephane.pdf","https://espace.etsmtl.ca/id/eprint/2650/7/GAZAILLE_Stephane-web.pdf"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description","label":"Description","values":["The Canadian company Croesus offers a virtual dashboard for financial advisors. Their flagship product, a software called Croesus Advisor, brings together financial portfolio management and customer relationship management (CRM) functionalities. Croesus is leading a Research & Development (R&D) project whose aim is to automate the generation of reports about the performance of financial portfolios over a given period. The generated reports will have to measure and explain variations of portfolio performances with respect to benchmark market(s) using natural language. Such a project is in the domain of Natural Language Generation (NLG). This R&D project involves more than one research problem. First, the time series describing the different assets in a given portfolio must be analyzed in order to identify significant events. Then, the publications describing these events must be identified and retrieved. Finally, the sequence of publications describing significant events should be summarized in a coherent and informative narrative of controllable length. This work focused on a particular aspect of the R&D project, that is, the unsupervised training of automatic abstractive summarization models with the capacity to control their output’s length. Experiments were performed using Recurrent Neural Network and Transformers trained with and without supervision. A Transformer-based model trained without supervision yielded ROUGE-1, ROUGE-2 and ROUGE-L F1 scores of 21.29, 6.4 and 19.41, respectively. Furthermore, a similar model trained in a supervised fashion achieved ROUGE-1, ROUGE-2 and ROUGE-L F1 scores of 40.27, 18.46 and 37.59. To the best of this author’s knowledge, the model trained under supervision constitutes the current state-of-the-art in abstractive summarization of controllable length. These results, alongside research conducted in parallel to this work, mitigated the risk with respect to the automatic summarization aspect of the R&D project to a tolerable level. Work on the remaining research problems of the project is set to start soon."]},{"key":"dc:format","label":"Dc Format","values":["application/pdf"]},{"key":"dc:title","label":"Title","values":["Unsupervised abstractive summaries of controllable length"]}]}],"canonical_facts":{"dc:creator":["Gazaille, Stéphane"],"dc:date":["2020-08-17"],"dc:description":["The Canadian company Croesus offers a virtual dashboard for financial advisors. Their flagship product, a software called Croesus Advisor, brings together financial portfolio management and customer relationship management (CRM) functionalities. Croesus is leading a Research & Development (R&D) project whose aim is to automate the generation of reports about the performance of financial portfolios over a given period. The generated reports will have to measure and explain variations of portfolio performances with respect to benchmark market(s) using natural language. Such a project is in the domain of Natural Language Generation (NLG). This R&D project involves more than one research problem. First, the time series describing the different assets in a given portfolio must be analyzed in order to identify significant events. Then, the publications describing these events must be identified and retrieved. Finally, the sequence of publications describing significant events should be summarized in a coherent and informative narrative of controllable length. This work focused on a particular aspect of the R&D project, that is, the unsupervised training of automatic abstractive summarization models with the capacity to control their output’s length. Experiments were performed using Recurrent Neural Network and Transformers trained with and without supervision. A Transformer-based model trained without supervision yielded ROUGE-1, ROUGE-2 and ROUGE-L F1 scores of 21.29, 6.4 and 19.41, respectively. Furthermore, a similar model trained in a supervised fashion achieved ROUGE-1, ROUGE-2 and ROUGE-L F1 scores of 40.27, 18.46 and 37.59. To the best of this author’s knowledge, the model trained under supervision constitutes the current state-of-the-art in abstractive summarization of controllable length. These results, alongside research conducted in parallel to this work, mitigated the risk with respect to the automatic summarization aspect of the R&D project to a tolerable level. Work on the remaining research problems of the project is set to start soon."],"dc:format":["application/pdf"],"dc:identifier":["INSERT-YOUR-LIBRARY-ID-HERE2650","https://espace.etsmtl.ca/id/eprint/2650/1/GAZAILLE_Stephane.pdf","https://espace.etsmtl.ca/id/eprint/2650/7/GAZAILLE_Stephane-web.pdf"],"dc:language":["en"],"dc:publisher":["École de technologie supérieure"],"dc:subject":["traitement automatique du langage naturel, génération automatique de résumés abstractifs, apprentissage non-supervisé"],"dc:title":["Unsupervised abstractive summaries of controllable length"],"dc:type":["Electronic Thesis or Dissertation"],"thesis:degree_level":["masters"],"thesis:degree_name":["Mémoire de maîtrise électronique"],"thesis:institution_name":["École de technologie supérieure"]},"updated_at":"2026-11-18T22:08:38Z"}