This thesis is about the development of a technology to make feasible and viable the creation of conversational interfaces for improving learning processes. It is based on a two-fold research question: A) are empathic conversational interfaces a valid mean to support learners? B) is it possible to develop a sustainable, cost-effective conversational technology? (not only for educational conversations but also in general sense). As far as “A” is concerned, this thesis is part of a larger research effort (iMOOC), that postulates the quest for “adaptive education” tailoring each learning experience upon the specific needs of each learner. Adaptivity involves the amount of content, the level of difficulty of the content, the “kind” of content, and the arrangement of content items. The contribution of this thesis work is to enrich this approach with a conversational interface: a “tutoring chatbot”. The chatbots can engage learners in conversations, helping them through specifically tailored learning paths. Elaborating the “B” research question, the work for this thesis has developed a Data-driven Conversational Technology for Educational Chatbots (DCTEC). “Data-driven” means that “ properly tagged content” and “configuration tables” control a number of features, including how the chatbot speaks and how to interpret the utterances of the learner (user). This technology enables content and conversation experts to design and deploy chatbot’s conversational logic, without the involvement of ICT specialists. This possibility reduces (at some degree) the chatbot’s production/maintenance costs, as well as drawing new expertise into the task of educational chatbot production. Both research questions are addressed, in this thesis, in a rather general sense; but in order to ground the work upon suitable requirements, a real-life test case has been tackled: a MOOC (about “recommender systems”) developed by Politecnico di Milano, and derived by a course currently being offered at graduate level. The course consists of 8 modules for a total of 230 video lessons. The DCTEC technology uses a conversation meta-model for modeling time-extended, proactive, conversations. The meta-model focuses on defining the chatbot’s wording and the conversation flow. The DCTEC technology compiles and deploys a conversation, defined with the mean of intuitive spreadsheet tables, to the IBM Watson Assistant cloud service. This compiler is designed to be easily extensible to meet changes in the output format. This is achieved using a domain specific language with a powerful macro system to encode the compilation’s logic. An important result of DCTEC is that it allows the development of “adaptive conversations”: the chat adapts its turns and its wording to different learner’s stereotypes, and collects feedback from the learner about the course’s material. A second crucial result (concerning “B” research question) is that a similar adaptive conversation can be developed (at a negligible cost) for different courses. The chatbot developed with DCTEC has been integrated into a larger environment, i.e. an educational platform for adaptive courses’ fruition, and deployed in the cloud. The resulting platform can be used to deploy new courses using the same chatbot tutor. This thesis also investigates the authoring workflow to create new courses, as well as to customize/improve the conversation.
Il lavoro di questa tesi riguarda lo sviluppo di una tecnologia abilitante, allo scopo di rendere fattibile e sostenibile l’uso di interfacce conversazionali per migliorare il processo d’apprendimento. La domanda di ricerca é doppia: A)le interfacce conversazionali empatiche possono rappresentare un valido metodo per supportare lo studente ? B) é possibile sviluppare una tecnologia conversazionale sostenibile ed economicamente vantaggiosa ? (non limitata a conversazioni in campo educativo, ma in senso generale). Per quanto riguarda la domanda “A”, questa tesi é parte di un piú largo sforzo di ricerca (iMOOC), orientato al perseguimento della “educazione adattativa”, personalizzando l’esperienza di apprendimento sugli specifici requisiti del singolo studente. L’adattativitá in questione coinvolge la quantitá di contenuti, il loro livello di difficoltá, il “colore” (tipologia) del contenuto, e l’organizzazione dei vari atomi di contenuto. Il contributo di questo lavoro di tesi consiste nell’arricchire tale approccio con un’interfaccia conversazionale: un “chatbot tutore”. Il chatbot é in grado di coinvolgere gli studenti nelle conversazioni, guidandoli attraverso percorsi di apprendimento specificatamente personalizzati. Elaborando la domanda di ricerca “B”, questa tesi ha sviluppato una tecnologia conversazionale (DCTEC), guidata da dati, per lo sviluppo di chatbot educativi. Per “guidata dai dati” (data-driven) si intende che “contenuti propriamente annotati” e “tabelle di configurazione” controllano una serie di funzionalitá, tra cui il modo in cui il chatbot parla e quello in cui interpreta le frasi dello studente (utente). Questa technologia abilita gli esperti dei contenuti e della conversazione alla progettazione e allo sviluppo di logiche conversazionali per chatbot, senza richiedere l’intervento di specialisti informatici. Questo presenta il potenziale per ridurre (in una certa entitá) i costi di sviluppo e manutenzione dei chatbot, oltre ad attirare nuovi tipi di esperti nella produzione di chatbot educativi. Entrambe queste domande di ricerca sono state affrontate, nella tesi, in senso generale; tuttavia allo scopo di basare il lavoro su requisiti corretti, un caso d’uso reale é stato affrontato: un MOOC (sui “Sistemi di raccomandazione”) sviluppato dal Politecnico di Milano, derivante da un corso attualmente offerto a livello universitario. Il corso é costituito di 8 moduli, per un totale di 230 video lezioni. La tecnologia DCTEC sfrutta un metamodello per rappresentare conversazioni proattive ed estese temporalmente. Tale metamodello si concentra sulla definizione del flusso della conversazione e delle “formulazioni linguistiche” del chatbot. Questa tecnologia é utilizzata per compilare ed eseguire conversazioni, definite attraverso intuitive tabelle in Excel, sfruttando il servizio cloud di “IBM Watson Asssistant” per la loro esecuzione. Tale compilatore é stato progettato per essere facilmente estensibile a fronte di cambiamenti nel formato di uscita. Questo é reso possibile da un linguaggio di dominio specifico (DSL) dotato di un sistema di metaprogrammazione (macro), usato per definire la logica di compilazione. Un importante risultato di DCTEC consiste nell’abilitare lo sviluppo di “conversazioni adattative”: il chatbot adatta i suoi turni e il suo modo di parlare ai diversi stereotipi di studente, oltre a raccogliere commenti dagli studenti riguardo il materiale del corso. Un secondo risultato cruciale (riguardo la domanda “B”) é rappresentato dal fatto che una conversazione adattativa simile puó essere sviluppata (a costi trascurabili) per corsi differenti. Il chatbot sviluppato con DCTEC é stato integrato in un ambiente piú ampio, nella fattispecie, una piattaforma educativa per la fruizione di corsi adattativi, e reso disponibile pubblicamente in cloud. La piattaforma risultante puó essere utilizzata per pubblicare nuovi corsi usando lo stesso chatbot tutore. Questa tesi investiga inoltre lo sforzo editoriale richiesto agli autori per creare nuovi corsi, cosí come per personalizzare/migliorare la conversazione.
Adaptive conversations for education supported by data-driven chatbots
LODI, LUCA
2017/2018
Abstract
This thesis is about the development of a technology to make feasible and viable the creation of conversational interfaces for improving learning processes. It is based on a two-fold research question: A) are empathic conversational interfaces a valid mean to support learners? B) is it possible to develop a sustainable, cost-effective conversational technology? (not only for educational conversations but also in general sense). As far as “A” is concerned, this thesis is part of a larger research effort (iMOOC), that postulates the quest for “adaptive education” tailoring each learning experience upon the specific needs of each learner. Adaptivity involves the amount of content, the level of difficulty of the content, the “kind” of content, and the arrangement of content items. The contribution of this thesis work is to enrich this approach with a conversational interface: a “tutoring chatbot”. The chatbots can engage learners in conversations, helping them through specifically tailored learning paths. Elaborating the “B” research question, the work for this thesis has developed a Data-driven Conversational Technology for Educational Chatbots (DCTEC). “Data-driven” means that “ properly tagged content” and “configuration tables” control a number of features, including how the chatbot speaks and how to interpret the utterances of the learner (user). This technology enables content and conversation experts to design and deploy chatbot’s conversational logic, without the involvement of ICT specialists. This possibility reduces (at some degree) the chatbot’s production/maintenance costs, as well as drawing new expertise into the task of educational chatbot production. Both research questions are addressed, in this thesis, in a rather general sense; but in order to ground the work upon suitable requirements, a real-life test case has been tackled: a MOOC (about “recommender systems”) developed by Politecnico di Milano, and derived by a course currently being offered at graduate level. The course consists of 8 modules for a total of 230 video lessons. The DCTEC technology uses a conversation meta-model for modeling time-extended, proactive, conversations. The meta-model focuses on defining the chatbot’s wording and the conversation flow. The DCTEC technology compiles and deploys a conversation, defined with the mean of intuitive spreadsheet tables, to the IBM Watson Assistant cloud service. This compiler is designed to be easily extensible to meet changes in the output format. This is achieved using a domain specific language with a powerful macro system to encode the compilation’s logic. An important result of DCTEC is that it allows the development of “adaptive conversations”: the chat adapts its turns and its wording to different learner’s stereotypes, and collects feedback from the learner about the course’s material. A second crucial result (concerning “B” research question) is that a similar adaptive conversation can be developed (at a negligible cost) for different courses. The chatbot developed with DCTEC has been integrated into a larger environment, i.e. an educational platform for adaptive courses’ fruition, and deployed in the cloud. The resulting platform can be used to deploy new courses using the same chatbot tutor. This thesis also investigates the authoring workflow to create new courses, as well as to customize/improve the conversation.| File | Dimensione | Formato | |
|---|---|---|---|
|
000_thesis.pdf
non accessibile
Descrizione: Thesis text
Dimensione
2.07 MB
Formato
Adobe PDF
|
2.07 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/150427