Customer journey analysis is a hot topic in marketing. Understanding how the customers behave is crucial and is considered as one of the key drivers of business success. To the best of our knowledge, a data-driven approach to analyze the customer journey is still missing. For instance, web analytics tools like Google Analytics provide an oversimplified version of the user behavior, focusing more on the frequency of page visits rather than discovering the process of the visit itself. On the other hand, customer journey maps have shown their use- fulness, but they need to be created manually by domain experts. This thesis contributes a novel approach for applying process mining techniques to web log customer journey analysis. Through process mining it is possible to (i) discover the process that better describes the user behavior, (ii) find useful insights, (iii) discover and compare the processes of different behavioural clusters of users. Moreover, this thesis aims to make a second contribution by bridging the gap between process mining and recommender systems worlds. Indeed, recommender systems techniques provide personalized recommendations to users based on their behaviour and tailor the customer journey on their preferences. This concept is tightly linked with the process mining world, which allows to visualize, analyze and improve the customer journey. Consequently, this thesis proposes a novel formalization to ensemble these two worlds, in order to use recommender system algorithms in the process mining scenario. Firstly, through pro- cess mining, it is possible to identify particular customer journey paths that can be enforced to optimize some KPIs (Key Performance Indicators). Then, with recommender systems algorithms, it is possible to recommend to users particular actions that will optimize the selected KPIs, using the customer journey as an implicit user feedback. Finally, the results can be analyzed to discover the new customer journey model and check if the recommendations were actually able to divert the journey in the desired way. The proof of the correctness of the introduced concepts is demonstrated through a real-life case study, by showing and evaluating the discovered process models from a real web log, and then using the extracted information from the process models to select and optimize a KPI via personalized recommendations. In particular, including event sequence as additional information, we were able to increase the recommendation click trough rate of 10 % compared to state-of-the-art collaborative filtering algorithms. A position paper [61], containing the preliminary results of this thesis, has been accepted and published in IEEE 6th International Conference on Future Internet of Things and Cloud (FiCloud-2018), 6-8 August 2018, Barcelona, Spain. Moreover, another paper, containing the complete results of this thesis, has been sent to The 34th ACM/SIGAPP Symposium On Applied Computing: SAC 2019, 8-12 April 2019, Cyprus.
Oggigiorno, il customer journey è considerato uno degli hot topics nel mondo del marketing. Difatti, capire come i customers si comportano all’interno di un sito web è cruciale ed è considerato uno dei fattori più importanti per il successo di un web business. Nella letteratura, un approccio data-driven per analizzare il customer journey non è ancora stato sviluppato. Per esempio, tools di web analytics come Google Analytics forniscono una versione troppo semplicistica dello user behaviour e non sono in grado di estrarre un process model che descriva il customer journey nella sua interezza. Di conseguenza, questa tesi propone un approccio innovativo che utilizza tecniche di process mining per analizzare il customer journey estratto da un web log. Infatti, tramite il process mining è possibile: (i) scoprire il process model che descrive meglio lo user behaviour, (ii) estrarre informazioni utili da questo modello e (ii) scoprire e comparare i comportamenti di diversi behavioural users clusters. Inoltre, questa tesi fornisce un secondo contributo cercando di fare da ponte tra il mondo process mining e quello recommender systems. Difatti, le tecniche di recommender systems forniscono raccomandazioni personalizzate agli utenti basandosi sul loro comportamento, personalizzando così il journey in base alle loro preferenze. Questo concetto è strettamente legato alle tecniche di process mining, che permettono di visualizzare, analizzare e ottimizzare il customer journey. Di conseguenza, questa tesi propone un innovativo framework in grado di unire questi due mondi, formalizzando l’uso di algoritmi di raccomandazione in un contesto di process mining analysis. Per prima cosa, tramite tecniche di process mining, è possibile identificare dei particolari customer journey paths che possono essere indotti per ottimizzare una o più KPIs (Key Performance Indicators). In seguito, considerando il customer journey come un feedback implicito fornito dall’utente, tramite algoritmi di recommender systems è possibile raccomandare agli utenti delle particolari azioni che ottimizzino la KPI selezionata precedentemente. Infine, i risultati possono essere analizzati nuovamente con tecniche di process mining, controllando che le raccomandazioni fornite abbiano effettivamente cambiato il customer journey nella maniera desiderata. Per dimostrare la validità dei concetti proposti, il framework è stato testato su un real case study: partendo da un web log reale, diversi process models sono stati estratti e testati; successivamente, le informazioni ottenute sono state usate per selezionare e ottimizzare una KPI tramite raccomandazioni personalizzate. In particolare, includendo il concetto di sequenza negli algoritmi di raccomandazione, siamo riusciti a migliorare il CTR (click through rate) del 10% rispetto agli algoritmi che caratterizzano lo stato dell’arte. Un position paper, contenente i risultati preliminari di questa tesi, è stato pubblicato nella IEEE 6th International Conference on Future Internet of Things and Cloud (FiCloud-2018), 6-8 Agosto 2018, Barcellona, Spagna. Inoltre, un altro paper, contenente i risultati completi, è stato inviato al 34th ACM/SIGAPP Symposium On Applied Computing: SAC 2019, 8-12 Aprile 2019, Cipro.
Analyzing customer journey with process mining : from discovery to recommendations
TERRAGNI, ALESSANDRO
2017/2018
Abstract
Customer journey analysis is a hot topic in marketing. Understanding how the customers behave is crucial and is considered as one of the key drivers of business success. To the best of our knowledge, a data-driven approach to analyze the customer journey is still missing. For instance, web analytics tools like Google Analytics provide an oversimplified version of the user behavior, focusing more on the frequency of page visits rather than discovering the process of the visit itself. On the other hand, customer journey maps have shown their use- fulness, but they need to be created manually by domain experts. This thesis contributes a novel approach for applying process mining techniques to web log customer journey analysis. Through process mining it is possible to (i) discover the process that better describes the user behavior, (ii) find useful insights, (iii) discover and compare the processes of different behavioural clusters of users. Moreover, this thesis aims to make a second contribution by bridging the gap between process mining and recommender systems worlds. Indeed, recommender systems techniques provide personalized recommendations to users based on their behaviour and tailor the customer journey on their preferences. This concept is tightly linked with the process mining world, which allows to visualize, analyze and improve the customer journey. Consequently, this thesis proposes a novel formalization to ensemble these two worlds, in order to use recommender system algorithms in the process mining scenario. Firstly, through pro- cess mining, it is possible to identify particular customer journey paths that can be enforced to optimize some KPIs (Key Performance Indicators). Then, with recommender systems algorithms, it is possible to recommend to users particular actions that will optimize the selected KPIs, using the customer journey as an implicit user feedback. Finally, the results can be analyzed to discover the new customer journey model and check if the recommendations were actually able to divert the journey in the desired way. The proof of the correctness of the introduced concepts is demonstrated through a real-life case study, by showing and evaluating the discovered process models from a real web log, and then using the extracted information from the process models to select and optimize a KPI via personalized recommendations. In particular, including event sequence as additional information, we were able to increase the recommendation click trough rate of 10 % compared to state-of-the-art collaborative filtering algorithms. A position paper [61], containing the preliminary results of this thesis, has been accepted and published in IEEE 6th International Conference on Future Internet of Things and Cloud (FiCloud-2018), 6-8 August 2018, Barcelona, Spain. Moreover, another paper, containing the complete results of this thesis, has been sent to The 34th ACM/SIGAPP Symposium On Applied Computing: SAC 2019, 8-12 April 2019, Cyprus.| File | Dimensione | Formato | |
|---|---|---|---|
|
2018_09_Terragni.pdf
accessibile in internet per tutti
Descrizione: Thesis text
Dimensione
9.15 MB
Formato
Adobe PDF
|
9.15 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/10589/142944