A predictive model for recurrent consumption behavior: An application on phone calls
Introduction
With the adoption and the proliferation of e-services, businesses rely on the collected data in order to analyze and understand consumers behaviors. Indeed, understanding consumption patterns helps businesses tailor their services, products and strategies to consumers’ preferences and trends. In addition to personalization, such analysis enable businesses and service providers to anticipate changes in customers’ expectations and consumption patterns and thus to proactively deal with them. In fact, analyzing and modeling individuals’ behaviors and interactions is gaining an increasing interest among industrials, especially in highly concurrent fields where understanding and anticipating consumers preferences and trends is the key of success. In order to achieve this goal, researchers resort to analytical approaches such as datamining, statistical modeling and knowledge based techniques to overcome the uncertainties behind individuals behaviors and decisions. Such approaches were largely adopted in recommender systems in order to analyze customers’ interests and predict their future consumptions and purchase decisions [1], [2]. Nevertheless, other studies may also rely on empirical methodologies for behaviors modeling, simulation and analysis [3].
Some of consumers behaviors are periodic and thus, can be modeled and predicted in order to help businesses plan their offers and future strategies. In fact, modeling and consequently predicting individuals consumption with regard to goods and/or to brands may benefit both businesses and consumers. On the one hand, being able to predict consumers preferences makes businesses able to satisfy their needs with targeted products and personalized communication which increases their loyalty. Besides, modeling users recurrent purchases makes it possible to analyze and predict their loyalty over the time. On the other hand, consumers would benefit from personalized information and offers while spending less effort on searching for relevant content or goods [4]. Recurrent or periodic consumers behaviors can be seen in various domains such as energy consumption, food industry, clothing and communication. For example, consumers visit several types of stores periodically (e.g. coffee shops, restaurants, clothing stores, etc.). Even websites are visited periodically by users searching for information or for items to buy. In fact, some goods are purchased periodically by consumers (clothes, grocery, movies, etc.). Moreover, in the communication field, phone calls are partly recurrent and predictable and individuals may show similar consumption patterns but with some variations due to personal needs, contexts and intentions.
With current advances in communication and smartphones, service providers are looking to offer innovative and useful services for their customers via mobiles. Currently, telecommunication services providers, are becoming more interested in the adoption of recommender systems as a tool that enables them to personalize their products and services based on their customers’ preferences similarly to e-commerce businesses [5]. In this context, recent researches in personalization and recommendation systems are focusing on hybrid approaches that unify existing theories. Hereby, Zhang et al. developed an hybrid recommendation system that helps customers selecting the most suitable mobile products and offerings based on their preferences [5]. Besides products and services recommendation, predicting users’ future phone calls and consumption behaviors can also be seen as a new service that can benefit both individuals and businesses. On the one hand, it enables for instance telecommunication services providers to anticipate their network workload and increase their quality of service. Moreover, companies can use predictive models to forecast demands in order to better allocate their resources and investments. Furthermore, it enables them to analyze their customers communication habits in order to address the consumers churn problem and accordingly better tailor their offers [6]. On the other hand, predicting future phone calls helps customers in planning their communications, anticipating incoming calls and remembering important phone calls.
In this article, we propose a statistical model that predicts, for a given user, his/her future incoming and outgoing calls towards each of his/her contacts based on calls periodicity. The originality of the proposed model comes from its fine grained predictions, autonomy and small footprint. The model is experimented on a real life dataset in order to evaluate its performances and predictions quality in terms of precision and adaptability with the evolution of consumption behaviors.
The remainder of this paper is organized as follows. In Section 2, the main related works and their prediction methodologies are presented. Section 3 discusses our motivations with regard to the calls prediction problematic and enumerates the objectives of the aimed solution. Section 4 presents and formalizes our calls prediction model. The implementation of the proposed model and its architecture are detailed in Section 5. Section 6 details the experimental results of our proposition and compares the model’s performances to several existing works. Finally, in Section 7, we analyze and interpret the obtained numerical results in order to present some insights regarding the different patterns of recurrent phone calls in several cultural and spatio-temporal contexts. The paper is then concluded with several potential extensions and improvements.
Section snippets
Overview of recurrent behaviors prediction
Consumers behaviors models and prediction approaches represent an important decision aid support helping businesses to harness the huge amount of available data and to foresee future consumption demand. In this context, several social and psychology researches are interested in understanding and modeling social interactions and human behaviors patterns [3]. In such researches, the main assumption is that individuals behaviors, interactions and decisions are largely recurrent, context-dependent
Calls prediction motivations and objectives
In this section, first, we present the motivation of our work and enumerate some of the objectives and constraints the calls prediction model should satisfy. Then, we discuss the privacy implications that can arise while mining users phone calls.
Calls prediction model
This section presents our calls prediction model. First, we state and formalize the main problematic of our work which consists in modeling and predicting users future phone calls. Then, we detail the two main steps of our proposed model which are the calls frequencies estimation and the future occurrences prediction.
Calls prediction framework overview
The calls prediction framework can be implemented on the service provider infrastructure in order to better understand and anticipate subscribers trends and behaviors. Moreover, it may serve as a loyalty monitoring system and used to analyze customers churn problem. Besides, when implemented on phone devices, the framework acts as an intelligent phone book reminding users of the numbers that they will probably dial and helping them predict the most probable incoming calls.
In this work, we
Experimentation and evaluation
In this section, we present our experimentation environment and the collected data. Afterwards, the main results of the proposed calls prediction model are presented and discussed.
Insights on recurrent calling patterns
In order to guarantee users privacy, only the prediction errors were collected in our work. Consequently, we are only able to analyze consumers behaviors through the model’s prediction error with regard to the spatial and temporal dimensions.
Fig. 8 represents our model’s prediction errors grouped respectively by country and region. Hereby, data shows that consumer’s phone usages in America are stable whereas those in Europe are highly unpredictable. At a more detailed level, data shows that
Conclusion and future improvements
In this article we presented a statistical model for recurrent consumption behaviors prediction. The model was applied on phone calls prediction aiming at proposing new services for both customers and mobile operators. The model considers both incoming and outgoing calls as inhomogeneous Poisson processes where calls frequencies are variable and dependent on both the caller and the callee. Consequently, in order to anticipate their changes and future trends, calls rates are estimated using an
References (24)
- et al.
A highly adaptive recommender system based on fuzzy logic for {B2C} e-commerce portals
Expert Syst. Appl.
(2011) - et al.
Recommender systems survey
Knowledge-Based Syst.
(2013) - et al.
A hybrid fuzzy-based personalized recommender system for telecom products/services
Inf. Sci.
(2013) - L. Cao, In-depth behavior understanding and use: the behavior informatics approach, Inf. Sci. 180 (17) (2010)...
- M.R. Haddad, H. Baazaoui, D. Ziou, H. Ben Ghezala, Towards a new model for context-aware recommendation, in: 6th IEEE...
- et al.
Social ties and their relevance to churn in mobile telecom networks
- W. Pan, N. Aharony, A. Pentland, Composite Social Network for Predicting Mobile Apps Installation, CoRR...
- et al.
Predicting spending behavior using socio-mobile features
- et al.
Reality mining: sensing complex social systems
Pers. Ubiquitous Comput.
(2006) - M. Azam, J. Loo, A. Lasebae, S. Khan, W. Ejaz, Behavioural analysis of low entropy mobile people using contextual...
Statistical analysis of a telephone call center: a queuing-science perspective
J. Am. Stat. Assoc.
Modeling daily arrivals to a telephone call center
Manage. Sci.
Cited by (13)
Fluid models for call centers with delay announcement and retrials
2018, Knowledge-Based SystemsCitation Excerpt :In practice, blocked or abandoned calls may be attempted later, that is, customers were unable to enter the system on the first attempt. This type of retrial behavior and its influence is modeled by Haddad et al. [14] and Shin and Moon [26]. Mandelbaum et al. [22] consider multiserver systems with abandonments and retrials and propose a fluid approximation for their analysis.
A novel pricing strategy for mobile broadband carriers using two-stage Stackelberg model
2018, Knowledge-Based SystemsCitation Excerpt :In addition, it is interesting to examine the impacts of different consumer consumption behaviors [13,19] or different criteria preferences [21,22] on the strategy choices.
Data analysis and call prediction on dyadic data from an understudied population
2017, Pervasive and Mobile ComputingSequential behavior prediction based on hybrid similarity and cross-user activity transfer
2015, Knowledge-Based SystemsCitation Excerpt :The prediction task can also be extended to other domains, such as phone human behavior, stock return, bank failure, customer consumption, and so on [4,2,12–14]. Haddad et al. [4] proposed a prediction model for recurrent consumption behaviors based on inhomogeneous Poisson processes, which could effectively predict future incoming/outgoing phone calls. Baumann proposed to predict human mobility based on adaptive sensor cooperation [15].
Text and phone calls: user behaviour and dual-channel communication prediction
2020, Human-centric Computing and Information Sciences