Designing a Smart Consumption Tracking Model *

Transcription

1 , pp Designing a Smart Consumption Tracking Model * Boeun Jung and Sora Lim Hankuk University of Foreign Studies qingtian88@naver.com, angelalim@hufs.ac.kr Abstract Big data analytics has received wide attention by information technology industries. It has been done in quantitative and statistic viewpoints. Observing huge amount of data, it is possible without doubt to establish a model that may predict purchase behaviors of consumers. But this approach can neither explain what brings the consumers to such decisions nor predict future purchase behavior of other product categories. Furthermore, it is not possible to reason about consumers preferential differences that make choose or avoid certain places and shops. To answer this question, this paper argues that a qualitative analysis based on consumption values will be an alternative, and proposes a conceptual model of extracting consumption values from big data using clothing purchase as a case study. Keywords: consumption value, big data, decision-making, preferential difference, smart consumption map 1. Introduction Tracking consumption patterns has become one of the most highlighted issues in present day marketing. Before and during the era of industrialization, what had triggered production and consumption was the lack of commodities necessary to maintain social life. 1 But in the post-industrial era in which shortages of basic essentials no longer exist, consumers do not spend just for necessities. On the contrary, in the post-industrial era, what triggers consumption is the additional values assigned to goods which are not essential to be a good. In this new post scarcity society, what we consume is not a good itself but the values added on it. These values can be assigned and determined by various social relations. Since they are closely related with our identities, it is reasonable to assume that what we consume is those personal indwelling values. In this aspect, extracting those values is worthwhile to track and analyze consumption so as to optimize efficiency and profitability. Statistical analyses using big data sources are mostly explored to figure out consumer preferences and consumption patterns. With these big data approaches, it may be possible to predict future consumption. Namely, it is a case study of forecasting consumer behavior through past consumption patterns based on big data analytics. However, these statistical methods are weak to clarify what are the relevant decision-making factors of such consumption. And as they are based only on probability and mathematical statistics which have been obtained from an observation of purchasing attitudes and behaviors for certain * This paper is a revised and expanded version of a paper entitled "Extracting Consumption Value from Big Data and Its Application" presented at International Conference on SUComS 2014, occurred at Budapest, Hungary, on August 14-17, The term "social life", presented here, means not only physical life in the sense of survival but also the minimum quality of life in a relative sense. ISSN: IJSEIA Copyright c 2014 SERSC

2 categories of products, it is difficult to use them to predict purchasing behaviors concerning other product categories. This problem is chiefly due to a lack of a relevant logical connection between those two separate choice structures. For this reason, further study is needed to develop a logic model based on qualitative research instead of statistical approach to quantitative data. It may allow reasoning about future consumption behavior and planning persuasive marketing approach according to factors of consumer decision-making. In this context, we suggest the utilization of Floch's [1] theory of consumption values for the baseline of this discussion. 2. Basic Consumption Values Scholars and researchers suggest different approaches to estimate consumption values. For example, Wiedmann, Hennigs and Siebels [2] suggest eight different forms of value: price value, usability value, uniqueness value, self-identity value, hedonic value, materialistic value, conspicuousness value and prestige value. On the other hand, Floch suggests four types of valorizations for consumption values: critical value, practical value, ludic value and utopian value. He proposed following semiotic square with these four types of valorizations that could reveal what consumers want or what factors influence their buying behaviors. Figure 1. Semiotic Square Each of these ideas has merits and demerits. But they deserve to be regarded as fundamental for identifying key factors affecting consumer decision-making process which are summed up by Melo and De Lencastre [3] in the following way: Practical valorization: corresponds to the utilitarian values with specific and realistic aims. Accordingly, the product will be appreciated for being practical, functional and adequate to its function; Critical valorization: corresponds to the non-existential values, characterized by separation and comparison. Quality/price, economy, innovation/cost will be important criteria, frequent in critical evaluation; Ludic valorization: corresponds to the non-utilitarian values such as luxury, refinement, impulsive act. It lies at an emotional and sensorial level; 168 Copyright c 2014 SERSC

3 Utopian valorization: corresponds to the existential values such as the identity and lifestyle. According to this valorization, the product will always be considered an accomplishment of something to satisfy customers' expectations. Consumption values are subjective since they are measured by emotional evaluations. In fact, they are very personal and therefore each index of these values can vary from person to person. It should also be noted that evaluation and interpretation of these values can be changeable not only according to consumers, but also according to contexts in which they make purchasing decisions. Namely, even if interpreted by the same person, a same index can be understood differently due to various contextual factors [4]. 3. Consumption Values from Big Data and Its Interpretation 3.1. Big Data Analytics (BDA) Big data analytics (BDA) refers to the process of examining large sets of data sources to discover systematic relationships between variables and other useful information. These data sources may include web server logs, clickstream data, social media activity, call detail records and information captured by sensors. Briefly speaking, they display a collection on complex and unstructured information from everything that is related to consumer buying behavior. For this reason, it actually is a challenge to analyze such a sheer volume of data and its many different formats. BDA should therefore be performed using specialized software tools and applications in order to process and transform data into meaningful information. This process can be summarized in the following diagram. Figure 2. BDA Process [5] The software tools and applications associated with BDA are commonly used as part of advanced analytics such as data mining. In fact, data mining is an analytic process designed to explore large amounts of data, and it usually consists of three stages: (a) the initial exploration with data preparation which involves web crawling, scrapping, text mining, and data taxonomy, (b) model building or pattern identification with validation of the findings in previous stages, and (c) deployment by applying the model selected or the detected patterns to new data in order to predict future behaviors [6-13]. While data mining is generally concerned with the detection of patterns in numeric data, useful information is stored in the form of text and it can be a valuable business asset. Due to its importance, the text mining, which until very recently was not commonly recognized as a field of interest, emerges as a rapidly growing and major Copyright c 2014 SERSC 169

4 area where significant methodological and theoretical advances are being made [14]. Among all analytical methods that are available for BDA process, text mining is the most important key steps for handling qualitative data such as consumption values, and its process can be illustrated as detailed in the diagram below. Figure 3. Text Mining Process [15] For text mining, Natural Language Processing (NLP) is used as a screening tool to identify the sentence, extract its literal meaning, infer the implied intent, and lastly discern what should be done or assumed given that it was uttered. For example, suppose there is a sentence like: "I'm not comfortable wearing Levi's skinny jeans, but it's so gorgeous. I couldn't resist buying it". Firstly, NLP separates words in this sentence into individual morphemes and then identifies them by using morphological analyzer (see Figure 4). Figure 4. Example of Morphological Segmentation Secondly, given a sentence, NLP determines the part of speech for each word (see Figure 5). This is called the part-of-speech (POS) tagging. POS tagging is considered an inseparable part of NLP, because some words can represent more than one part of speech, depending on the context. For instance, as seen in the figure above, "able" can 170 Copyright c 2014 SERSC

5 be an adjective or an adjectival suffix, and " s" can be a possessive adjectives or a contraction of the English words "is" and "has". Therefore, it's not easy to identify the correct part of speech without understanding the semantics. Figure 5. Example of POS Tagging By using Name Entity Recognition (NER), NLP determines which items in the given sentence to proper name, such as people, places, brands or product names, and what the type of each such name (see Figure 6). Figure 6. Example of NER After that, NLP restates the meaning of the given sentence by paraphrasing (see Figure 7). This process helps explain or clarify the context in which it appeared, and facilitates conveying its intended meaning. Figure 7. Example of Paraphrasing And lastly, NLP parses the sentence giving a structural representation and checking for correct syntax (see Figure 8). Figure 8. Semantic Analysis Copyright c 2014 SERSC 171

6 This semantic analysis process can be converted into computational method as below: (ROOT (S (S (NP (NP (NNP Levi) (POS 's)) (JJ skinny) (NNS jeans)) (VP (VBP are) (ADJP (JJ uncomfortable)))) (,,) (CC but) (S (NP (PRP I)) (VP (VBP like) (NP (DT the) (NN design)))) (..))) (ROOT (S (NP (DT That)) (VP (VBZ 's) (SBAR (WHADVP (WRB why)) (S (NP (PRP I)) (VP (VBD bought) (NP (PRP it)))))) (..))) As such, NLP allows converting large volumes of unstructured data into meaningful information, and converting also human language into readable representations by computer programs. As it facilitates the application of linguistic principles to extract purchase decision context, can be used as a basis for consumer profiling Extracting Consumption Values: A Case Study on Clothing Purchase Decision Process Theoretically and practically it is possible to make forecasting model for purchase behavior of customers by using statistics to analyze a customer's purchase history. But this approach does not show what is decision maker while guess future purchase action by statistics. Besides, this kind of forecasting model based on quantitative analysis say nothing about the preferential behavior of consumers. Simply, consumers purchase what please them. It is not easy to define and formulate please because this is personal feeling that is not quantifiable. However, consumption values seem to be a clue to solve the problem. In this respect, this paper suggests a conceptual model that has reasoning capabilities of future purchase behavior. For this purpose, based on the behavioral profiling obtained from BDA process, we have tried to identify how to extract consumption values from big data using clothes purchase as a case study. 172 Copyright c 2014 SERSC

7 Figure 9. Conceptual Model of Clothing Purchase Decision Process and Consumption Values Based on empirical evidence, Figure 9 shows a general and a simplified approach for classification of clothing purchase decision process and consumption values. Every value dimension presented above is widely acceptable under normal circumstances. Obviously, several influencing factors may be related to these value dimensions, such as customer s individual variables, but they are not considered here. As it is noted earlier, a value can have different interpretations from person to person. For example, practical value for one customer can be ludic or utopian value for others. Furthermore, a same customer can judge differently facing the same product according to situations. From long-term point of view, it is absolutely necessary to develop formulas that can extract consumption values more correctly matching with situations, contexts and consumer s particular variables. But that's beyond the scope of this paper and perhaps we could deal with that later Interpretation of Consumption Values Regarding matching those extracted consumption values with empirical human behaviors, it is possible to propose a matrix interpretation as follows: Table 1. Empirical Interpretation of Consumption Values Combinatory consumption value Consumption patterns Involvement Ludic > Utopian Overspending Impulse buying Ostentation Low Loyalty to brand Low Utopian > Ludic Overspending Self-accomplishment Low High Utopian > Practical Rational consumption Self-satisfaction High Low Practical - Critical Rational consumption Quality/Price High Low Copyright c 2014 SERSC 173

8 The first group that shows predominant ludic values has priority in the following order: ludic > utopian > critical > practical. For this group, quality/price, usability or necessities of goods are not under consideration. There are other considerations we take into account such as reputation or social values added to products. This group shows very impulsive spending and easily affected by what goes viral. Since consumers belong to this group care less about price, quality and usability, emotional approach could be the appropriate marketing strategy to stimulate consumer purchase decision. High price policy, giving superiority feeling, arousing jealousy are therefore some of the most effective strategies for this group. Figure 10. Ludic Dominant Patterns According to Figure 1, as noted earlier, it is expected that utopian-critical and practical-ludic combinations do not occur, as they are contradictory values. Those who have great utopian orientation, show tendency to purchase goods which are beyond their financial condition but fulfill their utopian values. For them, critical values, related to price or quality, do not matter. Because of this reason, practical-ludic and ludic-critical combinations are also excluded from discussions. These combinations are observed when people buy goods just for fun without further consideration. For the second group where utopian values are predominant, there are two different subgroups according to the valorization of self-implementation. The first subgroup has priorities in the order of utopian > ludic > practical > critical while the second has utopian > practical > critical > ludic order. 174 Copyright c 2014 SERSC

9 Figure 11. Utopian Dominant Patterns Firstly, utopian > ludic oriented consumers have a tendency of self-implementation and self-satisfaction. And they often ignore someone s eyes. For consumers of this type, price is not under consideration. It may be expensive but cheap too. They want to be different and stand out from others by possessing special items regardless of the price. Sometimes relatively inexpensive products can be used as icons of special social values or of particular social classes and groups. In fact, their self-realization is not always relied on high quality/high price products, but rather they tend to insist particular brand showing strong loyalty to them. On the other hand, utopian > practical oriented consumers are inclined to realize themselves through special activities instead of purchase. For them, purchase is only complementary fulfillment of their primary goal of self-realization. One such example is sportswear and equipment brands. And lastly, practical > critical oriented consumers tend to spend very rationally considering necessity and quality/price. This is the most common consumption value of usual consumers under normal circumstances if there are no variables. For consumers of practical > critical orientation, informative advertising is the most effective approach to induce purchases. Figure 12. Practical - Critical Dominant Patterns Copyright c 2014 SERSC 175

10 3.4. Application The conceptual model previously presented focuses not on frequency but on meaning of purchase behind. As shown earlier, it is possible to extract consumers consumption values in combinatory ways from big data. And once value orientation types are tracked for individual or collective consumers, it will be possible to offer customized smart services for consumers in omnidirectional ways. Before, statistical analyses using big data sources have shown only purchasing patterns of consumers but do not say about their preferential differences between same category of products and shops. For example, in quantitative approach, purchase patterns of consumers can only predict possible future purchase behavior for the same categories. If consumption patterns is extracted from clothes buying, it is only applicable for clothing purchase behavior because there are no universal features that can describe or explain consumer s behaviors. But in qualitative approach proposed here, the consumption values extracted from big data to track consumers' purchase behavior can be considered as universal properties of consumers in question. Thus, these values can be extended to predict purchase behavior of other categories. Furthermore, it makes possible to explain why consumers choose or avoid certain places and shops where sell same kind of products. Figure 13. Using Big Data Modeling to Create Smart Consumption Tracking Tool This conceptual model, then, can be used to offer customized smart map for individuals according to their consumption values. 4. Conclusion Statistical analysis of big data to predict purchase patterns of consumers is meaningful in the sense that it can predict future customer behavior by past performance. However, it is not an appropriate approach because its implication is limited to the purchase of same category and fails to explain consumers' decisionmaking process and preferential difference. On the other hand, qualitative approach based on consumption values solves those difficulties because certain values that affect more significantly consumer's attitudes and purchase decisions can be considered as decision-making factors. Consumption values based on qualitative analysis of big data can also allow creating smart cultural map reflecting consumers preferential differences. 176 Copyright c 2014 SERSC

11 Acknowledgements This work was supported by Research Fund of Hankuk University of Foreign Studies. References [1] J. Floch, Sémiotique, marketing et communication: sous les signes, les strategies, Presses Universitaires de France, Paris (1990). [2] K. Wiedmann, N. Hennigs, and A. Siebels, Value-based segmentation of luxury consumption behavior, Psychology & Marketing, vol. 26, no. 7 (2009), pp [3] C. Melo and P. de Lencastre, Values underlying the consumption of perfumes: social-semiotic approach, Proceedings of the 7 th International Congress Marketing Trends, ESCP-EAP Paris and Università Ca' Foscari Venezia, Edited J. Andreani and U. Collesei, (2008) January 17-19, Venice, Italy, online version available at [4] S.Y. Lee and J. O. Lee, Human Values for Authorizing Persuasive Multimedia Contents, International Journal of Multimedia and Ubiquitous Engineering, vol. 8, no. 3 (2013), pp [5] Saltlux, Big Data Analytics, Saltlux Inc., Seoul (2012). [6] M. J. A. Berry and G. S. Linoff, Mastering Data Mining: The Art and Science of Customer Relationship Management, Wiley Computer Publishing, New York, Chichester, Weinheim, Brisbane, Singapore, Toronto (1999). [7] H. A. Edelstein, Introduction to Data Mining and Knowledge Discovery, Two Crows Corp., Potomac, MD (1998), Second Edition. [8] U. M. Fayyad, G. Piatetsky-Shapiro, P. Smyth and R. Uthurusamy, Advances in Knowledge Discovery and Data Mining, MIT Press, Cambridge, MA (1996). [9] J. Han, M. Kamber and J. Pei, Data Mining: Concepts and Techniques, Morgan-Kaufman, New York (2000), Third Edition. [10] T. Hastie, R. Tibshirani and J. Friedman, The Elements of Statistical Learning: Data Mining, Inference, and Prediction, Springer, New York: Springer (2009), Second Edition. [11] S. M. Weiss and N. Indurkhya, Predictive Data Mining: A Practical Guide, Morgan-Kaufman, New York (1997). [12] C. Westphal and T. Blaxton, Data Mining Solutions: Methods and Tools for Solving Real-World Problems, Wiley Computer Publishing, New York (1998). [13] I. H. Witten, E. Frank, and M. A. Hall, Data mining: Practical Machine Learning Tools and Techniques, Morgan-Kaufman, Amsterdam, Boston, Heidelberg, London, New York, Oxford, Paris, San Diego, San Francisco, Singapore, Sydney, Tokyo (1997), Third Edition. [14] S. Sakurai, Editor, Theory and Applications for Advanced Text Mining, InTech, open access book (2012). [15] K. Y. Song, Yeogiea Dangshineui Yokmangyi Boinda: Big Dataga Tchtanen 70 Eok Yokmangeui Jido (Big Data Says You Want: Tracking Desires of 7 Billion Population), Sam & Parkers, Seoul (2012). Authors Boeun Jung, she is a Chief Researcher of Institute of Chinese Studies Center for International Are Studies Hankuk University of Foreign Studies. She has Ph.D. degrees in Graduate School of International Area Studies Hankuk University of Foreign Studies. Sociology Ph. D. Chinese majors Studies. Her recent articles are Have the Tools on Modern Chinese Intellectuals Study (2011); Portugal Cultural Transplant and Formation of Cultural Pluralism in Macao (2012) etc. Copyright c 2014 SERSC 177

12 Sora Lim, she is an assistant professor of Hankuk University of Foreign Studies (HUFS), Seoul, Rep. of Korea. From 2009 to 2012, she was a senior researcher for the Institute of Latin American Studies at HUFS. She has conducted research on multimedia learning materials for university students. She is currently interested in designing AR-based language learning model and digital cultural mapping. She graduated from HUFS with a B.A. degree in Portuguese. She received her M.A. and Ph.D. degrees in Comparative Literature from Federal University of Rio Grande do Sul, Brazil. 178 Copyright c 2014 SERSC