Association Rules Relating Food Taste and Food Characters with Kikei Roles

全文

(1)Kansei Engineering International Journal Vol.11 No.4 pp.235-240 (2012) Special Issue on 2011 ICBAKE (International Conference on Biometrics and Kansei Engineering). ORIGINAL ARTICLE. Association Rules Relating Food Taste and Food Characters with Kikei Roles Linfu LI* and Takashi UOZUMI** * Satellite Venture Business Laboratory, Muroran Institute of Technology, 27-1 Mizumoto-cho, Muroran, Hokkaido 050-8585, Japan ** Department of Computer Science and Systems Engineering, Muroran Institute of Technology, 27-1 Mizumoto-cho, Muroran, Hokkaido 050-8585, Japan Abstract: Based on the theory of Yakuzen, food as medicial treatment has affected dietary over a long history . Especially, there are food care principle which take the advantage of food taste and food character to support health care and treat the disease. Yakuzen theory is still regarded as a hypothesis. Many aspects of herbal treatments’ effectiveness remain to be clarified. Nevertheless, research results show that many rules have been advanced for foods’ associated medical roles. Kikei, one promising hypothesis, holds that food taste will play a special medicinal role in different organs. This study, using data mining techniques, particularly addresses the relation between the food taste, food character and Kikei role to extract the associated rules. Results of cluster analysis and chi-square tests show a therapeutic effect, which encourages our establishment of a systematic methodology related to food care. Keywords: food taste, Yakuzen, food character, kikei, association rule, cluster, support, confidence, lift. 1. INTRODUCTION. 2. FOOD TASTE, FOOD CHARACTER AND KIKEI ROLE. Food is a basic material to support life and health, which not only offers the necessary calories, but also supplies necessary nutrients to maintain normal activity every day. In recent years, as lifestyle diseases have increased, health-conscious nutrition is recognized as important. Consciousness of food care and food support without nutritional guidance from a dietitian is growing in popularity. One of the methods is Yakuzen food theory [1], which is a Chinese food medical theory based on ‘food as medicine’, and have became to be a new paradigm. Based on theories of traditional Chinese food medicine, Yakuzen has affected dietary treatment over China’s long culinary history. However, many such hypotheses of food dietary treatment remain to be verified. Yakuzen theory is still regarded as a hypothesis. Many aspects of herbal treatments’ effectiveness remain to be clarified. This study uses data mining techniques –– particularly addressing the relation between food taste, food character and its Kikei role [2], a Chinese medicinal concept –– to extract rules associating foods with their physiological effects. Based on cluster analysis [3] and chi-square tests, we verified therapeutic effects to support our eventual aim of establishing a systematic methodology of food care theory.. Received 2012.03.15 Accepted 2012.09.22. Yakuzen theory, originally from China, was translated into Japanese directly based on the theories of Chinese medicine ingredients. It is a food-based health care method based on food characteristics and a person’s health status. Normally, foods that have medicinal roles can be regarded as Yakuzen food. The key point for food, to be regarded as a medicinal material or not, is related to whether the food itself has the role of heating or cooling the body. The roles which elicit body heat or cold, designated as the five characters (translated from Chinese into Japanese as Goki), include warm, hot, neutral, cool, and cold. According to the natural level of cooling, the food characters are separable into cool and cold types; by the warming level, foods are classifiable into warm and hot types. Being neither warm nor cold implies a neutral character. For example, after eating ginger or hot pepper, a person’s body will feel hot, and perspiration is likely to appear. Warming foods are of two types: hot and warm, which make the body work actively, improve blood flow, and remove cold from the body to warm the body. Cooling character foods have two types –– cool and cold –– which reduce body heat, hydrate the body, and eliminate poisons from the body. Neutral foods can balance body heat, making the body neither warm nor cool.. 235. Copyright © 2012 Japan Society of Kansei Engineering. All Rights Reserved..

(2) Kansei Engineering International Journal Vol.11 No.4. Another feature of food is taste. According to Yakuzen theory, foods have five tastes, translated from Chinese into Japanese as Gomi. There are sweet, sour, bitter, spicy, and salty. Each taste has distinctive physiological effects. For example, bitterness can discharge extra components from the body; its drying role can discharge heat and water from the body, which is helpful for high fever and constipation. Sourness can produce muscle tightening, which have the function of stopping perspiration or heavy urination. Therefore, sourness is effective for hyperhidrosis, urinary frequency, and involuntary ejaculation. Tastes have effective roles for human organs, named Kikei, which means that the food can be located selectively at one organ and can affect the organ. In a body, there is a meridian system. The meridian system is a transmission course of energy and information. The meridian system reaches each internal organ. For a medicine or a food, Kikei means which meridian in particular internal organ it can reach and be effective. Table 1 presents the Kikei relation between tastes and affected organs. The association between taste and Kikei from 138 kinds of foods has been analyzed based on previous study. In this study [4], 156 kinds of foods were be selected, besides the taste, the association between food character with kikei role will be veritified too.. together, the customer is likely to also buy ground beef. Such information is useful as a basis for decisions about marketing activities such as promotional pricing or product placements. Here, the association rule will be used to verify the association between the Kikei character of Yakuzen to human organs. To select the association rules, constraints on various measures of significance and interest are useful. The bestknown constraints are minimum thresholds of support, confidence, and lift. The support supp(X) of an item set X is defined as the proportion of transactions in the dataset which contain the item set. The confidence of a rule is:. 3. EVALUATION FEATURE VALUE OF ASSOCIATION RULE. 4. ANALYSIS OF ASSOCIATION RULE BETWEEN THE FOOD TASTE, FOOD CHARACTER WITH KIKEI ROLE. conf(X=>Y) = supp(X∪Y)/supp(X) For example, {Ginger}=>{Lung} has a confidence of 0.2/0.4 = 0.5 in the database, which means that it is applicable for 50% of the transactions under the condition. The lift of a rule is defined as lift(X=>Y) = supp(X∪Y)/supp(Y)*supp(X). The rule appearance ratio is low if the support is lower. However, if the items are in numerous databases, then a high support degree of individual items cannot be expected. Rule evaluation must be considered comprehensively for support, confidence, and lift.. Association rule is one of the most popularly methods of data mining for discovering interesting relations among elements in large databases. Piatetsky-Shapiro [5] describes analyses and precedent strong rules discovered in the database. Based on the concept of strong rules, association rules are used for discovering regularities between products in large scale transaction data recorded by supermarket POS systems. The rule “{onions, potatoes} => {ground beef}” found in the sales data of a supermarket would indicate that if a customer buys onions and potatoes. The study aims to clarify the association rules for food taste, food character and Kikei medicinal effects. In this study, another taste, ‘bland’, was also regarded as important feature too, and will be regarded as the other five tastes. From the reference of “Chinese food treatment” [2], there are 156 kinds of food be selected fir the food taste. The Kikei medicinal effects are shown in Table 2. Table 2: Food Taste, Food Character, and Kikei Medical Effects. Table 1: Kikei exemplifying for food taste with Organs Taste. Sweet. Sour. Spicy. Salty. Bitter. Organ Spleen. Liver. Lung. Kidney. Heart. Effect. Stress reduction Weakness improvement Pain alleviation. Constipation Constipation Hyperhidrosis Poor Appetite Diarrhea circulation Anemia improvement Frequent Depression Digestion urination. Food. Grains Fruit. Apricot Plum Pomegranate. Ginger Garlic Onion Pepper. Kelp Laver Prawn. Bitter gourd Lettuce. 236.

(3) Kansei Engineering International Journal Vol.11 No.4 Association Rules Relating Food Taste and Food Characters with Kikei Roles. We can extract 191 rules based on formula (1) if the parameters were be adjusted as Support = 0.04, confidence = 0.05, extracted association rules are 136. In the extracted association rule, the left side is antecedent, marked as lhs. The right is consequent, marked as rhs. The rule can be sorted based on support, or confidence, or lift. Fig. 2 shows the subset of the associated rule, sorted by support and limited on 20. There are 191 rules were extracted and the highest support item is “sweet”, the highest support association rule is {sweet -> stomach}, {sweet -> spleen}. There is a new and more faster algorithms which named Eclat (Equivalence class transformation) [6, 7] to extract association rules. This algorithms can extract the frequency item from data set. Here, the function of Eclat, was employed as:. Based on the apriori algorithm, the package rules installed in R language is used to extract the association rule between food taste and the Kikei role. Prior to be processed, the data of Table 2 are translated into a database format as in Table 3, in a transaction, 1 represents presence, and 0 denotes the absence of an item. The item frequency is depicted in Fig. 1. Here the function of “apriori” is used to extract the association rule between food taste and Kikei medical effects. Parameters of the function include support, confidence, and the maximum length of the frequency item. The association rules can be taken up by funtion of “inspect” by setting proper threshold evaluation value. The extracted rules were arranged in the order column from left to right is as lhs (left-hand-side), rhs (right-handside), support, confidence, lift. Here, the parameter of support is given as 0.02, confidence is given as 0.03. ajii_kikei_rule = apriori (ajiki_kikei_transdata, parameter = list(maxlen = 2, support = 0.02, confidence = 0.03, ext = TRUE)). aji_ki_ec = eclat (ajiki_kikei_transdata) aji_ki_ec2 = aji_ki_ec[size(items(aji_ki_ec)) == 2] inspect (SORT (aji_ki_ec2, by = “support”)[1:5]) (2). (1). The Result of frequency item is shown as Table 4. From Table 3, the first rank is {Sweet, Stomach}, the second rank is {Sweet, Spleen} based on support value, it shows the same result that apriori function extracted association rule is matched with frequency item which extracted by. Figure 1: Item frequency bar graph of food taste, food character and Kikei. Figure 2:. Extracted association rules sorted by support. Table 3: Transaction format data of food taste, food character, and Kikei medical effects item cold wheat 0 barley 0 buckwheat 0 bubble 0 adlay 0 rice 0 corn 0. cool neutral warm 1 0 0 1 0 0 1 0 0 1 0 0 1 0 0 0 1 0 0 1 0. hot 0 0 0 0 0 0 0. sweet 1 1 1 1 1 1 1. salt 0 1 0 1 0 0 0. bland sober bitter 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0. sour 0 0 0 0 0 0 0. spicy 0 0 0 0 0 0 0. heart spleen kidney stomach 1 1 1 0 0 1 0 1 0 1 0 1 0 1 1 1 0 1 1 0 0 1 0 1 0 0 0 1. lung 0 0 0 0 1 0 0. bladder 0 0 0 0 0 0 0. liver 0 0 0 0 0 0 0. ~~ meat pig hearts ham beaf lamb ~~~. 0 0 0 0 0. 0 0 0 0 0. 1 1 1 1 0. 0 0 0 0 1. 0 0 0 0 0. 1 1 0 1 1. 1 1 1 0 0. 0 0 0 0 0. 0 0 0 0 0. 0 0 0 0 0. 0 0 0 0 0. 0 0 0 0 0. 0 0 0 0 0. 1 0 0 0 1. 1 1 1 0 0. 1 0 0 0 1. 0 0 0 1 0. 0 0 0 0 0. 0 0 0 0 0. salmon octopus. 0 0. 0 0. 1 1. 0 0. 0 0. 1 1. 0 1. 0 0. 0 0. 0 0. 0 0. 0 0. 0 0. 0 1. 0 1. 1 0. 0 0. 0 0. 0 1. 237.

(4) Kansei Engineering International Journal Vol.11 No.4. Table 4: Extracted frequency item by eclat function Items {Sweet, Stomach} {Sweet, Spleen} {Neutral, Sweet} {Spleen, Stomach} {Sweet, lung}. Table 6: Extracted association rules of Kikei with food character. Support 0.4230769 0.4166667 0.3397436 0.2500000 0.2307692. Eclat function result. The association rule can be deduced from the evaluation parameter using the function of subset after extracted. aji_kikei_asrule = subset (ajiki_kikei_rule, subset = (lhs %in% “sweet” | lhs %in% “salt” | lhs %in% “bland” | lhs %in% “sober” | lhs %in% “bitter” | lhs %in% “sour” | lhs %in% “spicy”) & (rhs %in% “heart” | rhs %in% “stomach” | rhs %in% “spleen” | rhs %in% “kidney” | rhs %in% “large.intestin” | rhs %in% “small.intestine” | rhs %in% “lung” | rhs %in% “bladder” | rhs %in% “liver”)) (3). Support Confidence Lhs.supp 0.0256. 0.0315. 0.8141. 1.2283. Bland. Lung. 0.0256. 0.8000. 0.0320. 2.6000. Bitter. Lung. 0.0384. 0.6666. 0.0576. 2.1660. Sweet. Small. intestine. 0.0641. 0.0787. 0.8141. 1.1166. Spicy. Lung. 0.0769. 0.8000. 0.0961. 2.6000. ~~. ~~. ~~. ~~. ~~. ~~. Sweet. Liver. 0.1731. 0.2126. 0.8141. 1.0528. Sweet. Kidney. 0.2115. 0.2598. 0.8141. 0.9651. Sweet. Lung. 0.2307. 0.2834. 0.8141. 0.9212. Sweet. Spleen. 0.4166. 0.5118. 0.8141. 1.0505. 0.0666. 0.3846. 0.9454. Cold. Heart. 0.05128. 0.2963. 0.1731. 2.7189. Neutral. Heart. 0.02564. 0.0666. 0.3846. 0.6117. Cold. Larg. intestine. 0.02564. 0.1481. 0.1731. 1.5204. Cold. Lung. 0.03205. 0.1851. 0.1731. 0.6018. ~~. ~~. ~~. ~~. ~~. ~~. Cool. Stomach. 0.0897. 0.5384. 0.1666. 0.9473. ~~. ~~. ~~. ~~. ~~. ~~. Neutral. Stomach. 0.1794. 0.4666. 0.3846. 0.9454. Neutral. Spleen. 0.1858. 0.4833. 0.3846. 0.9921. ANALYSIS. OF. dd1 = dissimilarity (aji_kikei_asrule) dd2 = dissimilarity (ki_kikei_asrule). lift. ASSOCIATION. (5). Based on results of hierarchical cluster analysis, association rules between food taste and kikei are divisible into five classes. Class 1 includes rule 1, rule 4, rule 8, rule 18, rule 19, rule 20, rule 21, rule 22, rule 23, which shows sweets associated with organs. Class 2 includes rule 14, rule 15, rule 16, rule 17, which shows are salt associated with organs. Class 3 includes rule 9 and rule 10, rule 12, rule 13, which shows sour associated with organs. Class 4 include rule 2, rule 3, rule 5, rule 11, which shows bitter and spicy associated with organs. Class 5 includes rule 6, rule 7, which shows spicy associated with organ. A cluster graph is shown as Fig. 3. Based on results of hierarchical cluster analysis, association rules between food character and kikei are divisible into four classes. Class 1 includes rule 1, rule 3, rule 16, rule 18, rule 23, rule 24, rule 25, rule 26, which shows neutral associated with organs. Class 2 includes rule 2, rule 4, rule 5, rule 6, rule 7, rule 8, which shows cold associated with organs. Class 3 includes rule 9 and rule 10, rule 11, rule 12, rule 13, rule 14, it shows cool associated with. Table 5: Extracted association rules of Kikei with food taste Bladder. 0.0256. The extracted association rules were analyzed by the cluster method [8]. From the extracted association rule, and combined with the R languages function of “dissimilarity”, food taste and food character with kikei association rules can be clustered by formula (5).. Here Table 6 is the result of subset from the extracted rules. The following results can be drown that 1) Neutral has more higher associated relation with organ based on support and confidence feature value. 2) Cool, cold, warm and hot, have lower associated relationship with organ.. Rhs. Support Confidence Lhs.supp. Small. intestine. 5. CLUSTER RULES. ki_kikei_asrule = subset (ki_kikei_rule, subset = (lhs %in% “cold” | lhs %in% “cool” | lhs %in% “neutral” | lhs %in% “warm” | lhs %in% “hot”) & (rhs %in% “heart” | rhs %in% “stomach” | rhs %in% “spleen” | rhs %in% “kidney” | rhs %in% “large.intestin” | rhs %in% “small.intestine” | rhs %in% “lung” | rhs %in% “bladder” | rhs %in% “liver”)) (4). Sweet. Rhs. Neutral. 3) The support and confidence show a lower value than the taste with kikei, which means lower association between food character with kikei role. Based on Chinese medicinal theory, food taste near to neutral is good for health, which is verified through the extracted association rule.. In fact, 23 rules can be extracted using formula (3). The extracted rules are presented in Table 5. The association rule between food character and kikei also had been extracted in this study. In fact, 26 rules can be extracted using formula (4).. Lhs. Lhs. lift. 238.

(5) Kansei Engineering International Journal Vol.11 No.4 Association Rules Relating Food Taste and Food Characters with Kikei Roles. organs. Class 4 include rule 15, rule 17, rule 19, rule 20, rule 21, rule 22, which shows warm associated with organs. The cluster graph is shown as Fig. 4.. The proposed associated rule is X=>Y. The support of X, Y, and X∪Y is marked as SX, SY, and SXY. The amount of transaction is proposed as M. In addition, the chi-square test [9] value of Tdep is calculated using formula (6):. 6. CHI-SQUARE TEST OF THE EXTRACTED ASSOCIATION RULES RELATED TO FOOD TASTE AND KIKEI ROLE. Tdep = M(Sxy-Sx)2/SxSy(1-Sx)(1-Sy). (6). If Tdep is close to 0, then the relation between X and Y relation is independent. If Tdep is larger more, there is stronger correlation between X and Y. Propose an established level of significance, if Tdep < χ2(α), then X and Y are inferred to be independent. Here, based on α = 0.05, χ2(α) = 3.84, which is related to the significance that is shown in Table 7. Based on chi-square test results, the associations between ‘salty with kidney’, ‘sweet with spleen’, ‘sour with liver’, ‘bitter with heart’, are inferred as significant, but the association between ‘hot with lung’ is not inferred as significant. For the support and confidence of associated rules of food character related with kikei role is obviously lower value, and based on Yakuzen theory, there are no hypothesis about the associated relation between food character with kikei, so therefore, the chi-suqare test could not be verified.. As a criterion of associated rules, only support and confidence evaluation are not always appropriate. To evaluate the extracted association rules properly, chisquare tests were performed in addition to the support and confidence.. 7. CONCLUSION This study performed association rules associated with food tastes, food character and affected organs. The Kikei role of Yakuzen was verified based on data mining. The association rule between food taste, food character and the Kikei role was extracted based on the support and confidence. A stronger association rule is shown between sweet and spleen, as well between sweet and stomach. The other associated relationships are also shown by the other Kikei role, including sour with liver, bitter with heart, salt with kindney, spicy with lung. In this study, we try to verify whether there are association rule between food character related with Kikei, 26 association rules of food character with Kikei are extracted, neutral show more higher associated relation with organ based on support and confidence feature value. cool, cold,. Figure 3: Cluster tree of the extracted association rule between food taste and kikei role. Table 7: Chi-square test result Associated item. Figure 4: Cluster tree of the extracted association rule between food character and kikei role.. 239. Chi-square test value. Significance. Salty: kidney. 23.458. Yes. Sour: liver. 39.861. Yes. Bitter: heart. 5.045. Yes. Sweet: spleen. 660.061. Yes. Hot: lung. 3.156. No. Sweet: stomach. 638.625. Yes. Faint: kidney. 9.572. Yes.

(6) Kansei Engineering International Journal Vol.11 No.4. warm and hot, show lower associated relationship with organ. The result can be certified that neutral food is good for health of Yakuzen theory. The association rules are verified using the chi-square test. Aside from hot with lung, other association rules between tastes and organs show significance. This time, only 156 samples were used, perhaps explaining the lack of an association between hot and the lungs. The next step of research will involve collecting more food samples to verify the association rule.. Linfu LI Japan Kansei Engineer member. received his B.E. degree from Suzhou University, Suzhou, China in 1992. He received his M.E degree from Suzhou University, Suzhou, China in 1998. And his D.E. degree from Shinshu University, Ueda, Japan in 2003, he was a post doctoral researcher of SVBL, Muroran Institute of Technology, Japan from 2003-2006, and he was a post doctoral researcher of Kyushu University in 2006-2007. Now he is a post doctoral researcher of SVBL, Muroran Institute of Technology. His research field is Kansei Engineer, Ontology, and Textile fiber. Contact email address: li@mmm.muroran-it.ac.jp. REFERENCES 1. Tatsumi, H. “Yakuzen material dictionary”, Gensou Sya, pp.15-27 (2006). 2. Seo, K., Murakata, A., and Inata, E. B. Chinese medicine food treatment, Toyo gakujutsu syuppansya, pp.2-12 (2003). 3. Kim, A., Data Science by R, Morikita Syuppan sya, pp.107-126 (2007). 4. Linfu Li, Takashi Uozumi, “Association Rules Relating Food Tastes and Kikei Roles Based on Yakuzen Theory”, ICBAKE2011, KAGAWA, Japan, pp.39-43, 2011. 5. Agrawal, R. and Srikant, R. Fast algorithms for mining association rules. Proc. 20th. Conf. VLDB, pp.487-499. 6. Charu C. Aggarwal, Finding localized associations in market basket data. IEEE Trans. On knowledge and Data Engineering, 14(1), pp.51-62. 7. M. J. Zaki, Scalable algorithms for association mining, IEEE Transactions on Knowledge and Data Engineering, 12(3), pp.372-390, 2000. 8. Shinnou, H, Learning cluster analysis by R, Ohmsha, pp.72-73 (2007). 9. Aoki, S, Statistic analysis by R, Ohmsha, pp.119-120 (2009).. Takashi UOZUMI Takashi Uozumi received his B. E. degree from Muroran Institute of Technology, Muroran, Japan, in 1973 and his D. E. degree from Hokkaido University, Sapporo, in 1980. Since 1988, he has been an associate professor at the Department of Computer Science and Systems Engineering, Faculty of Technology, Muroran Institute of Technology. His research fields include image understanding, computer vision, machine learning and Kansei Engineering. He is a member of the Kansei Society of Japan, IEEE, the Information Processing Society of Japan and the Japan Society of Medical Electronics and Biological Engineering. Contact email address: uozumi@epsilon2.csse.muroran-it.ac.jp. 240.

(7)