Article can be accessed online at http://www.publishingindia.com Utility Mining Algorithms - A Comparative Study Sivamathi Chokkalingam*, Vijayarani S** Abstract Utility mining is an emerging topic in data mining. The aim of utility mining is to discover the itemsets that have maximum utilities. Here utility refers number of items bought, cost of an item or it can be any other user choice in a transaction database. Frequent itemset mining is starting point of utility mining. In frequent itemset mining most often occurring itemsets in a transaction are retrieved. The discovery of such frequent itemsets can help in many business decision making process. Frequent itemset mining concentrates on the number of occurrence of items in a transaction, but not the value of items. But utility mining considers importance of itemsets like the profit it earns in a transaction, quantity in a transaction. In this paper various utility mining algorithms like MEU (Mining with expected utility), FUM (Fast Utility Mining), Two-Phase, CTUMine, UP-Growth (Utility Pattern Growth), and FHM (Faster High Utility itemset Mining) MHUI-BIT (Mining High-Utility Itemsets based on BIT vector), MHUT-TID (Mining High-Utility Itemsets based on TIDlist), and THUI (Temporal High Utility Itemsets) are discussed. Keywords: Utility mining, High utility itemset mining, MEU, MHUI-BIT & MHUT-TID, THUI-Mine, FUM, TwoPhase CTU-Mine, UP-Growth, FHM Introduction Data mining refers to extracting or mining knowledge from large amounts of data (Pujari, 2001). Data mining mainly concerns with the analysis of large volumes of data andretries previously unknown relationships or patterns from data. Frequent Itemset mining discovers the items based on how frequent the item occurs in a database without considering its importance. Frequent Itemset mining reflects statistical correlation of an item, but not its semantic significance (Kanimuthu et al., 2011). But for many real world applications utility of an item is also equally important. Consider an electronic super store. Assume that the profit of a mobile phone is 5000 INR and the profit of a memory card is 15 INR. In a transaction database, memory card occurs in 10 transactions and mobile phone occurs in 3 transactions. The total profit of memory card is 150 INR and the total profit of a mobile phone is 15000 INR. As per frequent itemset mining memory card has a higher frequency. But the total profit of a mobile phone is much greater than a memory card. Hence, traditional frequent itemset mining cannot discover the most important itemsets. This is because frequent itemset mining does not consider the profit(i.e. utility) of an item, which is also highly important in decision making. As per utility mining by considering profit mobile phone has high utility. Utility mining is an important task in marketing decision making process. Mining high utility itemsets from databases is an important task has a wide range of applications such as website click stream analysis, business promotion in chain hypermarkets, cross-marketing in retail stores, online e-commerce management, mobile commerce environment planning and even finding important patterns in biomedical applications (Tseng, Wu, Shie & Yu, 2012). Rest of the paper is organised as follows: second section * Ph.D Research Scholar, Department of Computer Science, Bharathiar University, Coimbatore, Tamil Nadu, India. Email: c.sivamathi@gmail.com ** Assistant Professor, Department of Computer Science, School of Computer Science and Engineering, Bharathiar University, Coimbatore, Tamil Nadu, India. Email: vijimohan_2000@yahoo.com 38 Journal of Applied Information Science discusses utility mining process. Third section presents literature survey on utility mining algorithms, while fourth section describes various utility mining algorithms and its pseudo code developed by researchers. Fifth section presents the conclusion. Utility Mining Process In all utility mining algorithms utility of each itemset is calculated based on its internal utility. Each algorithm uses its own data structure to store the item utility information. In general, utility mining algorithms work as follows: Volume 4, Issue 1, June 2016 Table 1: An Example Transaction Database Transaction id transaction T1 (i1,1) (i2,3) (i3,2) 19 T2 (i1,2) (i3,3) (i4,1) 21 T3 (i2,2) (i3,1) 10 T4 (i2,3) (i3,1) (i4,3) (i5,2) 32 T5 (i1,1) (i2,3) (i3,2) (i4,1) (i5,2) 28 Table 2: Utility Table Step 1: Take items from a transaction database. Step 2: Calculate Utility value for a transaction. Utility of an Transactions Item Profit (utility) Step 3: Get minimum utility threshold. i1 2 Step 4: Output the high utility itemset whose utility is greater than minimum utility threshold. i2 3 i3 4 i4 5 i5 2 High Utility Itemset and Frequent Itemset Consider the database of Table 1. This database contains five transactions T1, T2, T3, T4, and T5. The second column of the database is transaction items appear in a transaction. The third column is utility (say profit) of each item in a particular transaction. The last column is utility of a transaction. For example consider T1, which has items i1, i2 and i3 with occurrence 1, 3 and 2 respectively. Then the utility of this transaction is (1 * 2) + (3 * 3) + (2 * 4) = 19. Similarly utility of remaining transaction are calculated and is given in the last column. If the minimum utility threshold is chosen as 20, the transactions T2, T4 and T5 are having high utility itemset. High utility itemset differs from frequent itemset mining. To clearly understand the difference, the same database can be used to calculate association rule mining. Association rule can be evaluated using support and confidence metrics. Support calculates fraction of transactions contains itemset i1, i2. Confidence measures how often item i2 occurs in transaction that contains i1. The aim of association rule mining is to find all rules having, support≥minimum_suppport threshold, confidence≥ minimum_confidence threshold. Support count (α) is frequency of occurrence of an itemset. Support count of itemset i1 and i2 αis, α = ({i1,i2}) = 2. Consider n = number of transactions. Support of an itemset = α /n, Support of itemset ({i1, i2}) = 2/5 = 0.4. Confidence of an itemset({i1, i2})conf({i1, i2})is, conf({i1, i2}) = α ({i1, i2}) / α (i1) = 2/2 = 1. Similarly, Support of itemset ({i1, i3}) =3/5 = 0.6. Confidence of an itemset ({i1, i3}) = 3/2= 1.5 Support of itemset ({i2, i3}) =4/5 = 0.8. Confidence of an itemset ({i2, i3}) = 4/4= 1 If minsup_threshold is chosen as = 0.5, minconf_threshold = 1 then, ({i1, i3}), ({i2, i3}) are frequent itemset. Hence traditional frequent itemset mining techniques cannot be applied to utility mining (Tseng et al., 2012). Hence new algorithms are proposed for mining high utility itemset. Utility Mining Algorithms – A Comparative Study High Utility Itemset Mining High utility itemset mining refers to the discovery of high utility itemsets. The main objective of high utility itemset mining is to identify the itemsets that have utility values above given utility threshold (Agrawal, Imielinski & Swami,1993). The term utility refers number of itemset, profit of an item , weight, popularity or it can be even measures of user’s choice (Agrawal & Srikant, 1994). In this paper, a literature survey of various high utility itemset mining algorithms has been presented. Literature Review Many utility based algorithms have been developed. Association rule mining (ARM) identifies frequent itemsets from transaction databases and generates association rules (Agrawal et al., 1993). However, items are actually differs by its utilities. This utility helps to make a strong impact on the decision making in business applications. Therefore, traditional ARM cannot meet all JayanthiThe and Babu (2009)of these the demands of business Purusothoman, applications. base traditional ARM algorithms is the “downward closure property” (also known as anti-monotone property) is defined as any subset of a frequent itemset must also be Wu, Zida, frequent (Agrawal et al., 1993). This approach (n.d.) is wellorganised since more number of item combinations can be ignored at each level. However, this property doesn’t apply to the utility mining model. In utility mining, utility of a particular item may be low, but its superset may be a high utility itemset. An overview of various high utility itemset mining algorithms, its basic concepts and techniques used in various research papers have been given in this section. Fig. 1 displays of various high utility mining algorithms. Yao, Hamilton, and Butz (2004) proposed an algorithm called MEU (mining with expected utility). It finds all itemsets in a transaction database with utility values higher than the minimum utility threshold. Liu, Liao, and Choudhary (2005)proposed Two-Phase algorithm for finding high utility itemsets. Shankar, Purusothoman, Jayanthi and Babu (2009) presented an algorithm known as Fast Utility Mining (FUM). Tseng et al. (2012) proposed an efficient algorithm, namely UPGrowth i.e Utility Pattern Growth. All these algorithms calculate high utility values only for positive integers. There are algorithms like FHN (Faster High-Utility itemset miner with Negative unit profits)proposed by Fournier-Viger, Wu, Zida, and Tseng (2014) and HUINIV Mining (FUM). TsengNegative . (2012) Item Values)-Mine, (High Utility tyItemsets with by Chu, Tseng, and Liang (n.d.) to handle itemsets with negative profits. In supermarkets they may offer free items FHN (Faster High to promote the sales. Such free items (2014) and HUINIV (High Utility Itemsets with Negative Item have negative utility. The algorithms HUNIV and FHN can handle transaction with items that have positive and negative utilities. Fig. 1: Utility Mining Algorithms Utility mining Algorithms High utility Itemset with positive profits. Data Mining 1. MEU Data Streams 1. MHUI-BIT and MHUI-TID 2. Two phase 2. GUIDE 3. FHM 3. HUPMS 4. FUM 5. HUI 6. UP GROWTH 6. CTU-Mine 4. THUI 39 High utility Itemset with negative profits. 1. 2. FHN HUINIV 40 Journal of Applied Information Science Volume 4, Issue 1, June 2016 Utility Mining Algorithms Two-Phase Algorithm Utility mining is useful in marketing, retail and wholesale business. By knowing high utility itemset, marketers can promote their sales. Recent researchers developed many algorithms to calculate high utility itemset. MEU, Two-Phase, FHM, FUM, UP-Growth, CTU-Mine are algorithms that calculate high utility itemset with positive profit only. MHUI-BIT and MHUI-TID, GUIDE, HUPMS and THUI algorithms can discover high utility itemset in data streams. FHN and HUINIV algorithms can handle itemset with negative profits. Liu et al. (2005) proposed Two-Phase algorithm for finding high utility itemsets. In the first phase, a model called “transaction-weighted utilisation mining” was developed. At each level during the level-wise search the combinations of itemsets that have high transaction(Liu weighted utilisation are added into the candidate set. Phase I may overestimate some low utility itemsets, but it never underestimates any itemsets (Liuet al., 2005). In phase II, database is scanned again to remove overestimated itemsets. Fig. 2: Pseudo Code of the 2P Algorithm MEU Algorithm In the paper by Yao et al .(2004), a theoretical model called MEU (Mining with expected utility) is proposed, which finds all itemsets in a transaction database with utility values higher than the minimum utility threshold. MEU prunes the search space by predicting the high utility k-itemset, with the expected utility value. This expected utility value is predicted from the utility values of all items. If utility of an item is greater than this expected value, them it is added to the candidate set, otherwise, it is pruned. This laid the foundation for future utility mining algorithms. The drawback of this algorithm is that the prediction may overestimate. Such thing may lead to high memory space and more computation power. In addition, MEU may miss some high utility itemsets if the variance of the itemset supports is large. Input: database DB. constraintsminUtil and minSup Output: all utility-frequent itemsets /* Phase 1: find all quasi utility-frequent itemsets */ [1] CandidateSet = QUF-APriori(DB, minUtil, minSup) /* Phase 2: prune utility-infrequent itemsets */ [2] foreach c in CandidateSet: [3] foreach T in DB: [4] if c in T and u(c,T) >= minUtil: [5] c.count += 1 [6] return {c in CandidateSet | c.count>= minSup} <Figure head>Fig. 2:Pseudo code of the 2P algorithm THUI A novel method, namely THUI (Temporal High Utility Itemsets) An algorithm for frequent item set mining was presented (Hu & Mojsilovic, 2007). It uses a data structure called High yield partition tree. This algorithm finds segments of data, defined through certain rules. So this algorithm can be used to identify high utility items. It uses “rule-discovery” approach (Hu & Mojsilovic , 2007). This gives a group of patterns that contribute a predefined objective function. MHUI-BIT and MHUI-TID Algorithm In their paper,Li, Huang, Cheng Chen, Liu and Lee(2008) proposed two efficient sliding window based algorithms, MHUI-BIT (Mining High-Utility Itemsets based on BITvector) and MHUI-TID (Mining High-Utility Itemsets based on TID list), for mining high utility itemsets. Bitvector and TID list are used to represent item information. A tree-based summary data structure LexTree-2HTU was developed. This improves the efficiency of mining high utility itemsets. A novel method, namely THUI (Temporal High Utility Itemsets) –Mine was proposed by Tseng (2009), for mining temporal high utility itemsets from data streams. It can effectively discover the temporal high utility itemsets by generating a few temporal high transaction(Tseng, 2009) weighted. This algorithm not only considers profits and quantities but also common branches of items in a multidatabase environment. Utilisation 2-itemsets such that the execution time can be reduced substantially in mining all high utility itemsets in data streams (Tseng, 2009).Thus all temporal high utility itemsets in all time windows are discovered. The advantages of this algorithm are less candidate itemsets, less execution time, low memory space etc. Rare Utility Itemsets Algorithm G.C.Lan et al proposed a new kind of patterns, called Rare Utility Itemsets (Lan, Hong &Tseng, n.d.). The algorithm (2009) Utility Mining Algorithms – A Comparative Study TP-RUI-MD (Two-Phase Algorithm for Mining Rare Utility Itemsets in Multiple Databases) was proposed to retrieve rare utility itemsets in a multi-database environment. Here rare utility itemsets refer to itemsets that occur rarely in a database and have combination of high utility itemsets. The TP-RUI-MD algorithm uses the level-wise technique for discovering the rare-utility itemsets in a multi-database environment (Lan et al., n.d.). 41 Fig. 4: Pseudo Code of the FUFM Algorithm Input: database DB.constraintsminUtil and minSup Output: all utility-frequent itemsets [1] L = 1 [1] [2] L =find 1 the set of candidates of length L with support >= minSup [2] [3] findcomputeexteded the set of candidates of for length L with support support all candidates and >= minSup output utility frequent itemsets FUM Algorithm [4] Increment the value of L by 1 Shankar et al. (2009) present an algorithm known as Fast Utility Mining (FUM). This algorithm retrieves all high utility itemsets. It will not consider the entire set of itemsets in a transaction. FUM considers the distinct itemsets involved in a transaction. Thus the candidate set of combinations of distinct itemsets in a transaction is calculated. This reduces the execution time and improves performance efficiency. It also demands less main memory storage requirements, hence reduced hardware cost.By combining FUM and Fast Utility Frequent mining (FUFM) algorithms authors also propose different types of itemsets such as High Utility and High Frequency itemsets (HUHF), High Utility and Low Frequency itemsets (HULF), Low Utility and High Frequency itemsets (LUHF), and Low Utility and Low Frequency itemsets (LULF). [4] [5] L += use1 the frequent itemset mining algorithm to obtain new set of frequent candidates of length L from the old set of frequent candidates [6] stop if the new set is empty otherwise go to [3] [6] stop if the new set is empty otherwise go to [3] Fig. 5: Pseudo Code of the HUHF Algorithm :Database DB; Constraints minUtil and minSup Input :Database DB; Constraints minUtil and minSup Output : Ranked list utility of customers who buy HUHF items [1] Compute High and high frequent (HUHF) itemsets [1] Compute High utility and high frequent (HUHF) itemsets using FUFM algorithm ∈ [2] For each I ∈ HUHF itemset, scan the database DB to find the customers who buy that itemset [3] Increment the count value associated with the customer who is a buyer of I. [4] Stop if the HUHF is empty else Go to [2] [5] List the HUHF customers in descending order of the count value associated Fig. 3: Pseudocode of the FUM Algorithm with customer [6]each return (list of HUHF customers) [6] return (list of HUHF customers) Input: Database DB, Transaction T,minUtil Output: High Utility ItemsetsH [1] Compute the utility value ∀single itemset <Figure head>Fig. 5: Pseudo code of the HUHF algorithm Fig. 6: Pseudo Code of the HULF Itemsets Mining [2] for each T € DB [3] begin if T ∉S {where S ⊂ DB k S= [0 ..T – 1]} Input: Database DB; Constraints minUtil and minSup [5] begin Output : High Utility and Low Frequency Itemsets (HULF) [6] CandidateSet – CombinationGenerator(T); [1] Compute High utility itemsets HU using FUM algorithm. [7] for each C € CandidateSet [2] Compute High utility and high frequent itemsets HUHF using FUFM algorithm. [4] [8] [9] if(C ∉ H) ^ (U(C,T)>minUtil) [4] return (HULF) H.add(C); [10] [11] [12] [3] HULF = HU -HUHF /*set difference operation*/ begin Fig. 6 end end [13]end Input: Database DB; Constraints min [14]return(H); Output: Low Utility and Low Frequency Itemsets (LULF) CombinationGenerator(T) – Generate all possible combination of itemset€ T. ∈ ∀ ∉S {where ⊆ DB | I = [0 .. T 1]} Input: Database DB; Constraints minUtil and minSup Output : High Utility and Low Frequency Itemsets (HULF) 42 Journal of Applied Information Science Volume 4, Issue 1, June 2016 [3] HULF = HU [4] return (HULF) Fig. 7: Pseudo Code of the LULF Itemsets Mining Algorithm Fig. 6 Input: Database DB; Constraints min_Util and min_Sup, LUHF Output: Low Utility and Low Frequency Itemsets (LULF) [1] Compute the utility value ∀single itemset [2] For eachT ∈ DB [3] begin [4] if T ∉S {where I⊆ DB | I = [0 .. T-1]} [5] begin [6] Candidateset = CombinationGenerator (T) [7] For each C ∈CandidateSet [8] begin [9] if (C ∉ H) ) ^ U(C,T) <min_Util ) [11] Vincent S. Tseng proposed an efficient algorithm, namely UP-Growth i.e Utility Pattern Growth (Tseng et al., 2012), for mining high utility itemsets with a set of techniques for pruning candidate itemsets. UP-Tree (Utility Pattern Tree) is a data structure used in this algorithm to store information of high utility itemsets. Here candidate itemsets are generated within two scans i.e Utility Patternwhen Growth ( compared of the database. It has goodrowth performance Tree(Utility Pattern to other algorithms. Fig. 10: Pseudocode of UP-Span algorithm Input: LU.add (C); [10] UP-Growth Algorithm end [12] [1]CES:complex event sequence; [2]min_utility: minimum utility threshold; [3]MTD: maximum time duration; end [13] end Output: HUE_Set: The complete set of high utility episodes; [14] LULF = LU - LUHF/*set minus operation*/ [1] Scan CES once to find high utility 1-epsiodes and calculate their EWUs and catch [15] return (LULF) the associated minimal occurrences; CombinationGenerator(T) - Generate all possible combinations of itemset∈T <Figure head>Fig. 7: Pseudo code of the LULF itemsets miningalgorithm [2] [3] Fig.8: Pseudo Code of the LUHFI for each global event α do if ( EWU ( α ) min_utility)then MiningHUE( α , moSet ( α ), MTD, min_utility); [4] [5] end if; Input: Database DB; Constraints minUtil and minSup [6] ProcedureMiningHUE( episode α , moSet( α ),MTD, min_utility ) Output: Ranked list of customers who buy Low utility and high frequent items(LUHF) [7] MiningSimultHUE( α , moSet ( α ), MTD, min_utility ); [1] Compute LUHFitemsets using LUHFM algorithm [8] MiningSerialHUE( α , moSet ( α ), MTD , min_utility ); Input: Database DB; Constraints minUtil and minSup [2] For each I ∈ LUHF itemset, scan the database DB to (LUHF) find the customers who buy I. [3] Increment the∈ count value associated with the customer who is a buyer of I. [4] Stop if the LUHF is empty else go to [2] [5] List the LUHF customers in descending order of the count value associated with each customer [6] return (list of LUHF customers) <Figure head>Fig.8: Pseudo code of the LUHFI [6] return (list of LUHF customers) Fig.9: Pseudo Code of the HULFI Customer List Input: Database DB; Input: Database DB; minUtil and minSup [1] Compute High utility and low frequent (HULF) itemsets using HULFM algorithm Output: List of customers who buy High Utility Low Frequent items ∈ [1] Compute High utility and low frequent (HULF) itemsets using HULFM algorithm [2] For each I ∈HULF itemset, scan the database DB to find the customers who buy that itemset [3] Increment the count value associated with the customer whenever the customer buys I. [4] Stop if the HULF is empty else go to [2] [5] List the HULF customers in descending order of the count value associated with customer [6]each return (li [6] return (list of High Utility Low Frequent items customers) 9 9 FHM Algorithm Fournier-Viger et called al. FHM (2014) proposed anMining). algorithm (2014) proposed an algorithm (Faster High Utility itemset It uses called FHM (Faster High Utility itemset Mining). It uses Estimated Utility Co-occurrence Pruning. It pre-calculates the transaction weighted utility measures of 2 itemsets. If . Hence this algorithm avoids 95% of join operation and an itemset contains a 2 itemsetand its transaction weighted utility is less than minutil, then it is considered as low utility itemset. Hence supersets, join of these itemsets are not calculated. Hence this algorithm avoids 95% of join operation and thus has very high performance. Utility Mining Algorithms – A Comparative Study 43 Table 3: Comparison of Utility Mining Algorithms in Data Mining Algorithm Authors Features MEU Two-Phase FHM Yao et al. Liu et al. Fournier-Vigeret al. Based on heuristic value. High utility itemset in traditional database Uses Estimated Utility Co occurrence Pruning. FUM Shankar et al. Handles distinct set UP-Growth Tseng et al. Pruning candidate itemset with two scans. Uses tree structure High utility itemset for pattern growth and dense data Uses tree structure CTU-Mine Erwin et al. Limitation Slow Multiple scans of a database Static database It generates the combinations for the already generated subset of the itemsets. Table 4: Comparison of Utility Mining Algorithms in Data Streams Algorithm Authors Features Limitation MHUI-BIT & MHUT-TID Li et al. Item information, lexTree and HTU for data THUI-Mine GUIDE Tseng et al. Tseng et al. Generates few candidate and high performance Use join operation Individual models for sliding , landmark window and Use MUI tree structure time fading are given. High Performance and low memory usage. HUPMS Ahmed, Tanbeer, Jeong, and Choi Interactive and incremental mining is possible. Conclusion In data mining, association rule mining is one of the most important tasks, that considers mining of frequent itemsets only. It does not consider profit or cost of an itemset. In real life applications there may be some items, which may not be frequent but may have high profit. Utility mining mines high utility itemsets in a transaction databse. It is very beneficial in several real-life applications. In this paper, we have presented a brief overview of various algorithms for high utility itemset mining. Among them FHM algorithm has higher performance. However most of these high utility itemset mining algorithms focus on static databases. Today in many applications, the data is processed in the continuous dynamic data streams. So more research issues now focus on high utility itemset mining in data streams. Also more research focus on high utility itemsets with negative profit. Use join operation Use HUS tree structure References Agrawal, R., Imielinski, T., &Swami, A. (1993). Mining association rules between sets of items in large databases. In Proceedings of the ACM SIGMOD International Conference on Management of Data, (pp. 207-216). Agrawal, R., & Srikant, R.(1994). Fast Algorithms for Mining Association Rules. In Proceedings of the 20th International Conference Very Large Databases, (pp. 487-499). Fournier-Viger, P., Wu, C., Zida, S., Tseng, V. S. (2014). FHM: Faster High-Utility Item-set Mining using Estimated Utility Co-occurrence Pruning. In Proceeding of 21st International Symposium on methodologies for intelligent systems (ISMIS 2014), (pp.83-92). Erwin, A., Gopalan, R. P., & Achuthan, N. R. (2008). Efficient mining of high utility item-sets from large datasets. Advances in Knowledge Discovery and 44 Journal of Applied Information Science Data Mining, 5012, (pp. 554-561). Utility Mining Algorithms - A Comparative Study. Hu, J., &Mojsilovic, A. (2007). High-utility pattern mining: A method for discovery of high-utility item sets. Pattern Recognition, 40(11), 3317-3324. Kanimuthu, S., Premalatha, K., & Shankar, S. (2011). iFUM - Improved Fast Utility Mining. International Journal of Computer Applications, August, 27(11), 32-36. Lan, G. C., Hong, T. P., &Tseng, V. S. (n.d.). An Algorithm for Mining Rare-Utility Itemsets in a Multi-Database Environment. Li, H. F., Huang, H. Y., Cheng Chen, Y., Liu, Y., &Lee, S. (2008). Fast and Memory Efficient Mining of High Utility Item-sets in Data Streams. 8th IEEE International Conference on Data Mining. Liu, Y., Liao, W., & Choudhary, A. (2005). A Fast High Utility Item-sets Mining Algorithm. In Proceedings of the Utility-based Data Mining Workshop. Pujari, A. K. (2001). Data mining Techniques. A text book published by University Press (India) Private limited, (pp. 42). Volume 4, Issue 1, June 2016 Shankar, S., Purusothoman, T. P., Jayanthi, S., & Babu, N. (2009). A Fast Algorithm for Mining High Utility Itemsets. Proceedings of IEEE International Advance Computing Conference (IACC 2009), Patiala, India, (pp.1459-1464). Tseng, V. S., Chu, C. J., & Liang, T. (2006). Efficient mining of temporal high utility item-sets from data streams. In Proceedings of ACM KDD Workshop Utilitybased Data Mining (UBDM’06), Philadelphia, Pennsylvania, USA. Tseng, V. S. (2009). Efficient Mining of Temporal High Utility Itemsets. Tseng, V. S., Wu, C., Shie, B., &Yu, P. S. (2012). Efficient Algorithms for Mining High Utility Itemsets from Transactional Databases. IEEE Transactions on Knowledge and Data Engineering, March, 25(8), 1772 - 1786. Yao, H., Hamilton, H. J., &Butz, C.,J. (2004). A Foundational Approach to Mining Itemset Utilities from Databases. Proceedings of the 3rd SIAM International Conference on Data Mining, Orlando, Florida, (pp. 482-486).