In time recommendations through an Associative Classifier and LookBackApriori: a case study Anna Dalla Vecchia1,*,† , NiccolΓ² Marastoni1,† and Elisa Quintarelli1,† 1 Dept. of Computer Science, University of Verona, Verona, Italy Abstract Recommender systems are becoming essential tools in many scenarios, as they help users extract hidden knowledge and useful insights from datasets. In many real domains, the temporal order between events, combined with their contextualization, improves the accuracy of provided suggestions. In this paper, we introduce a framework designed to mine personalized, in time, contextual, and explainable sequential rules useful to provide recommendations for a predefined target parameter. Specifically, this framework is composed of the 𝐿3 Associative Classifier and LookBackApriori, a modification of Apriori algorithm. Our proposal takes historical data and contextual information as input and generates two sets of rules: the first set comprises rules that allow enhancement of the target parameter, and the second makes it worse. The proposed technique is applied to a real-world scenario involving data collected by Fitbit wearable devices to improve the user’s sleep score after performing fitness activities in different contexts. The idea has been evaluated on two real datasets, and the results confirm the positive effects of the combination of 𝐿3 with LookBackApriori. Keywords Recommendations, Associative classifier, Explanation, Data Mining 1. Introduction conditions in the user location and holidays. More in detail, our use case is based on data gathered The widespread popularity of sensors and wearable de- with Fitbit and focuses on suggesting the intensity of vices, like smartwatches and fitness trackers, has in- physical activities and rest periods to carry out during creased the amount of available data that can be lever- the current day to sleep better. This is done by mining aged to monitor and enhance various aspects of their the historical contextualized physical activities during a users’ well-being. Such devices are often equipped with specified temporal window, which represents the number intuitive apps for activity tracking that mainly provide of consecutive observation days taken into account. aggregate parameters and trend analysis, thus leaving We use two datasets consisting of activity logs from room for more personalized and insightful suggestions to Fitbit wearable devices: PMDataset [1] and a Custom raise the end-users’ awareness about what affects certain dataset. The latter has been collected from four will- monitored parameters and habits. ing participants in the past two years to integrate more To achieve advanced insights, historical data, possi- specific information about the user context. bly integrated with external information describing the The main aim of this paper is the construction of a user context, needs to be analyzed for each user to offer novel recommender system that combines the strengths tailored and context-aware suggestions to improve their of two algorithms: the 𝐿3 associative classifier [2, 3] and life beyond generic recommendations. LookBackApriori (LBA) [4, 5, 6]. The first one allows us In this work, we propose a framework that aims to to predict a specific target parameter based on associative give personalized, and in time, contextual suggestions to classification. It takes as input all the historical physical a specific user to improve a target parameter (e.g., sleep activity and the related contextual information (i.e., the quality) along with an explanation about the provided context at the time the physical activity was performed) suggestions. To achieve this goal we integrate monitored and outputs the predicted sleep score. We leverage the data with contextual information, e.g., current weather second algorithm to provide an explainable recommen- dation about what activity to do and what to avoid to Published in the Proceedings of the Workshops of the EDBT/ICDT 2024 increase the predicted sleep quality and not decrease it. Joint Conference (March 25-28, 2024), Paestum, Italy With this framework, we overcome the limitations of * Corresponding author. † the two algorithms and, in particular, the state explo- These authors contributed equally. $ anna.dallavecchia@univr.it (A. Dalla Vecchia); sion problems of LBA when managing wide temporal niccolo.marastoni@univr.it (N. Marastoni); windows are less severe in L3 . In addition, LBA allows elisa.quintarelli@univr.it (E. Quintarelli) the production of explainable recommendations; indeed,  0000-0001-7026-5205 (A. Dalla Vecchia); 0000-0001-6988-1203 since LBA is based on Apriori, it mines sequential rules (N. Marastoni); 0000-0001-6092-6831 (E. Quintarelli) that contain in their antecedent the explanation of the Β© 2024 Copyright for this paper by its authors. Use permitted under Creative Commons License CEUR Attribution 4.0 International (CC BY 4.0). CEUR Workshop Proceedings (CEUR-WS.org) Workshop Proceedings http://ceur-ws.org ISSN 1613-0073 provided sleep score present in the consequent. CEUR ceur-ws.org Workshop ISSN 1613-0073 Proceedings Prediction Historical Data Associative OutC R+ + Ordering Classifier Contextual Data and Splitting R- Contextual Data Rule Generaor R at time zero Explanation Data at time zero Figure 1: The main components in our framework The document is organized as follows: Section 2 out- LookBackApriori The second part of our framework lines the architecture of our recommender system, while uses the LBA algorithm, which was initially proposed Section 3 presents the case study. Section 4 reports the in [4] and extended in [5, 6]. It is an algorithm based validation of the proposed framework, and Section 5 ex- on Apriori that mines totally ordered sequential rules plores related work. Finally, Section 6 summarizes the and provides timely and explainable recommendations. main contributions and outlines future directions. More in detail, LBA mines association rules where the antecedent is a sequence of itemsets that represent past events or actions by the user with an explicit relative 2. Architecture order w.r.t. the current itemset (representing current day events and actions). The framework proposed in this paper combines an as- For our specific scenario related to wearable devices, a sociative classifier, 𝐿3 , to predict the value of a target rule can be formalized as follows: parameter (e.g., the sleep score for the current day, the stress level) and an algorithm based on Apriori, called π‘Ÿ : πΌβˆ’(πœπ‘€ βˆ’1) ∧ Β· Β· Β· ∧ πΌβˆ’2 ∧ πΌβˆ’1 ∧ 𝐼0𝑓 β†’ 𝐼0𝑠 [𝑠𝑖 , 𝑐𝑖 ] LookBackApriori (LBA), to generate timely and explain- able contextual recommendations helpful to suggest what The rule π‘Ÿ consists of a sequence of itemsets, each rep- to do to improve the predicted value (i.e. the fitness ac- resenting either fitness activities (𝐼 𝑓 ), sleep quality (𝐼 𝑠 ), tivities to undertake in the current day to increase the or both (𝐼), for a specific day, where 0 represents the sleep score, whenever it is possible). current day. The parameter πœπ‘€ defines the temporal win- dow, that is, the number of consecutive days that the Associative classifier The first part of our framework algorithm can consider and thus may be present in a rule. comprises the 𝐿3 associative classifier [2, 3]. 𝐿3 uses The antecedent can contain all the itemsets up to πΌβˆ’1 a technique of lazy pruning to discard those rules that (the day before the current one) or some of them (i.e., the classify training data incorrectly. sequence may be incomplete), while only the physical Then, the classification of unlabeled data is executed activity data is present for the current day. Indeed, this in two steps: first, by considering a subset of high-quality itemset, 𝐼0𝑓 , represents the physical activity suggested to rules for the classification process, and second, by adding the user by our framework to improve their sleep score a larger set of rules when it fails to find rules for certain for the same night, represented in the consequent as 𝐼0𝑠 . data points. The associative classifier 𝐿3 is used for historical data In the green section of Fig. 1, we show the classifier processing and prediction to address memory-related is- component, which is employed for predicting the value sues faced by LBA that stem from the size of the input. of a target parameter, represented in our scenario by the After the sleep score prediction step, LBA is used to pro- sleep score for the current day. It takes historical data vide explainable positive and negative recommendations (i.e., the user’s log of their fitness activities and sleep thanks to the form of the mined rules. To this end, the scores) as input and integrates it with past contextual antecedent contains the sequence of events that will lead information, including the context of the current day at to the sleeping score in the consequent; thus, it provides the time of the prediction. an explanation for the recommendation. Three possible rules mined by the LBA Algorithm are the following: From both datasets, we make use of the logs about "light", "medium", and "heavy" activity, along with rest π‘Ÿ1 : {𝐻𝐴 : 3, 𝐿𝐴 : 2}βˆ’1 ∧{𝐻𝐴 : 3, 𝐿𝐴 : 2}0 β†’ {𝑆𝐿 : 1}0 periods and the sleep score for each day. Fitbit records π‘Ÿ2 : {𝐻𝐴 : 2, 𝑅 : 3}βˆ’2 ∧ {𝐿𝐴 : 1}βˆ’1 β†’ {𝑆𝐿 : 3}0 these features as minutes spent in each activity type; thus, π‘Ÿ3 : {𝐻𝐴 : 2, 𝑅 : 3}βˆ’2 ∧ {𝑀 𝐴 : 1}0 β†’ {𝑆𝐿 : 3}0 we discretize them to obtain categorical data as described in [4]. During this discretization process, the activity π‘Ÿ1 states that if yesterday the user performed a high levels and sleep scores are further split into three sub- level of heavy physical activity (𝐻𝐴 : 3) and a medium levels according to set thresholds, e.g., a heavy activity level of light activity (𝐿𝐴 : 2), and today they perform (𝐻𝐴) can be encoded into three possible labels: 𝐻𝐴 : the same activities, the resulting sleep score will have a 1, 𝐻𝐴 : 2, and 𝐻𝐴 : 3. These represent, respectively, a low value (𝑆𝐿 : 1). low level, medium level, and high level of heavy activity, π‘Ÿ2 is an incomplete rule since it does not contain informa- all decided by the amount of time spent undertaking the tion about the physical activity the user should perform specific physical activity during the day. during the current day. Although the rule is valid, it is Regarding the context, for both datasets, we also have not helpful for making a recommendation to improve information on whether a day falls on a weekend (π‘Š 𝐸) sleep quality, as it does not have any itemset labeled 0 in or not (π‘Š 𝐷). the antecedent. In addition, for the Custom dataset, we integrate the π‘Ÿ3 is another incomplete rule, but it gives us information information about the user’s vacations (𝑉 𝐴/π‘›π‘œπ‘‰ 𝐴) and about the physical activity the user should do during the the weather conditions. For this last aspect, we have current day to sleep well; thus, it can be used to provide simplified its representation as follows: if there has been a recommendation. rain, snow, fog, or other bad weather conditions, the label In the orange part of Fig. 1, we show the contribution is π΅π‘Žπ‘‘. In all the other cases, it is πΊπ‘œπ‘œπ‘‘. Regarding the of LBA to our framework. Firstly, it mines a set of rules temperature, we use the average daily temperature as a 𝑅 using the Rule Generator, which takes as input the feature, and the result is labeled into πΆπ‘œπ‘™π‘‘ or π»π‘œπ‘‘, using current context, also used by 𝐿3 for the prediction step, the yearly average temperature as a threshold. and data at the time of the prediction. Secondly, taking advantage of the label produced by the classifier, the rules generated are split into two sets: those that improve the 3.1. Practical example target parameter value w.r.t. the predicted one is labeled 𝑅+ , and those that do not are labeled π‘…βˆ’ . Then, the rules in both sets are ordered according to the completeness of the rule antecedent, confidence, and support. The rules t-3 t-2 t-1 t0 recommended to the user are the most complete ones with the highest confidence and support. Associative Classifier t0 t-3 t-2 t-1 t0 3. A case study Our experiments focus on wearable device data: their Rule Generator R t0 logs contain information about daily physical activity t0 t0 levels and sleep scores. Whenever possible (i.e., when Training R which we have enough data about the user), we integrate such improve t0 logs with additional information, e.g., holidays, day of R t0 Ordering and Splitting the week, and weather conditions related to the user’s R which deteriorate t0 location, to better contextualize the gathered fitness and sleep quality data. Figure 2: An intuitive workflow of our framework We consider two datasets for this domain: PMdata [1] and Custom. PMdata consists of logs from 16 users, 13 male and 3 female, all aged 23 to 60 years old. The data Fig. 2 shows the workflow of the framework in the was collected for 149 days between November 2019 and specific case of study of this paper. March 2020. First, we need to decide the dimension of the temporal The Custom dataset was collected from 4 users specifi- window to consider to make the recommendation. The cally for this study, the earliest of which started recording temporal window shown in Fig. 2 is 4 days, i.e., the three in August 2021 and ended in September 2022. The partic- previous days are considered along with the current one. ipants are evenly split between males and females; their All the information about the past days is fed to the ages vary from 16 to 55. associative classifier 𝐿3 along with the context. The The relevance of historical data Only last timeslot data Historical data 0.4 0.3 Accuracy 0.2 0.1 0.0 p01 p02 p03 p04 p05 p06 p07 p08 p09 p10 p13 p14 p15 p16 USER1USER2USER3USER4 User Figure 3: The importance of historical data for each user classifier then returns a sleep score for the current day. β€’ testing the efficacy of the proposed framework The rules generated during the training phase state on 𝐿3 w.r.t. to its ability to predict the value of correlations that are only related to the current day, and the predefined target parameter. are in the form shown in the purple rectangle, i.e., con- textual information together with physical activity in the For all the experiments, we have reserved the first antecedent of the rule and the sleep score in the conse- 80% of the Fitbit logs of each user as a training set and quent. the remaining 20% of the data for testing. Due to the Some examples of mined rules are the following: sequential nature of the problem, we cannot randomize the sampling of the two sets. Thus, we maintain the π‘Ÿ1 : {πΆπ‘œπ‘™π‘‘, πΊπ‘œπ‘œπ‘‘, π‘Š 𝐷, 𝐻𝐴 : 1, 𝐿𝐴 : 2}0 β†’ {𝑆𝐿 : 1}0 sequential order based on the timestamp of the logs and π‘Ÿ : {π΅π‘Žπ‘‘, π‘Š 𝐷, 𝐿𝐴 : 2, 𝑀 𝐴 : 3} β†’ {𝑆𝐿 : 3} select the last 20% of the dataset for the tests. 2 0 0 Rule π‘Ÿ1 tells us that on a weekday with clear weather, if the user performs low levels of heavy activity (𝐻𝐴 : 1) 4.1. Relevance of historical data and medium levels of light activity (𝐿𝐴 : 2), their sleep To conduct this part of the experiments, we use the 𝐿3 score for the same night will be low. Whereas rule π‘Ÿ2 associative classifier to predict the sleep label related to states that, in the case of a stormy weekday, the user will the current day 𝑑_0. sleep well after performing medium levels of light activity First, we perform the prediction by selecting only the (𝐿𝐴 : 2) and high levels of medium activity (𝑀 𝐴 : 3). physical activity and context of the current day as input Thanks to the sleep score obtained by the associative to the classifier without considering historical data. An classifier, it is possible to split the rules into those that example of input data for a Custom user is: increase or decrease the predicted sleep score. At this stage, the user can explain why their sleep quality may (πΊπ‘œπ‘œπ‘‘_𝑑0, π»π‘œπ‘‘_𝑑0, π‘Š 𝐷_𝑑0, 𝑉 𝐴_𝑑0, 𝐿𝐴_3_𝑑0, improve or not by looking at the antecedent of the rules. 𝑀 𝐴_1_𝑑0, 𝐻𝐴_1_𝑑0, 𝑅_2_𝑑0) This can be interpreted as: on clear weather and hot 4. Evaluation weekdays during a holiday, the user performs a high To test the validity of our framework, we have performed level of light physical activity, low levels of both medium several experiments on the following aspects: and heavy activity, and a medium level of rest. Then, we add the historical data (i.e., physical activity, β€’ verifying the relevance of historical data and their context, and sleep score of the past days) to the input context in the ability to predict the value of a used for the first set of experiments. target parameter, i.e., the sleep score. Fig. 3 depicts the recorded accuracies of these two β€’ testing the performance of the two algorithms experiments for each user in both PMdata and Custom used by the framework w.r.t. execution time and datasets. These results show that for 77% of users, using memory consumption. historical data instead of only using data from the current Fig. 6 and Fig. 7 depict the elapsed time and the mem- day improves the accuracy of the classifier. ory consumption of the framework when using data from Fig. 4 shows that, regardless of the presence of con- one of the users of the Custom dataset, showing the im- textual information, having historical data improves the portant contribution of 𝐿3 . During the experiments, the accuracy of most users. The performance of the algo- input is the physical activity log and the available con- rithm decays quickly as the temporal window increases, textual information of the chosen user. We maintain the especially in the absence of contextual data. Thus, it same support and confidence and gradually increase the seems clear that sleep quality does not depend on data temporal windows, i.e., increase the historical data given that is temporally distant from the current day. in input. LBA shows an exponential trend for memory consumption and time elapsed, until its memory alloca- Percentage of users where the historical data is better than the only last timeslot data Without context tion fails when the temporal window reaches value 4. 75 70 With context On the other hand, 𝐿3 can manage at most a temporal 65 window of 6, maintaining a relatively constant trend. Percentage 60 55 User: USER2 50 L^3 45 LBA 2 3 4 Temporal Window 5 6 300 Time (s) Figure 4: Percentage of users that obtain higher accuracy with 200 the use of historical data, with different lengths of temporal windows, w.r.t. the use of data of the current day 100 Additionally, the accuracy value obtained by some 0 users increases gradually in the presence of contextual 1 2 3 4 5 6 Temporal window information as the length of the temporal window in- creases. One example is shown in Fig. 5, where we can Figure 6: Execution time with different temporal windows also confirm that sleep does not depend on the activi- for USER2 of Custom dataset. ties performed six days earlier. We can also observe that contextual information improves the final result. User: USER2 Accuracy of the classifier for user p03 with different temporal windows 3500 L^3 0.450 Without Context 3000 LBA 0.425 With Context 2500 Memory (MB) 0.400 0.375 2000 Accuracy 0.350 1500 0.325 1000 0.300 500 0.275 0 1 0.250 2 3 4 5 6 2 3 4 5 6 Temporal window Temporal Window Figure 7: Memory usage with different temporal windows Figure 5: Accuracy trend when we add more historical data, for USER2 of Custom dataset. comparison with and without context 4.3. Evaluation of the framework 4.2. Time and memory performance The last set of experiments performed is on the complete Despite LBA’s intrinsic capability to provide explainable framework, as explained in Sec. 3.1. The idea is to val- recommendations for achieving a better sleep score, we idate, on real user logs, the prediction of 𝐿3 enriched still employ the associative classifier 𝐿3 to process his- with the recommendation produced by LBA. In order to torical data. Thanks to the associative classifier, we can validate the results of the whole framework, the best ap- process significantly larger volumes of data, e.g., longer proach would be to ask for inputs directly from the users. sequences of data with their contextual information, with- Due to time constraints, the evaluation of the framework out the risk of memory errors. Additionally, 𝐿3 is signifi- is strictly empirical. cantly faster in obtaining the results. Accuracy of L^3 associative classifier 0.4 0.3 Accuracy 0.2 0.1 0.0 p01 p02 p03 p04 p05 p06 p07 p08 p09 p10 p13 p14 p15 p16 USER1USER2USER3USER4 User Figure 8: Accuracy of the 𝐿3 associative classifier As before, we use the first 80% of the dataset to train 5. Related Work both 𝐿3 and LBA. The first step is setting the length of the temporal window, which is done empirically by With the spread of smart devices and the availability of analyzing the experiments in the previous case study. their large datasets, we have the possibility to extract The input of the classifier is composed of the physical both explicit and implicit knowledge about monitored activities, contextual information, and sleep score of the parameters. For this reason, there are many intelligent past days and the contextual information of the current techniques proposed in the literature to improve the cus- day. The classifier then predicts the sleep score on the tomization of data exploitation. Recommender Systems current day. (RS) offer suggestions on items, services, or news that Separately, the rule generator produces a set of rules may interest users and affect their decisions based on correlating the physical activities and contextual infor- their profile, history, and preferences [7]. For instance, in mation for the current day (antecedent) and the related [8], the authors develop an RS that can suggest activities sleep score (consequent). targeted to specific users to improve their health condi- At this point, a trained classifier and a set of rules exist tions starting from data collected by a Fitbit wearable for each user. To test the complete framework, we take device. The physical activity information collected by from the remaining 20% of the dataset a temporal window Fitbit is also used in [9] to correlate daily physical activity of observation at a time: all the past data, together with levels with predictions of sleep quality. Neither of the the current context (i.e., the contextual information of mentioned works considers contextual information. the current day), are used by the associative classifier to In the literature, there are many methodologies for predict the sleep score. sleep prediction. In particular, [10] introduces an explain- In Fig. 8, we report both the accuracy of the classifier, able sleep model that exploits the correlation between without any knowledge about physical activity for the daily activities and sleep quality, providing recommenda- current dayIt can be noted that the accuracy of 𝐿3 for tions to improve sleep quality. While the outcome of this most users is less than 0.45. framework aligns closely with our proposal, the approach The output of 𝐿3 produced is used as a threshold to does not account for sequences of events that occurred separate the rules mined by the Rule Generator in the in the days leading up to the prediction intended for the two sets 𝑅+ and π‘…βˆ’ of positive and negative recommen- user. Furthermore, the model presented fails to incor- dations, respectively. Due to the nature of the problem porate external contextual information beyond sensed at hand, it would not be accurate to use historical data to humidity and temperature. As highlighted in Subsection check whether the recommendation given by the frame- 4.1, historical data are important to improve the quality work will actually result in a change in sleep score. of provided predictions. In the state of the art, contextual information is often integrated into RS to improve the precision of recommen- dations. In general, user preferences may vary depending on the environment and the situation in which they are acting [11, 12]. Therefore, Context-Aware Recommender and how to improve it when fitness and sleep parameters Systems (CARS) use contextual information, such as time, are monitored through wearable devices. As future work, location, and social situation, to add knowledge during we are extending the proposal to other domains, like the the recommendation process, thus improving the per- correlation of fitness activities with blood glucose levels. sonalization and the relevance of the suggestion [13]. A systematic literature review is proposed in [14], where the authors describe the integration of the context in RS, References the main categories of contextual features, and the vali- [1] V. Thambawita, S. A. Hicks, H. Borgli, H. K. Stens- dation mechanism with respect to datasets, properties, land, D. Jha, M. K. Svensen, S.-A. Pettersen, D. Jo- metrics, and evaluation protocols. hansen, H. D. Johansen, S. D. Pettersen, et al., Pm- In a lot of real scenarios, the history of events and data: a sports logging dataset, in: Proceedings their order is important and can be leveraged in the pro- of the 11th ACM Multimedia Systems Conference, cess of knowledge extraction. Sequential pattern mining 2020, pp. 231–236. aims to extract hidden sequential patterns in sequential [2] E. Baralis, P. Garza, A lazy approach to pruning databases. These patterns can be used as a first step to classification rules, in: 2002 IEEE International Con- extract frequent sequential rules, usually formalized as ference on Data Mining, 2002. Proceedings., 2002, 𝑋 β†’ π‘Œ [15, 16, 17]. These algorithms propose partially pp. 35–42. doi:10.1109/ICDM.2002.1183883. ordered sequential rules where the antecedent 𝑋 is com- [3] E. Baralis, S. Chiusano, P. Garza, A lazy approach posed of events that happened before the consequent to associative classification, IEEE Transactions on π‘Œ and do not consider the relative order of the events. Knowledge and Data Engineering 20 (2008) 156– For this reason, we develop our approach where the an- 171. doi:10.1109/TKDE.2007.190677. tecedent is a sequence of events with a relative temporal [4] N. Marastoni, B. Oliboni, E. Quintarelli, Explainable label. recommendations for wearable sensor data, in: In- In general, when using algorithms based on Apriori to ternational Conference on Big Data Analytics and predict a target parameter, the dimension of sequences Knowledge Discovery, Springer, 2022, pp. 241–246. one can manage is limited. For this purpose, an analysis [5] A. Dalla Vecchia, N. Marastoni, B. Oliboni, of classifiers is mandatory. In the literature, there are E. Quintarelli, The synergies of context and data several associative classifiers, such as CBA [18], DeEPs aging in recommendations, in: International Con- [19], CMAR [20], CPAR [21], and iCAEP [22]. However, ference on Big Data Analytics and Knowledge Dis- these present two main drawbacks: either a loss of useful covery, Springer, 2023, pp. 80–87. information due to an overpruned set of rules or, con- [6] A. Dalla Vecchia, N. Marastoni, S. Migliorini, B. Oli- versely, a significant growth of the rule set in case of boni, E. Quintarelli, Mining totally ordered sequen- limited rule pruning. For these reasons, we decided to tial rules to provide timely recommendations, in: use the 𝐿3 associative classifier [2, 3], which addresses European Conference on Advances in Databases both problems. and Information Systems, Springer, 2023, pp. 197– It has also been recognized that when users under- 207. stand why certain items have been recommended, the [7] F. Ricci, L. Rokach, B. Shapira, Introduction to rec- suggestion becomes more persuasive. Hence, in recent ommender systems handbook, in: Recommender years, explainable RSs have been applied in real-world systems handbook, Springer, 2011, pp. 1–35. scenarios [23]. However, explanation techniques are of- [8] A. Calero Valdez, M. Ziefle, K. Verbert, A. Felfernig, ten not compatible with some machine learning or deep A. Holzinger, Recommender systems for health in- learning approaches due to their black-box nature [24]. formatics: state-of-the-art and future perspectives, Although there exist some studies on the explainabil- Machine Learning for Health Informatics: State-of- ity, interpretability, and trustworthiness of deep learning the-Art and Future Challenges (2016) 391–414. techniques [25, 26], data mining techniques remain more [9] A. Sathyanarayana, S. Joty, L. Fernandez-Luque, effective in offering comprehensive explanations for the F. Ofli, J. Srivastava, A. Elmagarmid, T. Arora, nature of the rules they are able to infer. S. Taheri, Sleep quality prediction from wear- able data using deep learning, JMIR mHealth and 6. Conclusion uHealth 4 (2016) e125. [10] V. Pandey, D. D. Upadhyay, N. Nag, R. C. Jain, Per- In this paper, we have combined the 𝐿3 associative clas- sonalized user modelling for context-aware lifestyle sifier and the LBA algorithm to provide explainable rec- recommendations to improve sleep., in: HealthRec- ommendations. Such an approach helps give insights to Sys@ RecSys, 2020, pp. 8–14. users who desire to know what affects their sleep score [11] G. D. Abowd, A. K. Dey, P. J. Brown, N. Davies, M. Smith, P. Steggles, Towards a better under- [24] M. Alshammari, O. Nasraoui, S. Sanders, Mining standing of context and context-awareness, in: semantic knowledge graphs to add explainability H. Gellersen (Ed.), Handheld and Ubiquitous Com- to black box recommender systems, IEEE Access 7 puting, First International Symposium, HUC’99, (2019) 110563–110579. Karlsruhe, Germany, September 27-29, 1999, Pro- [25] R. Guidotti, A. Monreale, S. Ruggieri, F. Turini, F. Gi- ceedings, volume 1707 of Lecture Notes in Computer annotti, D. Pedreschi, A Survey of Methods for Science, Springer, 1999, pp. 304–307. Explaining Black Box Models, ACM Comput. Surv. [12] G. Adomavicius, R. Sankaranarayanan, S. Sen, 51 (2018). doi:10.1145/3236009. A. Tuzhilin, Incorporating contextual information [26] R. Meo, R. Nai, E. Sulis, Explainable, Interpretable, in recommender systems using a multidimensional Trustworthy, Responsible, Ethical, Fair, Verifiable approach, ACM Trans. Inf. Syst. 23 (2005) 103–145. AI... What’s Next?, in: In Proceedings of 26th [13] G. Adomavicius, B. Mobasher, F. Ricci, A. Tuzhilin, European Conference on Advances in Databases Context-aware recommender systems, AI Mag. 32 and Information Systems, ADBIS, 2022, pp. 25–34. (2011) 67–80. doi:10.1007/978-3-031-15740-0_3. [14] N. M. Villegas, C. SΓ‘nchez, J. DΓ­az-Cely, G. Tamura, Characterizing context-aware recommender sys- tems: A systematic literature review, Knowledge- Based Systems 140 (2018) 173–200. [15] P. Fournier-Viger, J. C.-W. Lin, R. U. Kiran, Y. S. Koh, A survey of sequential pattern mining, Data Science and Pattern Recognition 1 (2017) 54–77. [16] J. Pei, J. Han, B. Mortazavi-Asl, J. Wang, H. Pinto, Q. Chen, U. Dayal, M. Hsu, Mining sequential pat- terns by pattern-growth: The prefixspan approach, IEEE Trans. Knowl. Data Eng. 16 (2004) 1424–1440. [17] R. Srikant, R. Agrawal, Mining sequential patterns: Generalizations and performance improvements, in: P. M. G. Apers, M. Bouzeghoub, G. Gardarin (Eds.), Advances in Database Technology - EDBT’96, volume 1057 of Lecture Notes in Computer Science, Springer, 1996, pp. 3–17. [18] B. Ma, B. Liu, Y. Hsu, Integrating classification and association rule mining, in: Proceedings of the fourth international conference on knowledge discovery and data mining, volume 50, 1998, pp. 55–00. [19] J. Li, G. Dong, K. Ramamohanarao, L. Wong, Deeps: A new instance-based lazy discovery and classifica- tion system, Machine learning 54 (2004) 99–124. [20] W. Li, J. Han, J. Pei, Cmar: Accurate and efficient classification based on multiple class-association rules, in: Proceedings 2001 IEEE international con- ference on data mining, IEEE, 2001, pp. 369–376. [21] X. Yin, J. Han, Cpar: Classification based on pre- dictive association rules, in: Proceedings of the 2003 SIAM international conference on data min- ing, SIAM, 2003, pp. 331–335. [22] X. Zhang, G. Dong, K. Ramamohanarao, Information-based classification by aggregating emerging patterns, in: International Conference on Intelligent Data Engineering and Automated Learning, Springer, 2000, pp. 48–53. [23] Y. Zhang, X. Chen, Explainable recommendation: A survey and new perspectives, Found. Trends Inf. Retr. 14 (2020) 1–101. doi:10.1561/1500000066.