Jing Fan

Mapping Elements with the Hungarian Algorithm: An Efficient Method for Querying Business Process Models

Efficient query processing over a large amount of business process models is important for managing the business process model repository. The structural similarity between two process models is considered as the main measurement for ranking the process models for a given search model. Current business process query methods are inefficient since too many expensive computations of the graph edit distance are involved for constructing the elements mapping as well as deriving the structural similarity. To address this, using Petri-net as the modelling method, this paper presents the Hungarian algorithm based query method, where we firstly define the context similarity for a pair of place nodes that are from different process models by taking into account both the common paths and common transitions, then transform the elements (e.g., The transitions and the places) mapping to classical assignment problem that can be solved by Hungarian algorithm efficiently. In this way, we can save a lot of time for searching the best combination of elements mapping. Finally, we use the common method of the graph edit distance to measure the structural similarity based on the found best combination of elements mapping.

Querying Similar Process Models based on the Hungarian Algorithm

The structural similarity between two process models is usually considered as the main measurement for ranking the process models for a given query model. Current process query methods are inefficient since too many expensive computations of the graph edit distance are involved. To address this issue, using Petri-net as the modeling method, this paper presents the Hungarian algorithm based similarity query method. Unlike previous work where the non-task nodes (i.e., place nodes in the Petri-net) were lightly studied or even ignored, we think these non-task nodes also play an essential role in measuring the structural similarity between process models. First, we extract the context for each place and define the similarity for a pair of place nodes that are from different process models from two perspectives: commonality and the graph edit distance. Then, the place mapping is transformed to classical assignment problem that can be solved by Hungarian algorithm efficiently. Furthermore, we propose a new process similarity measurement on the basis of the place similarity. The extensive experimental evaluation shows that our Hungarian based methods outperform the baseline algorithm in both retrieval quality and query response time.

Toward accurate energy-efficient cellular network: switching off excessive carriers based on traffic profiling

The carrier based energy saving (CBES) method, i.e., switching off the excessive carriers when traffic load is light, can help network operators obtain significant energy savings in cellular network. In practice, however, due to the lack of corresponding traffic classifications, current CBES method uses same traffic threshold for switching off all the BS carriers, which limits its effectiveness. In this paper, we seek to illuminate that the CBES method could be improved in a more precise way and make cellular network more energy efficient. Based on the collected large volume of 3G data set, we profile the traffic in sector granularity and classify them. With the help of machine learning approach, three traffic classes are derived and for each of them, we articulate corresponding strategy in high-level vision for improving CBES method. The simulation results for energy saving show the feasibility of our strategies.

Predicting e-book ranking based on the implicit user feedback

In this paper, we plan to predict a ranking on e-books by analyzing the implicit user behavior, and the goal of our work is to optimize the ranking results to be close to that of the ground truth ranking where e-books are ordered by their corresponding reader number. As far as we know, there exist little work on predicting the future e-book ranking. To this end, through analyzing various user behavior from a popular e-book reading mobile APP, we construct three groups of features that are related to e-book ranking, where some features are created based on the popular metrics from the e-commerce, e.g., conversion rates. Then, we firstly propose a baseline method by using the idea of learning to rank (L2R), where we train the ranking model for each e-book by taking all its past user feedback within a time interval into consideration. Then we further propose TDLR: a Time Decay based Learning to Rank method, where we separately train the ranking model on each day and combine these models by gradually decaying the importance of them over time. Through extensive experimental studies on the real-world dataset, our approach TDLR is proved to significantly improve the e-book ranking quality more than 10% when compared with the L2R method where no time decay is considered.

Dynamic Pricing for Resource Consumption in Cloud Service

This paper studies dynamic pricing for cloud service where different resources are consumed by different users. The traditional cloud resource pricing models can be divided into two categories: on-demand service and reserved service. The former only takes the using time into account and is unfair for the users with long using time and little concurrency. The latter charges the same price to all the users and does not consider the resource consumption of users. Therefore, in this paper, we propose a flexible dynamic pricing model for cloud resources, which not only takes into account the occupying time and resource consumption of different users but also considers the maximal concurrency of resource consumption. As a result, on the one hand, this dynamic pricing model can help users save the cost of cloud resources. On the other hand, the profits of service providers are guaranteed. The key of the pricing model is how to efficiently calculate the maximal concurrency of resource consumption since the cost of providers is dynamically varied based on the maximal concurrency. To support this function in real time, we propose a data structure based on the classical B+ tree and the implementation for its corresponding basic operations like insertion, deletion, split, and query. Finally, the experiment results show that we can complete the dynamic pricing query on 10 million cloud resource usage records within 0.2 seconds on average.

A Benchmark Dataset for Evaluating Process Similarity Search Methods

Process similarity search is an effective way tomanage a large number of business process models. However,there exists no benchmark dataset that can be used to evaluate theperformance of the existing process similarity search algorithms.To solve this problem, we have constructed a benchmark datasetthat modeled by Petri-net. In this paper, the benchmark datasettotally consists of 100 process models, where we have markedout 10 search models and their corresponding 10 relevant models(including itself). And for each search model, the ranking orderof its relevant models is derived from user studies. The datasetand the codes of corresponding similarity search algorithms areavailable to the public on a website1.

FB-Diff: A Feature Based Difference Detection Algorithm for Process Models

Detecting difference between process models is important for many business process management scenarios, such as process version control and process merging. However, it is far from trivial to detect the process difference. Existing work suffers from drawbacks like inappropriate data structure support or expensive computation. In this paper, we propose FB-Diff, a feature-based difference detection approach. Firstly, a semi-ordered tree model called task based process structure tree (TPST) is used to represent a process model, which can correctly describe the structure as well as the behavior (the execution sequence of task nodes). Then FB-Diff adopts a divide and conquer strategy to find the similar parts of two TPSTs. Specically, we divide the TPST into fragments that are represented by feature vectors. A feature vector consists of six features, and each feature describes a key characteristic of the fragment. Based on the similar parts, the edit script that can transform one TPST into the other is generated. The extensive experimental evaluation shows that our method can meet the real requirements in terms of precision and efficiency.

Covering the Optimal Time Window Over Temporal Data

In this paper, we propose a new problem: covering the optimal time window over temporal data. Given a duration constraint d and a set of users where each user has multiple time intervals, the goal is to find all time windows which (1) are greater than or equal to the duration d, and (2) can be covered by the intervals from as many as possible users. This problem can be applied to real scenarios where people need to determine the best time for maximizing the number of people to be involved in an activity, e.g., the meeting organization and the online live video broadcasting. As far as we know, there is no existing algorithm that can solve the problem directly. In this paper, we propose two algorithms to solve the problem, the first one is considered as a baseline algorithm called sliding time window (STW), where we utilize the start and end points of all users - intervals to construct time windows satisfying duration d. And then we calculate the number of users whose intervals can cover the current time window. The second method, named TLI, is designed based on the the data structures from the Timeline Index in SAP HANA. In TLI algorithm, we conduct three consecutive phases to achieve the purpose of efficiency improvement, namely construction of Timeline Index, calculation of valid user set and calculation of time windows. Within the third phase, we prune the number of time windows by keeping track of the number of users in current optimal time window, which can help shrink the search space. Through extensive experimental evaluations, we find TLI algorithm outperforms STW two orders of magnitude in terms of querying time.

Detecting Difference between Process Models Based on the Refined Process Structure Tree

The development of mobile workflow management systems (mWfMS) leads to large number of business process models. In the meantime, the location restriction embedded in mWfMS may result in different process models for a single business process. In order to help users quickly locate the difference and rebuild the process model, detecting the difference between different process models is needed. Existing detection methods either provide a dissimilarity value to represent the difference or use predefined difference template to generate the result, which cannot reflect the entire composition of the difference. Hence, in this paper, we present a new approach to solve this problem. Firstly, we parse the process models to their corresponding refined process structure trees (PSTs), that is, decomposing a process model into a hierarchy of subprocess models. Then we design a method to convert the PST to its corresponding task based process structure tree (TPST). As a consequence, the problem of detecting difference between two process models is transformed to detect difference between their corresponding TPSTs. Finally, we obtain the difference between two TPSTs based on the divide and conquer strategy, where the difference is described by an edit script and we make the cost of the edit script close to minimum. The extensive experimental evaluation shows that our method can meet the real requirements in terms of precision and efficiency.

Using Classification Method for Querying the Relevant Process Models

Operations management is important to a company, so more and more business process models are created. At the same time, how to manage such a large amount of process models is becoming a big challenge for companies. Querying the relevant process models is proposed as a business process management technology and it has attracted more and more attention by researchers. The existing methods query the relevant models for a query process model by measuring their similarities. And most of them measure the similarity by focusing on only one kind of feature, such as the structural features or behavioral features, while ignoring other features. In this paper, we consider both structural features and behavioral features to query the relevant process models for a query process model. In order to reach this goal, we use two classification methods named back propagation neural network (BPNN) and support vector machines(SVM) for classifying the candidate models in the repository into two classes: relevant and irrelevant. For the sake of classification, we summarize 7 features to represent the similar or dissimilar parts of two process models. The experiment result shows the precision and efficiency of the classification methods are acceptable.