http://dspace.cityu.edu.hk:80/handle/2031/743 Tue, 04 Jun 2013 16:21:56 GMT 2013-06-04T16:21:56Z http://dspace.cityu.edu.hk:80/handle/2031/6595 Title: Automatic testing of student programs using token patterns Authors: Tang, Chung Man ( 鄧頌民) Abstract: Assessing students' programming exercises manually can be tedious and error-prone. To relieve these problems, many universities have developed software systems to automatically assess students' programs, usually by executing them to test their functional correctness, typically through simple matching of the program output with expected output. This approach requires that the precise and detailed form of the output be given in the specification of the exercise, because even a small and insignificant deviation (such as an extra whitespace) from the expected output will cause the matching to fail. An undesirable pedagogical consequence of this technical limitation is that it may unnecessarily confine the way of implementation, perhaps inhibit creativity, and distract students from the essentials of the exercise. To handle slightly deviated but acceptable program outputs, some automated systems employ tailored output comparison techniques that match certain kinds of output variants. However, existing techniques are mostly ad hoc and limited in capabilities. There is a need for more sophisticated techniques that can tolerate reasonably acceptable output variants. Towards this goal, we propose a novel approach, based on token patterns, that splits the output texts into smaller units for more fine-grained output comparison. As a result, the program output comparison can be more flexible and the accompanying undesirable side effects on teaching and learning can be reduced. An experimental prototype was initially built to evaluate our proposed approach. The prototype was then further integrated into a production system used in a programming course for empirical evaluation. The evaluation results are encouraging, showing that our proposed new approach is feasible and promising in improving student satisfaction in the use of the automated system, as well as producing better outcomes in terms of teaching and learning. Notes: CityU Call Number: QA76.6 .T36 2011; xii, 187 leaves : ill. (some col.) 30 cm.; Thesis (M.Phil.)--City University of Hong Kong, 2011.; Includes bibliographical references (leaves 140-143) Sat, 01 Jan 2011 00:00:00 GMT http://dspace.cityu.edu.hk:80/handle/2031/6595 2011-01-01T00:00:00Z http://dspace.cityu.edu.hk:80/handle/2031/6224 Title: Topology control in wireless mesh networks with directional antennas Authors: Zheng, Zhongming (鄭忠明) Abstract: With the development of wireless networks, many studies have been done to improve the network performance. Among those studies, wireless mesh networks have become a hot research interest in recent years. A wireless mesh network is an infrastructure wireless network composed by gateway nodes, non-gateway nodes and end-users. The back bone of the mesh networks is formed by gateway nodes and non-gateway nodes with wireless links. Gateway nodes connect to Internet by wired links. The end users directly connect with mesh back bone by wireless links, and access the Internet by mesh back bone. We found that only limited studies about topology control by using directional antennas have been done. One interesting study to optimize performance by using directional antennas was proposed by Kumar et al. In their work, a topology control algorithm was presented to create a tree with bounded degree, based on the approximation algorithm of minimal degree spanning tree. However, the throughput is not the direct objective in their algorithm, which means that their algorithm may not be optimal in throughput. Based on the work of Kumar et al, we formulate the throughput of each node as traffic delivery ratio. With each node equipped with m non-steerable directional antennas, two topology control algorithms for the wireless mesh networks are presented . Our objective is to maximize the minimal traffic delivery ratio from each node to the gateway nodes in both algorithms, by adjusting the orientations of the directional antennas appropriately. The MTDR (maximizing traffic delivery ratio) algorithm is based on an approximating degree bounded algorithm. In every step, the algorithm chooses a pair of links to be added and removed, so that the number of the maximal degree nodes is decreased by one. Then, the maximal degree of the topology will be decreased gradually until it is equal or smaller than the number of equipped directional antennas. We show by simulation that the MTDR algorithm improves the throughput significantly, compared with Kumar et al’s work. Moreover, we study how each parameter, such as the beam width of antennas, and the number of antennas on each node, has impact on the network performance. Many advantages of directional antennas are not utilized in the MTDR algorithm. We have improved our topology control method and proposed the EMTDR (extended maximizing traffic delivery ratio) algorithm for the single channel wireless mesh networks with m directional antennas equipped on each node. The objective of the EMTDR algorithm is the same as the MTDR algorithm. By allowing each antenna to connect to multiple nodes, the EMTDR algorithm improves performance by 40% ~ 280%, compared with the MTDR algorithm. Notes: CityU Call Number: TK5103.2 .Z46 2010; ix, 55 leaves : ill. 30 cm.; Thesis (M.Phil.)--City University of Hong Kong, 2010.; Includes bibliographical references (leaves 50-55) Fri, 01 Jan 2010 00:00:00 GMT http://dspace.cityu.edu.hk:80/handle/2031/6224 2010-01-01T00:00:00Z http://dspace.cityu.edu.hk:80/handle/2031/6223 Title: Design and analysis on network throughput enhancement in integrated fiber-wireless (FiWi) access networks Authors: Zheng, Zeyu (鄭澤宇) Abstract: Nowadays, a hybrid fiber-wireless access network (FiWi) has been proposed to integrate the optical PON network and the wireless mesh network (WMN) to provide the high bandwidth, cost efficient and ubiquitous last mile Internet access. For the PON subnetwork of FiWi networks, it consists of an Optical Line Terminal (OLT) at the central office which sends traffic received from the access network to the Internet and vice versa, an Remote Node (RN) which multiplexes the upstream traffic destinated to the OLT and vice versa, and a group of Optical Network Units (ONUs) close to users' premises which send the upstream traffic to the RN and vice versa. For the wireless subnetwork of FiWi networks, WMN is applied to support ubiquitous and flexible communications in users' premises. Generally, WMN consists of multiple gateways connected to the wired Internet, a group of wireless mesh clients that associate with those routers. In FiWi networks, the integration of PONs and WMNs enables ONUs to combine functions of both traditional ONUs in PONs and gateways in WMNs together. For the newly emerged FiWi access networks, the network throughput is a very attractive issue and hasn't been quite investigated yet. Therefore in our work, we intend to study the achievable network throughput in FiWi networks and factors that affect the throughput. Considering the specific architectural features of FiWi networks, its throughput can be affected by the following aspects: the traffic demands pattern from wireless mesh clients, the traffic routing algorithm applied in the wireless mesh subnetwork, the TDMA schedule of ONUs in the PON subnetwork and the deployment of ONUs in FiWi networks. In this work, we will address above mentioned aspects for the design and analysis on the throughput enhancement in FiWi networks. As for traffic demands from wireless mesh clients, considering traditional traffic demand that goes to the Internet, the TDMA schedule of ONUs for the upstream traffic and the routing algorithm applied in the wireless mesh subnetwork play very important roles if the high throughput is expected. In the TDMA schedule, ONUs share the uplink capacity to the OLT and each ONU can only use its assigned time slots for the upstream traffic transmission. In order to achieve higher throughput, it's better if the ONU that occupies the current transmission time slot has traffic loads for transmission, otherwise, such time slot will be wasted without any contribution to the throughput. Therefore, it's better to send traffic to the ONU whose assigned time slots are nearly coming so that the capacity assigned to that ONU can be well utilized, thus enhancing the throughput. On the other hand, at each ONU, the actual arrival rate of traffic from the wireless subnetwork has much effect on the achievable throughput. It's better if traffic goes through the wireless path with less interferences in the wireless subnetwork, which may result in higher throughput. In our work, we proposed the Interference Aware and Delay Bounded Routing (IADBR) algorithm to send traffic to the ONU from which the traffic can be sent out to the Internet as soon as possible along the wireless path with reduced interferences. We propose both centralized and distributed algorithms for IADBR and simulation results show that the distributed algorithm performs quite closely to the optimal centralized algorithm and performs much better than the shortest path algorithm. Based on above work and many other work like Alichery, Kodialam on the network throughput optimization, we observe that when only traffic that goes to the Internet is considered, the achievable network throughput of FiWi networks is quite bottle-necked by interferences in the wireless subnetwork. However, when peer-to-peer communications from one wireless client to another wireless client is introduced, the integration of PONs and WMNs in FiWi networks provides an opportunity to reduce the impact of interferences on network throughput. In traditional WMNs, peer-to-peer communication is carried in the wireless network, which is subject to interferences in wireless communications. However, in FiWi networks, such traffic can be carried through the wireless-optical-wireless mode in which the traffic is sent from the source wireless client to its nearest ONU, which is then sent to the ONU close to the destination wireless client through the PON subnetwork and then delivered to the destination wireless client. Such wireless-optical-wireless mode introduced by FiWi networks can sustain the interference in wireless subnetwork, thus improving the network throughput. In our work, we study the network throughput gain in FiWi networks subject to peer-to-peer communications compared with traditional WMNs and parameters that can affect the throughput gain. Simulation results show that with heavy peer-to-peer communication traffic, the network throughput gain in FiWi networks is significant compared with the traditional WMN. Furthermore, based on above work and other work like Sarkar on the deployment of ONUs on FiWi networks, we also observe that the ONU deployment will have great impact on the network throughput in FiWi networks considering peer-to-peer communications and the ONU deployment will be different from that when only traffic to the Internet is considered. In our work, given the distribution of wireless mesh routers, we study where to place K ONUs in FiWi networks so that the overall network throughput can be maximized considering peer-to-peer communications. We proposed a Tabu Search (TS) based heuristic for the problem solving. Simulation results show that compared to the random deployment and the fixed deployment which performs well when only traffic to the Internet is considered, Tabu Search heuristic has a much better performance with much enhanced network throughput. Notes: CityU Call Number: TK5105.5956 .Z45 2009; xi, 65 leaves : ill. 30 cm.; Thesis (M.Phil.)--City University of Hong Kong, 2009.; Includes bibliographical references (leaves 60-64) Thu, 01 Jan 2009 00:00:00 GMT http://dspace.cityu.edu.hk:80/handle/2031/6223 2009-01-01T00:00:00Z http://dspace.cityu.edu.hk:80/handle/2031/6220 Title: Cooking recipe manipulations : modeling, organization, personalized search and recommendation Authors: Yu, Lijuan (余麗娟) Abstract: Cooking is a daily and necessary activity in our real life. Recipe relevant applications, such as resource organization, browsing, search and recommendations, are of great values to users who like or enjoy cooking greatly. In this thesis, we start our work from recipe data modeling. Cooking recipes can be viewed as a kind of complex data containing rich information like ingredients, seasonings, cooking methods, tastes, nutrition, etc, many of which are difficult to be represented by simple data structures. In our model, we divide recipe features into three categories: cooking features, nutrition features, flavor and other features according to users' concerning aspects over recipe information, and employ a hybrid scheme consisting of three parts to model these features comprehensively. The second problem addressed in this thesis is to provide a personalized content organization schema for recipe resources. We set up a folksonomy environment for collecting user annotations, and index the recipe resources using tags. Based on the semantic network, we propose three types of personal views for content organization, named as Media View, Semantic View, and Personalized View, respectively. Next, we move on to the personalized recipe search strategy in a folksonomy environment that can help a user quickly find out his/her desired recipe(s) with simple hints. In such an environment, users are invited to tag their favorite recipes using interested terms, and by aggregating such interactions it enables the system to build tag-based user profiles. Meanwhile, each recipe may receive a list of collaboratively edited tags from multiple users, describing its semantic Cooking is a daily and necessary activity in our real life. Recipe relevant applications, such as resource organization, browsing, search and recommendations, are of great values to users who like or enjoy cooking greatly. In this thesis, we start our work from recipe data modeling. Cooking recipes can be viewed as a kind of complex data containing rich information like ingredients, seasonings, cooking methods, tastes, nutrition, etc, many of which are difficult to be represented by simple data structures. In our model, we divide recipe features into three categories: cooking features, nutrition features, flavor and other features according to users' concerning aspects over recipe information, and employ a hybrid scheme consisting of three parts to model these features comprehensively. The second problem addressed in this thesis is to provide a personalized content organization schema for recipe resources. We set up a folksonomy environment for collecting user annotations, and index the recipe resources using tags. Based on the semantic network, we propose three types of personal views for content organization, named as Media View, Semantic View, and Personalized View, respectively. Next, we move on to the personalized recipe search strategy in a folksonomy environment that can help a user quickly find out his/her desired recipe(s) with simple hints. In such an environment, users are invited to tag their favorite recipes using interested terms, and by aggregating such interactions it enables the system to build tag-based user profiles. Meanwhile, each recipe may receive a list of collaboratively edited tags from multiple users, describing its semantic features. By building up connections between the tag-based user and recipe profiles, it facilitates the goal of personalized search. The fourth relevant issue addressed by this thesis is to devise a personalized recipe recommendation strategy. The basic idea of our approach is to blend the content-based and collaborative filtering methods, with the goal of exploring the folksonomy to identify interest-similar users. As a result, not only can it overcome the 'cold start' problem, but also will the resultant system keep improving over time with more users joinning and becoming members of it. As a part of this dissertation research, we have conducted some empirical studies on a real data set based upon a prototype recipe system that we have implemented, so as to evaluate our approach. The experiment results demonstrate the validity and efficiency of our proposed methods for both personalized recipe search and recommendation. Notes: CityU Call Number: TX643 .Y8 2010; vii, 68 leaves : ill. 30 cm.; Thesis (M.Phil.)--City University of Hong Kong, 2010.; Includes bibliographical references (leaves 61-68) Fri, 01 Jan 2010 00:00:00 GMT http://dspace.cityu.edu.hk:80/handle/2031/6220 2010-01-01T00:00:00Z