No full-text available
To read the full-text of this research,
you can request a copy directly from the authors.
Application of improved MFNN on dynamic computing for case-intelligence recommendation system
Abstract
Personalized recommendation involves a process of gathering and storing information about website visitors, from which user's characteristic knowledge is exploited to satisfy the personalized needs. Facing the difficulty of timely identifying new data computing in updating real-time user behaviors, we propose a case-intelligence system framework along with a feature-based multi-layer feed-forward neural networks (MFNN) approach to personalized recommendation that is capable of handling the massive with dynamic data effectively. Our experimental results indicate that better performance in our recommender comes from the both sides: the one is that our MFNN has understandable, constructive and reliable process, unlike the black box of the other ANN networks; the other is our covering algorithm can decrease the complexity of ANN algorithm effectively.
... The former cases can also be used to evaluate the new issues and new programs of problem-solving [9], and prevent the potential errors in the future. Cases can be reused by similarity computing as case knowledge space conversion, which is the glorious with exciting highlight in the construction of CBR intelligent system, and the characteristic advantage distinguishes CBR systems from RBR systems thoroughly [10]. ...
How to suggest a valid recommend within a reasonable time is the greatest technical challenge for the recommendation system, for which tremendous user cases with high dimension are generated while it runs in real time, and these massive data are too difficult to compute directly. This paper proposes a case-intelligence system framework along with a feature-based multi-layer feed-forward neural networks (MFNN) to succeed case-retrieval based on dynamic computing, which constructs the neural networks dependence on the real input vectors instead of the fixed and dull networks structure presupposed, and can apply many kinds of knowledge granularity from various levels effectively to help users for information retrieval and case adaptation. Our subsequent experimental results indicate that it is capable of handling the massive personalized data, and our covering algorithm can decrease the complexity of MFNN algorithm for dynamic computing, which performs adaptable knowledge granularity to enhance the system's efficiency of reasoning.
In recent years, the number of recommender systems used in online shops has strongly increased and is be-coming an important success factor for electronic com-merce applications. A recommender system can be uti-lized to suggest similar related or potentially interest-ing products for a given customer or a set of products for a marketing campaign. This paper shows how sea-sonal products can be identified and included in the rec-ommendation process to improve the quality of a rec-ommender system. The approach was tested using real life data from two companies selling working protection and tools.
In Web-based services of dynamic content (such as news arti- cles), recommender systems face the diculty of timely iden- tifying new items of high-quality and providing recommen- dations for new users. We propose a feature-based machine learning approach to personalized recommendation that is capable of handling the cold-start issue eectively. We main- tain profiles of content of interest, in which temporal charac- teristics of the content, e.g. popularity and freshness, are up- dated in real-time manner. We also maintain profiles of users including demographic information and a summary of user activities within Yahoo! properties. Based on all features in user and content profiles, we develop predictive bilinear regression models to provide accurate personalized recom- mendations of new items for both existing and new users. This approach results in an oine model with light computa- tional overhead compared with other recommender systems that require online re-training. The proposed framework is general and flexible for other personalized tasks. The supe- rior performance of our approach is verified on a large-scale data set collected from the Today-Module on Yahoo! Front Page, with comparison against six competitive approaches.
The equivalent between kernel functions based SVM (Vapnik) and the three-layer feedforward neural network is demonstrated. From the covering algorithms of neural networks that author proposed, a kernel function existence theorem is proved. The theory shows that given a set of training samples, there must exist a corresponding function such that the image of the training samples is linear separated in a high dimensional space under the mapping of the function. An algorithm of seeking the kernel functions is given. The computational complexity of the algorithm is polynomial growing with the sample size and the solution is the maximal margin one in the high dimensional space.
The acquisition of personalized need is key to effective recommendation. Case intelligence is a comprehensive expression which is integrated representation of human sense, logics and creativity. Through the former cases we can acquire users' preferences which are implicit in case-base, the process of recommendation is easy to understand and accept. As E-commerce is in complex environment, cases are regarded as the foundation for knowledge representation in case intelligent system and may be represented in semi-structured or unstructured model, or even in natural language texts. This paper presents a personalized recommendation system based on case intelligence. The system has good flexibility, uses modular components to integrate various artificial intelligence technologies, which is convenient to acquire revision knowledge from huge cases from multi-channels. At last, this article proposes how to explore the revision knowledge and characteristic adaptation methods, so we can improve the quality of recommendation and “support” the users effectively.
The task of recommending documents to knowledge workers differs from the task of recommending products to consumers. Variations in search context can undermine the effectiveness of collaborative approaches, while many knowledge workers function in an environment in which the open sharing of information may be impossible or undesirable. There is also the 'cold start' problem of how to bootstrap a recommendation system in the absence of any usage statistics. We describe a system called ResultsPlus, which uses a blend of information retrieval and machine learning technologies to recommend secondary materials to attorneys engaged in primary law research. Rankings of recommended material are subsequently enhanced by incorporating historical user behavior and document usage data.
In this paper, we investigate two novel indexing schemes called D- HS and D-HS+PSR(II) designed for use in case-based reasoning systems. D-HS is based on a matrix of cases indexed by their discretised attribute values. D- HS+PSR(II) extends D-HS by combining the matrix with an additional tree-like indexing structure to facilitate solution reuse. DHS+PSR(II)'s novelty lies in its ability to improve retrieval efficiency over time by reusing previously encountered solution patterns. Its benefits include its accuracy, speed and ability to facilitate efficient real time maintenance of retrieval knowledge as the size of the case-base grows. We present empirical results from an analyses of 20 case-bases and demonstrate the technique to be of similar competency to C4.5 yet much more efficient. Its performance advantages over C4.5 are shown to be especially apparent when tested on case-bases which grow in size over time.
The last three years have seen a dramatic increase in both awareness and exploitation of Web Application Vulnerabilities. 2008 and 2009 saw dozens of high-profile attacks against websites using Cross Site Scripting (XSS) and Cross Site Request Forgery (CSRF) for the purposes of information stealing, website defacement, malware planting, clickjacking, etc. While an ideal solution may be to develop web applications free from any exploitable vulnerabilities, real world security is usually provided in layers. We present content restrictions, and a content restrictions enforcement scheme called Content Security Policy (CSP), which intends to be one such layer. Content restrictions allow site designers or server administrators to specify how content interacts on their web sites-a security mechanism desperately needed by the untamed Web. These content restrictions rules are activated and enforced by supporting web browsers when a policy is provided for a site via HTTP, and we show how a system such as CSP can be effective to lock down sites and provide an early alert system for vulnerabilities on a web site. Our scheme is also easily deployed, which is made evident by our prototype implementation in Firefox and on the Mozilla Add-Ons web site.
educing the Dimensionality of Data Streams using Common Sense
- Catherine Havasi
- Jason Alonso
- Robert Speer
Case-based Reasonor Based on Multi-Layered Feedforward Neural Network
- Jianyang Li
- Zhiwei Ni