SECTION I: The product
1a 
The innovation potential of this product is related to:

Maniweb is a technology driven innovation. Many recent applications of machine learning in data mining, computer vision, and elsewhere require deriving a classifier or function estimate from an extremely large data set. Modern data sets of ten consist of a large number of examples, each of which is made up of many features. Though access to an abundance of examples is purely beneficial to an algorithm attempting to generalize from the data, managing a large number of features some of which may be irrelevant or even misleading – is typically a burden to the algorithm. Overwhelmingly complex feature sets will slow the algorithm down and make finding global optima difficult. To lessen this burden on standard machine learning algorithms (e.g. classifiers, function estimators), a number of techniques have been developed to vastly reduce the quantity of features in a data set i.e. to reduce the dimensionality of data.

1b 
What added value for endusers does the product hold? 
The proposed algorithms include Isomap , locally linear embedding (LLE) and its variations, manifold charting , hessian LLE and local tangent space alignment (LTSA) , and they have been successfully applied in several computer vision and pattern recognition problems. Several drawbacks and possible extensions of the algorithms have been pointed out in and the focus of this paper is to address two key issues in manifold learning: 1) how to adaptively select the neighborhood sizes in the knearest neighbor computation to construct the local connectivity; and 2) how to account for the variations in the curvature of the manifold and its interplay with the sampling density of the data set. Those two issues in the context of local tangent space alignment (LTSA), a variation of locally linear embedding (LLE) The basic ideas we proposed can be similarly applied to other manifold learning algorithms. The Web is an enormous information domain with its physical structure best described in the form of a graph. Distinct pages are represented via nodes and the related hyperlinks are represented via graph edges. The graph is readily available to be studied via Weblinkmining, with its outcomes expected to improve the overall experience for the users, save resources for the Web site owners, and provide a better understanding of the Web. Given a large scale graph we should be able to mine interesting patterns if they exist and represent them in way that they can be exploited. Such patterns include the communities (clusters), outlier nodes, and in general the intrinsic geometry/layout of the graph. The problem for smalltomedium graphs (i.e. thousand of nodes) only is dealt with traditional graph mining algorithms and manifold learning techniques with high time complexity.

1c 
What is the Unique Sales Proposition of the potential product? 
One of the stateoftheart algorithms in this area is the CiBC . According to CiBC the nodes of the graph are selected by preferring the ones with the smallest Betweenness Centrality at the bounds of the graph. Then, the nodes are being merged into possibly over lapping communities and the process iterates until no further merging can lead to a better community. Due to its polynomial complexity CiBC is suitable for medium scale graphs on the context of an entire Web site only. As it is focused on optimizing the content delivery of a CDN, CiBC detects communities that most likely capture many requests of a random surfer. Recently, there have been several advances in the machine learning and pattern recognition communities for developing manifold learning algorithms to construct nonlinear lowdimensional manifolds from sample data points embedded in highdimensional spaces. The algorithms that address two key issues in manifold learning: 1) the adaptive selection of the neighborhood sizes; and 2) better fitting the local geometric structure to account for the variations in the curvature of the manifold and its interplay with the sampling density of the data set. It is also illustrate the effectiveness of our methods on some synthetic data sets.

SECTION 2: The Market
2a 
What is the target market for the product? National ¨ European ¨ Global ¨x Please describe the characteristics of your target market. 
On the Web scene, we identify two major target groups that could benefit from ManiWeb: a) the Web site owners and b) the users accessing the Web site, in the following areas: • Web site administration. For instance, a Web site that contains thematic categories and embeds com munity organization could be more easily navigated by users, since once a user reaches a page of in terest, relevant pages could be easily accessed(due to the communities increased linkage density). In a communityabsent Web site, inferior user experience can be easily handled by the administrators who could use the ManiWeb maps to tune usage mining. • Caching and prefetching. Once a 2D display is pro duced, the administrator, by using a “lasso” tool, the revealed communities can be selected and become available as outsourcing units for the CDN. The selected communities are also suitable for caching and prefetching since they can predict users nav igation, due to their dense linkage and the fact that they deal with coherent topics. Therefore, the communities may reduce the latency significantly, if they are placed in a CDN or in a traditional Proxy server. • Information retrieval. The reducedgraph, pro duced by ManiWeb can be exploited as an index structure of the original full data representation graph. In the context of search engines, an algorithm could use the ManiWeb reducedgraph to speed up the initial filtering stage of a query and at the refine ment stage would have to partially retrieve more nodes according to the distributed flow. Another use of ManiWeb could be in interactive recommen dation engines since given an initial user choice, the system can quickly examine the position of this choice in the ManiWeb map and then provide a set of relevant choices ranked by the flow.

2b 
How the product is characterized from the following options? Number of companies producing similar products in the field. • Base – applied by all companies in the industry • Leading – applied by a single or limited number of competitive companies • Key –at a development stage, but has already proven its potential 
ManiWeb is a leading technology. A large number of data such as images and characters under varying intrinsic principal features are thought of as constituting highly nonlinear manifolds in the highdimensional observation space. Visualization and exploration of highdimensional vector data are therefore the focus of much current machine learning research. However, most recognition systems using linear method are bound to ignore subtleties of manifolds such as concavities and protrusions, and this is a bottleneck for achieving highly accurate recognition. This problem has to be solved before we can make a high performance recognition system. Recent years have seen progress in modeling nonlinear manifolds. Rich literature exists on manifold learning. 
2c 
What type of market demand will be satisfied? • Existing demand – the market is already developed • hidden (latent) demand – the market has yet to be developed 
Maniweb satisfies existing demand. Beyond visualization, a dimensionality reduction procedure may help reveal what the underlying forces governing a data set are. Dimensionality reduction has other, related uses in addition to simplifying data so that it can be effi ciently processed. Perhaps the most obvious is vi sualization; if data lies in a 100dimensional space, one cannot get an intuitive feel for what the data looks like. However, if a meaningful two or three dimensional representation of the data can be found, then it is possible to “eyeball” it. Though this may seem like a trivial point, many statistical and machine learning algorithms have very poor optimality guar antees, so the ability to actually see the data and the output of an algorithm is of great practical interest. 
2d 
What is the current stage of the product’s market life cycle? • Implementation, implementation in production (leading to a radically new product offers) • Growth (rapid spread within the industry or outside it) • maturity (parameters of the technical characteristics of manufactured products reached their maximum, highergrade products can be manufactured on the basis of technological substitution) 
Mani web is at development stage. Given a set of cooperating surrogate servers, a set of clients, and a Web site, the problem is to identify Web communities and to place them inside the limited capacity caches of the surrogate servers. By distributing the Webcommunities in the surrogate servers we pro actively predict the clients’ requests, as a random surfer will most likely visit pages of the same community. A good community is the one that predicts many requests and it is as small as possible. At this point we give a possible practical application of ManiWeb in the context of CDNs by comparison with the CiBC algorithm. Using the National Geographic graph we applied the CiBC algorithm resulting into 98 communities. Using the same number of communities we performed kmeans clustering on the ManiWeb 2D result and the resulting clusters are the considered communities for ManiWeb. Using the same tool that CiBC was originally tested on we have generated 150000 client transactions. A transaction is a sequence of Web pages that a random surfer could visit if she followed the Web graph linkage probabilistically.

2e 
Strategic partnerships (existing or potential). 
Strategic partnership can be made with large Internet service provider (ISP) and web analytics as the industry bodies have been trying to agree definitions that are useful and definitive for some time. The main bodies who have had input in this area have been JICWEBS (The Joint Industry Committee for Web Standards in the UK and Ireland), ABCe (Audit Bureau of Circulations electronic, UK and Europe), The WAA (Web Analytics Association, US) and to a lesser extent the IAB (Interactive Advertising Bureau). 
SECTION 3: The Competition
3a 
What is the competition within your target market? 
Best Web Analytics Tools: Clickstream Analysis Tools, Google Analytics (google.com/analytics) Yahoo Web Analytics (web.analytics.yahoo.com) Crazy Egg (crazyegg.com) Compete (compete.com) Google Website Optimizer (google.com/websiteoptimizer) Optimizely (optimizely.com) Kissinsights from Kiss Metrics 
3b 
What competitive advantages will the introduction of the new product ensue? • lower prices based on lower production costs • product differentiation (uniqueness of the product proposal) 
The product differentiation is the distinct competitive advantage of Mani Web. ManiWeb could be exploited by almost every topic relevant to dimensionality reduction, graph embedding, data mining, and visualization. This is true since when ever Laplacian Eigenmap is suitable, ManiWeb can be applied too with appropriate tuning.

3c 
Potential products relate to the following price range: • High price range • Average price range • Low price range 
Mani web will be marketed at average price range. 
3d 
Potential products will be marketed: • To regulated markets (e.g. heat supply, water supply, universal telecommunication services, agricultural products, fishing industry, architectural services) • To markets operating on the principle of free negotiation between agents on the market 
The market of Maniweb operating on the principle of free negotiation between agents on the market. For example Kissinsights from Kiss Metrics (kissinsights.com) – is Free to $29/month, ClickTale (clicktale.com) is Free to $99/month, Crazy Egg (crazyegg.com) – $9$99/month, Mani web will be marketed at 25 euros per month. 
SECTION 4: Indicators
Estimated cost of the new products 
100.000 
Expected market volume (potential / maximum number of users) 
5000 
Expected sales volume 
1.500.000 
Expected market share of the company (proportion between sales and total company sales in the relevant market) 
0.001% 
KEYWORDS QUANTITATIVE ASSESSMENT (05)..
Please put X as appropriate.
1 
2 
3 
4 
5 

Addedvalue potential 




x 
Size of future market demand 




x 
Competitive positioning of the product 




x 