Abstract—the personalized web page recommendation is much needed these

days. Generally, Web page recommendation systems are implemented in Web

servers. They use data implicitly obtained as a collection of

Web browsing patterns of the users for recommending webpages. The existing

system collects the Web logs and generates a cluster of similar users and

recommends pages to the user by actively analyzing it in online. However the

time complexity for analyzing it in online is more. In order to optimize this

and to improve the correctness of recommendation systems we

propose the method of applying Firefly based algorithm for recommending Web

pages along with Naïve Bayes clustering. It clusters Web logs in

offline using Naive Bayes clustering technique. To find the similarity

between the active user query and other users in the cluster Firefly algorithm

based similarity measure is used. The proposed approach uses a probability

based clustering which eliminates the odd records while forming clusters.

Firefly algorithm meticulously searches the generated web logs present in the

cluster of the active user and recommends the top pages. Firefly algorithm

utilizes time efficiently, thus it can be used for processing in online. When

pages are obtained, they are ranked and the top pages that are more relevant to

the query are recommended. The efficiency of proposed system can be evaluated

using the measures like precision, recall-Score, Matthews’s correlation and

Fallout rate. The proposed approach is expected to improve time

utilization in online process as well as recommends more accurate

Webpages.

Introduction- Web page recommendation system is a sub-domain of recommendation

systems that recommends a set of Web pages to the users based on their past

browsing patterns. It is done by applying special mining techniques on

the data that are previously gathered from the users which in turn discovers and

extract information from Web documents and services. The major concern is about

finding the most accurate recommendation algorithms. Recommendation system typically

produces the result by following one of the two ways – through collaborative and

content based filtering.

A.

Colloborative Filtering

Most recommendation system has wide use

of collaborative filtering for recommending items. This method lies on collecting

and processing the information’s on user’s behaviors or activities and then predicting

the items relating to their similarity with other users. Collaborative filtering

approaches building a structure from a

user’s past behaviors and decisions of other

similar users. This model is then used

to predict items that the user may have an interest in. Since collaborative

filtering does not rely on machine analyzable contents, it is capable

of recommending for complex items accurately without “understanding” of the item

itself.

B. Content Based Filtering

Content based filtering is another common approach

when designing recommendation systems. This technique is based

on a definition of the item and a user’s preferred profile. In a content based recommendation

systems, the keywords are considered as user’s interest. Content based filtering

approaches utilize a series of distinct property of an item in

order to obtain and recommend items with same properties.

These approaches are often combined as Hybrid Recommendation Systems.

These algorithm try to recommend items based on examining

the items that are liked by a user in the past or in the

present. In general, various candidate items are

compared with items previously rated by the user and the best matching

items are recommended.

II. Literature survey

Recommendation system plays a vital role in

recommending personalized items for the users based on their interest in a web services.

The web also contains a rich and dynamic information’s. The amount of

information on the web is growing rapidly, as well as the number of web sites

and webpages per web site. Predicting the needs of a web user as she visits web

sites has gained importance. Many webpage recommendation system were developed in

the past, since they compute recommendations in online process, their time utilization

should be efficient.

A system 4 that uses support

vector machine (SVM) learning based model was

developed for computing similarity between two items which

performed better than latent factor approach for group recommendations.

Since the matrix representation was followed, the

data sparsity problem was solved. However, the system was

not able to stably scale when size of the group

dynamically increased.

Hybrid recommender systems that combines

two or more recommendation techniques was designed 5. It eliminates any

weakness which exist when only one recommender system is used. There are

several ways in which the systems can be combined, such as weighted hybrid

recommender where the score of a recommended item is computed from the results

of all of the available recommendation techniques present in the system.

However, data sparseness was still a problem, the system may generate week recommendations

if few users have rated the same items and also the system doesn’t overcome the

cold start problem.

Hyperspectral sensors can acquire

hundreds of contiguous bands over a wide electromagnetic spectrum for each

pixel. The rich spectral information allows for distinguishing materials with

subtle spectral discrepancy, but it usually leads to the “curse of

dimensionality”. To address this, an improved firefly algorithm based band

selection method 8 was used. The Firefly algorithm is an evolutionary optimization

algorithm proposed by Yang 13. After the initializations of parameters, the

brightness is calculated with the objective function (2.1), where t is the

maximum iterations, ? is the step size and ? is the light absorbance of m

number of fireflies. The moment states are then evaluated and the bands are

selected. In order to avoid employing an actual classifier within the band searching

process to greatly reduce computational cost, criterion functions that can

gauge class separability are preferred which provided better results. Firefly

algorithm also had a faster convergence even at the size of the data is larger

To improve the accuracy of similarity

measure, a nature inspired algorithm which is based in the behaviour of

Fireflies were introduced 10.We consider separate effects for ratings of

users with similar opinions and conflicting opinions. In order to generate

initial population of fireflies, half of population randomly generated and the other

half of population are randomly generated. Mean absolute error was chosen as objective

function to measure recommendation accuracy which is obtained by difference between

predicted rating and real rating. An optimal similarity measure via a simple

linear combination of values and ratio of ratings for user-based collaborative

filtering provides better results. It increased speed of finding nearest neighbours

of active user and reduce its computation time. Similarity function equation

based on Firefly algorithm was simpler than the equation used in traditional

metrics therefore, the proposed method provided recommendations faster than

traditional metrics. Graph colouring problems are generally discrete.

Algorithms to discrete problems are quite complex. A new algorithm based on

Similarity and discretize firefly algorithm directly without any other hybrid

algorithm was developed 11. It was adoptable to dynamic graph sizes.

A

system for assigning an electronic document to one or more

predefined categories or classes based on its textual context and use of

agglomerative clustering algorithm was developed 6. This type of

clustering along with sample correlation coefficient as similarity measure,

allowed high indexing term space reduction factor with a gain of

higher classification accuracy.

In order to minimize noise and outlier

data, a modified DBSCALE algorithm using Naïve Bayes has been designed 7.

This algorithm is basically a prospect based utility. This function is used to

estimate the outlier cluster data and increase the correctness rate of

algorithm on given threshold value. Since Naïve Bayes is a probability based

function, it removes outlier cluster data and increases the correctness rate

according to threshold value. It also computes maximum posterior hypothesis for

outlier data. In order to minimize noise and outlier data, a modified DBSCALE

algorithm using Naïve Bayes has been designed 7. This algorithm is basically

a prospect based utility. This function is used to increase the

correctness rate of algorithm on given threshold value

and to estimate the outlier cluster data. Since Naïve Bayes is a

probability based function, it removes outlier cluster data

and increases the correctness rate according to

threshold value. It also computes maximum posterior hypothesis for

outlier data.

The memory based collaborative system

uses matrix based computation and solves data sparsity problem but, scalability

of the system cannot be stable when size of the group dynamically increases.

Hybrid system could be helpful in overcoming the scalability issue but it again

leads to cold start problem. To eliminate outliers as well as overcoming

other two

problems Naive Bayes clustering, a probability based method

was used in past. Firefly algorithm has a faster convergence and

searches all possible subsets with better time utilization. Thus, to design

an efficient recommendation system, Naïve Bayes method can be

followed for clustering in offline. Since the time complexity

should be less, Firefly algorithm that is more

efficient in terms of time utilization, it can be used for

calculating similarity in online. Combination of these two technique might

increase the accuracy of the recommendation system as well as results in efficient

time utilization.

III. Overview of the proposed

work

Initially, the web log files are obtained

from the 1 America Online Inc. The log files consists of five

fields i.e. anonymous ID for individual user, query of each user along with

query time, list of URLs which user proceeded and its

rank in the result. These logs are collected

and grouped based on anonymous ID. The URL among all

the users are obtained and its content are downloaded and

processed. The processing of data includes removal of stop

words from the URL’s data and keyword extraction. Similar users are clustered based

on fetched keywords by using Naïve Bayes clustering technique which provides efficient

clusters compared to clustering by the use of association rules.

The created clusters are given to online component.

In online process, when an active user gives a query, the keywords from

the query is extracted. The similarity between the extracted

keywords with the other users in the same cluster

of the active user is calculated using Firefly similarity

measure. The similarity values are sorted along with the web pages browsed by

similar users in the cluster. The top k web pages are recommended for the active user

as a result.

IV. proposed work

The proposed system follows a linear

process of initially collecting the web logs and processing them followed by

clustering similar users by Naïve Bayes clustering technique and finally

generating recommendations based on a similarity measure from firefly

algorithm.

A. Preprocessing of Web Logs

The web logs are collected form 1 AOL Inc.

It consists of 20 million web queries from 650 thousand real users over 3

months. The data set includes anonymous ID, query, query time,

item rank and click URL. The log file contains

many number of users along with the web pages visited by

them. It is validated and separated based on anonymous ID. The user

is separated into individual file using anonymous ID. The content from the URL

are fetched and downloaded. Those keywords are processed which undergoes stop

words removal and stemming process. The final keywords are then extracted. The

features like keywords, Timings, Frequency, Click URL and Revisit are fetched.

The user profile is constructed using those features. The user profile that

constructed is based on the features that are taken form the user log files.

·

Timing: The timing that the

user spent on that particular URL

·

Frequency: The amount of time

the user visited the URL

·

Clickstream: The number of

click stream that are visited by user

·

Revisit: Whether the user

visited the web page

The keywords are generated from the

data fetched form the URL. Timing for each URL is estimated

from the given date and time by calculating the difference between the each URL

that are searched in a single day by having some time constraints.

Frequency is hence calculated such that number of times the user clicked the URL.

The clickstreams are those that are clicked by

the user for additional information. The timing of revisit is

calculated such that to decide whether the user preferred it much or

not. Keywords: Keywords are those which are extracted from

the URL. The information from the URL is hence collected and

processed to obtain features of the user.

B. Naïve Bayes

Clustering

Clustering, also known as unsupervised

classification, is a descriptive task with many applications. Clustering is

decomposition or partition of a data set into groups in such a way that the

object in one group are similar to each other but as different as possible from

the object in other groups. Three main approach for clustering of data is

partition based clustering, hierarchical clustering and probabilistic model

based clustering. Probabilistic model based clustering is a soft clustering

were an object can belong to more than one cluster following a probability

distribution. A clustering is useful if it produces some interesting insight in

the problem that we are analysing. Naïve Bayes clustering is also a

probabilistic clustering technique that is based in Bayes theorem with strong

independent assumption between features. The feature variables can be discrete

or continuous. This probabilistic clustering lies on nominal and numeric

variables in the data set and its novelty lies in the use of mixture of

truncated exponential (MTE) densities to model the numeric variables. In Naïve

Bayes clustering the class is the only root variable and all the attributes

are conditionally independent given the class. The clustering problem

reduces to take a data set of instances and a previously

specified number of clusters (k), and work out each cluster’s distribution

and the population distribution between the clusters. To obtain these

parameters the expectation maximization (EM) algorithm is used. Since

Naïve Bayes clustering is a probability based techniques. The items

belongs to the cluster if and only if it has a relation to it.

This helps in eliminating outlier data in the process of clustering. It also

provides proper clustering with less computations. The given dataset is divided

into two parts, one for the training and other for testing. For each record in

the test and train databases, the distribution of the class variable is

computed. According to the obtained distribution, a value for the class variable

is simulated and inserted in the corresponding cluster. The log-likelihood of

the new model is computed. If it is higher than the initial model, the process

is repeated. Otherwise, the process is stopped, obtained clusters are returned.

C. Optimisation Using

Firefly Algorithm

Firefly algorithm is an evolutionary

algorithm that is based on the behaviour of fireflies. Fireflies live in

colonies and cooperate for the survival of the colony. Generally, in order to

model the behaviour of fireflies, three assumptions will always be considered i.e.

all fireflies are homogeneous, Attractiveness of each firefly is related to its

level of brightness, rightness of firefly is determined with an exponential

objective function. Each firefly always emits a kind of light that by which

attracts other fireflies. The amount of accessed light depends on parameters

such as distance and absorption coefficient of the surroundings. The longer the

distance the lesser the amount of accessed light will be. Also in surroundings

with high light absorption coefficient such as foggy weathers, the intensity of

light decreases. The certain issue is that every firefly regardless of its

gender has always been attracted to and moved toward the brighter firefly. Firefly

has a light intensity of its own. The key concept is, the firefly with low light

intensity is always attracted to the firefly with high light intensity. This

concept can be incorporated for calculating similarity. By using firefly based

similarity measure unique and distinguished results can be obtained which is a

useful feature for ranking. It can deal with highly non- linear, multi-modal

optimization problems naturally and efficiently. It does not use velocities,

and there is no problem as that associated with velocity in PSO. The speed of

convergence is very high in probability of finding the global optimized answer.

It has the flexibility of integration with other optimization techniques to

form hybrid tools. It does not require a good initial solution to start its iteration

process.

Each web pages visited by the user i are

considered a firefly. The number of user visited the particular page is assumed

as the light intensity of the firefly. The objective function is formulated

based on the frequency and duration. Frequency is calculated as the ratio to

the number of visits per page to the average vests of all pages. The duration

is the ratio of duration of page to the total duration of all the pages visited

by the user.Thus, the objective function can be defined as in equation 5.1

Interest (i)= 2*Frequency

(i)*Duration (i)

Frequency (i)+Duration (i) (5.1)

The interest of all users in the cluster

is calculated. Then the pages to be recommended are found by using page rank

algorithm 2 on the obtained result. The results after applying page rank

algorithm is given as the recommended web page to the user.

D. 5.2.6 Rankng The Web

Pages

The result, set of web pages obtain

should ranked in an order that the user might have higher interest. Thus, they

are ranked in a sorted order based on the interest of the active user. The

association rule checks the maximum possible combinations which provides more

accurate pages.

E. 5.2.7 Recommendaiton

Process

The URL that are to be recommended will

be identified based on ranking and similarity measure. The similarity measure

is calculated among the users by comparing their similar interest. From the

obtained result of pages, page rank algorithm is used to rank the most relevant

pages to the user. Thus, resultant URL’s are recommended to the users. Hence

the web page that is to be recommended to the user will be more relevant. The

use of Naïve Bayes clustering will eliminate the outliers and Firefly based

similarity calculation will check all the subsets of the clusters.