International Journal of Computer Trends and Technology- May to June Issue 2011

ONLINE MESSAGE CATEGORIZATION USING APRIORI ALGORITHM

G.SenthilKumar, 2S.Baskar, , 3 M. Rajendran. Associate Professor, , Panimalar Engineering College, Chennai 2 PG Scholar, Panimalar Engineering College, Chennai 3 Associate Professor, , Panimalar Engineering College, Chennai
1 1

ABSTRACT Online message categorization using Apriori algorithm analyzes the selection and assignment of topical context. It presents several techniques to decrease the hypothesis space and heuristics that apply specifically to military reporting environments. This allows accurate assignment of topical context even when the context is only implied rather than given explicitly. With the explosive growth in online Communication is necessary to organize the information for faster and easier processing. Usually, a company receives huge number of emails to the single mail address. Automatic categorization of all incoming emails should be extremely useful because it can help route an email to the right person. is to find out whether this classification can be used to automatically route incoming mails to the person in charge using an association rule mining methodology and apriori algorithm. I. INTRODUCTION The objective of is to find out whether the classification of emails can be used to automatically route incoming Emails to the right person in charge. Categorizing messages into different topics or groups introduces a structure which helps a user to prioritize the E-mail. Categorization can also facilitate the search for a particular message as it can limit search to only several topics .As a Sample, considered 3 main categories for classification, which includes enquiry, purchase orders, and feedback about the organization/company products. The Association based classification technique is used to classify the incoming emails. The main phases are a. E-mail Preprocessing b. Building the Associative Classifier c. Testing the Associative Classifier

Figure: 1.1.Text Categorization Flow art E-mail preprocessing should be done before the email is given as input to any of the other two phases.It involves the process

ISSN:2231-2803

- 75 -

IJCTT

International Journal of Computer Trends and Technology- May to June Issue 2011
of transforming the training dataset into a representation which is suitable for Apriori Algorithm.To prepare the dataset a sample of 3000 emails from Gmail with enabled POP3 protocol are captured through Postie plug-in (PHP based plug-in).These e-mails are parsed to collect the words II. RELATED WORK In this, it integrates methods from two different areas of research: email classification and temporal data mining. Numerous studies on email classification appeared in the machine learning literature in the last few years . Most of the approaches put in use only word-based features, completely ignoring the temporal aspect of the email domain. At the same time, Sahami et al. It showed that bringing in other kinds of features (spam-specifc features in their study) could improve the classifcation results. To alter spam messages, they join together two sets of features, traditional textual and non-textual such as overemphasized punctuation, the domain type of the sender, time the message was received, etc. Our work explores the possibility of the usage of more complex time-related features in a general email classifcation context. In the data mining field there have been a lot of research on mining sequential patterns in large databases of customer transactions: from the first Apriori-like algorithms AprioriSome and AprioriAll to their numerous modifications and extensions. Similar approaches were proposed for Discovering sequential patterns, also called frequent episodes, in sequences. Recently, more complex temporal sequential patterns were investigated. Works consider interval-based sequences where events last over a period of time (as opposed to point-based events). In this study, the work with point-based events but concentrate on time elapsed between the events. Proposal of a new algorithm MINTS that intends temporal sequential patterns consisting of not only event types, but also the time intervals between the events. Therefore, the method predicts not only the expected event in a sequence, but also when the event is likely to happen. Work by Kleinberg applies a temporal analysis in the context of email. His assumption is that the appearance of a new topic in a document stream is signaled by a "burst of activity". He presents a formal framework for identifying such "bursts" and shows that the document structure imposed by this analysis has a natural meaning in terms of the content. However, research on temporal data mining focuses only on the temporal aspect of data and does not take into account any contentbased features. The main contribution of this paper is an attempt to integrate methods from the two areas in one powerful heterogeneous system. III. APRIORI ALGORITHM Apriori is designed to operate on databases containing transactions. Other algorithms are designed for finding association rules in data having no transactions or having no timestamps. The algorithm attempts to find subsets which are common to at least a minimum number C of the itemsets. Apriori uses a "bottom up" approach, where frequent subsets are extended one item at a time (a step known as candidate generation), and groups of candidates are tested against the data. The algorithm terminates when no further successful extensions are found. Apriori, while historically significant, suffers from a number of inefficiencies or trade-offs, which have spawned other algorithms. Candidate generation generates large numbers of subsets (the algorithm attempts to load up the candidate set with as many as possible before each scan). (a) Phases of knowledge discovery Data selection. Data cleansing.

ISSN:2231-2803

- 76 -

IJCTT

International Journal of Computer Trends and Technology- May to June Issue 2011
Data enrichment(integration with additional resources). Data transformation or encoding. Data mining. Reporting and display (visualization) of the discovered knowledge. Confidence(Certainity): Certainty of a rule can be measured with a threshold for confidence.This parameter lets to measure how often an events item set that matches the left side of the implication in the association rule also matches for the right side. Rules for events whose item sets do not match sufficiently often the right side. While matching the left (defined by a threshold value) can be excluded. Database D consists of events T1, T2, Tm, that is: D = {T1, T2,, Tm} Let there be a rule Xa Xb so that itemsets Xa and Xb are subregions of event Tk, that is: Xa Tk Xb Tk Also let Xa Xb = . The confidence can be defined as sup(Xa Xb) conf(Xa,Xb) = -------------------sup(Xa) This relation compares number of events containing both itemsets Xa and Xb to number of events containing an itemset Xa .

(b) Objective Measures Based on threshold values controlled by the user. Some typical measures: o o o Simplicity: Focus on generating simple association rules. Length of rule can be limited by userdefined threshold. With smaller item sets the interpretation of rules is more intuitive. Unfortunately this can increase the amount of rules too much. Quantitative values can be quantized (for ex. age groups). Support(Utility): Usefulness of a rule can be measured with a minimum support threshold. This parameter lets to measure how many events have such itemsets that match both sides of the implication in the association rule. Rules for events whose itemsets do not match boths sides sufficiently often (defined by a threshold value) can be excluded. Database D consists of events T1, T2, Tm, that is D = {T1, T2,, Tm}. Let there be an itemset X that is a subregion of event Tk, that is X Tk. The support can be defined as | {Tk D | X Tk} | ---------------------------|D| This relation compares number of events containing item set X to number of all events in database. sup(X) = Simplicity Support (utility) Confidence (certainty)

Figure 1.2. - Architecture of Associative Classifier

ISSN:2231-2803

- 77 -

IJCTT

International Journal of Computer Trends and Technology- May to June Issue 2011
Support and confidence: If confidence gets a value of 100 % the rule is an exact rule. Even if confidence reaches high values the rule is not useful unless the support value is high as well Rules that have both high confidence and support are called strong rules. Some competing alternative approaches (other that Apriori) can generate useful rules even with low support values IV GENERATING RULES ASSOCIATION Any (k 1) size itemset that is not frequent cannot be a subset of a frequent k size itemset, hence should be removed. Frequent set Lk has been achieved. Algorithm uses breadth-first search and a hash tree structure to make candidate item sets efficiently. Then occurrence frequency for each candidate item set is counted. Those candidate item sets that have higher frequency than minimum support threshold are qualified to be frequent itemsets APRIORI ALGORITHM IN PSEUDOCODE L1= {frequent items}; for (k= 2; Lk-1 !=; k++) do begin Ck= candidates generated from Lk-1 (that is: cartesian product Lk-1 x Lk-1 and eliminating any k-1 size itemset that is not frequent); for each transaction t in database do increment the count of all candidates in Ck that are contained in t Lk = candidates in Ck with min_sup end return k Lk; Main steps of iteration are: Find frequent set Lk-1 . Join step: Ck is generated by joining Lk-1 with itself (cartesian product Lk-1 x Lk-1).

Prune step (apriori property):

Usually consists of two sub problems (Han and Kamber, 2001): Finding frequent itemsets whose occurences exceed a predefined minimum support threshold. Deriving association rules from those frequent itemsets (with the constrains of minimum confidence threshold). These two subproblems are soleved iteratively until new rules no more emerge. The second subproblem is quite straightforward and most of the research focus is on the first subproblem. Use of Apriori algorithm Initial information: transactional database D and user-defined numeric minimun support threshold min_sup. Algortihm uses knowledge from previous iteration phase to produce frequent itemsets. This is reflected in the Latin origin of the name that means from what comes before. Creating frequent sets Lets define: Ck as a candidate itemset of size k Lk as a frequent itemset of size k

ISSN:2231-2803

- 78 -

IJCTT

International Journal of Computer Trends and Technology- May to June Issue 2011

Application Of Apriori Algorithm On Online Messages The Apriori algorithm for nding all large item sets makes multiple passes over the database. In the rst pass, the algorithm counts item occurrences to determine large one-item sets. The other passes consist of two steps. First, the large item sets L k1 found in the (k 1)-th pass are used to generate the candidate item sets Ck ; next, all those item sets which have some k 1 subset that is not in Lk1 are deleted, yielding Ck . Once the large item sets are obtained, rules of the form a (l a) are computed, where a l and l is a large item set.

Figure 1.3. - Categorization Flow chart The Apriori Algorithm is used for mining frequent item-sets in transactional databases .The output generated from above 2 phases is used as an input for Apriori Algorithm. By considering these various items and categories in emails. I1->request/data/informational/others I2->discount/product/problem/others I3->delivery/defect/others I4->balance/satisfy/feedback/others etc., The output of Apriori algorithm is: I1->I3->I5->I7->I9->I11->C1 (Enquiry) I2->I4->I13->I15->I17->C2 (Purchase) I3->I120->I21->I12->C3 (Feedback) These rules will be used in testing stage.

ISSN:2231-2803

- 79 -

IJCTT

International Journal of Computer Trends and Technology- May to June Issue 2011
V. PROPOSED SYSTEM In Proposed system, Email preprocessing involves the process of transforming the process of transforming the training dataset. In Emails certain words are most frequent and are not discriminative of message contents. Hence reduce the size and complexity of the data in the system, which is always advantageous. While working on the experiment I have observed that the execution time is more than 60 seconds to generate the output files for 50,000 email sets. As a future work, it can be reduced by writing special algorithms. Automatic reply to the incoming mails that are categorized is also being carried out along with enabling this application to work in Mobile Environment. In Existing system, the Emails classification with the increase in growth of Email communication. Usually, the companies receive huge no of emails to the single email address. This kind of emails should be responded within a certain amount of time by a responsible person VI. MODULES USED (a) LOGIN Module (b) CLIENT Module (c) ADMIN Module (d) MAIL CLASSIFICATION Module (a) Login Module First user want to login through the Username and password. Only Authorized Person Enter into the Login page Transfer the file.

(c) Admin Module Admin receives the file. Collect all details about transmission. Finally send all the details.

the

(d ) Mail Classification Module Admin Receive the all files classify the subject vise Details. The mails separate the wanted unwanted mails easily find out then wanted mails only send destination. and and and the

VII. REPORT GENERATION The user shares the online messages and the details are stored in database. We can generate the report and display from the database. VIII. PERFORMANCE ANALYSIS The reason behind using different traffic classes is due to the observation that subscribing content providers can be highly heterogeneous in terms of their traffic patterns and the type of content they handle. Hence, end-users request for content of varying sizes (ranging from small to large). The processing requirements also vary based on size of the content requested. The use of different traffic types allows us to reflect different user preferences and content request types. After using available number of resources in the current site based on the additional demand given by the users. The user shares the media and the details are stored in database. We can generate the report and display from the database IX. CONCLUSION Use cases for personal email assistant, multi-facet classication. Email classication approaches, ideas for new

(b) Client Module First user want to login through the name and password. Have to browse the file and select destination. Click on send button.

ISSN:2231-2803

- 80 -

IJCTT

International Journal of Computer Trends and Technology- May to June Issue 2011
ones. Information extraction (in emails). Emails and knowledge intensive tasks in workows.Emails are automatically extracted and folderised, so that the corresponding person will handle the emails.This reduces the burden of the manager of organization/company and faster services will be provided due to the earliest reply of incoming emails.Hence this proves that association rule mining can be used in automatic categorization effectively and efficiently. Training is relatively fast.Advantage of the association rule based classifier is that it does not assume that the terms are independent and also all the rules are human understandable. X. REFERENCES [1]. Aas, k. and A.Eikvil, 1999.Text categorization: A survey. Technical report, Norwegian Computing Center, June [2] Agarwal, R. T.Imielinski and A.swami, 1993. Mining association rules between sets of items in large databases., Proceedings of 1993 ACM-SIGMOD International Conference on Management data, Washington, DC.,pp: 207-216 [3] Han, J.and M.Kamber,2001.Data Mining:Concepts and Techniques,Morgan Kanfmann publishers [4] Jie Tang,Hang Li,Yunbo Cao and Zhaohui Tang,2005.Email data cleaning.KDD05,Chigago,USA. [5] Mccallum, A.and K.Nigam,1998. A Comparison fo event Models for Native bayes text classification, AAAI WS Mehtodology of applying Machine Learning. [6] Segal,R.B.and Jeffrey O.kephart,1999. Mail Cat: An intelligent assistant for organizing email. Third International Conference on Autonomous Agent. Yang, Y. and X. Liu,1999.

ISSN:2231-2803

- 81 -

IJCTT