Event Pattern Analysis and Prediction at Sentence Level using Neuro-Fuzzy Model for Crime Event Detection

Abstract

Classifying the sentences that describe Events is an important task for many applications. In this paper, Event patterns are identified and extracted at sentence level using term features. The terms that trigger Events along with the sentences are extracted from web documents. The sentence structures are analyzed using POS tags. A hierarchal sentence classification model is proposed by considering specific term features of the sentence and the rules are derived. The rules fail to define a clear boundary between the patterns and create ambiguity and impreciseness. To overcome this, suitable fuzzy rules are derived which gives importance to all term features of the sentence. The fuzzy rules are constructed with more variables and generate sixteen patterns. Artificial Neuro-Fuzzy Inference System (ANFIS) model is proposed for training and classifying the sentence patterns for capturing the knowledge present in sentences. The obtained patterns are assigned linguistic grades based on previous classification knowledge. These grades represent the type and quality of information in the patterns. The membership function is used to evaluate the fuzzy rules. The patterns share the membership values between [0–1] which determines the weights for each pattern. Later, higher weighted patterns are considered to build Event Corpus, which helps in retrieving useful and interested information of Event Instances. The performance of the proposed approach classification is evaluated for ‘Crime’ Event by crawling documents from WWW and also evaluated for benchmark dataset for ‘Die’ Event. It is found that the performance of the proposed approach is encouraging when compared with recently proposed similar approaches.

Appendix- I

1.1 Verification of proposed Fuzzy Rules using Fuzzy Petri Nets (FPN)

In general, anomalies in the rules are referred to as verification errors. They are inconsistency, incompleteness, redundancy, etc., and should be identified and removed for the error free rules. Inconsistency in rules generates conflict results, say, the rules R1: p1$\wedge$ p2 → p3, R2: p3$\wedge$ p4 → p5, R3: p1$\wedge$ p2$\wedge$ p4 → ¬p5 are said to be inconsistent. Incompleteness in rules is generated by missing rules. For example, R1: p1$\wedge$ p2 → p3 is said to be incomplete if p1 is a fact and p2 is neither a fact nor a conclusion of other rule. Consider rules R1: p1$\wedge$ p2 → p3, R2: p2$\wedge$ p1 → p3 and these rules are redundant, since, they refer unnecessary rules in a rule base. Similarly, rules are said to be subsumed when two rules have identical conclusions. Consider the rules R1: p1$\wedge$ p2 → p3, R2: p2 → p3 where the antecedents of one rule is subset of the antecedents of another. Finally, if the rules have circular dependency say, R1: p1 → p2, R2: p2 → p3, R3: p3 → p1, it is said to be circular rule.

Based on the above discussion, it is imperative that the rule base constructed for any application domain should be validated for having a perfect rule set. In this section, we have used Fuzzy Petri Nets (FPN) for verifying the proposed sentence classification model (SCM). The FPN is a combination of Petri Nets and rule-based knowledge representation. Discussing and explaining FPN is out of scope of this paper and interested reader can refer related research materials [26]. The FPN for the proposed SCM is introduced within a 5-tuple consisting of the input property set (IPS), internal property set (InPS), output property set (OPS), and rule set (RS). Question 1 (Q1) to Question 13 (Q13) represent Input Properties such as term feature(s, et, ct _(I) , and ct _(NI) ) and POS features (verb, noun, adjective, and adverb) for the Internal Properties “NQASS”, “SQASS”, “QASS”, “HQASS” patterns, respectively.

• i) The Input Properties

The Input Properties are gathered within 13 questions and represented as Q1 to Q13 below:

Q1: is s is verb?	Q8: is ct (I) is adverb?
Q2: is s is non-verb?	Q9: is ct (I) is not adverb?
Q3: is et is noun?	Q10: is ct (NI) is adjective?
Q4: is et is adjective/verb?	Q11: is ct (NI) is not adjective?
Q5: is et is adjective?	Q12: is ct (NI) is adverb?
Q6: is ct (I) is adjective?	Q13: is ct (NI) is not adverb?
Q7: is ct (I) is not adjective?

Though complete Input Properties set for all 16 rules of Subjective and Objective classes are given, for want of space and clarity, we consider the rules for the patterns of Subjective Active class (ASS) alone in further discussion and, however, can be extended for other class also.

• ii) The Internal Properties

The Internal Properties of the SCM are derived using various Input Properties as shown below:

• 1. The Input Properties Q1, Q4, Q9, and Q13 form an Internal Property called “NQASS” pattern.

• 2. The Input Properties Q1, Q4, Q9, and Q12 form an Internal Property called “SQASS” pattern.

• 3. The Input Properties Q1, Q4, Q8, and Q13 form an Internal Property called “QASS” pattern.

• 4. The Input Properties Q1, Q4, Q8, and Q12 form an Internal Property called “HQASS” pattern.

Below, Input and Internal Properties are deduced and presented in two levels:

1) Level 1:

If Q1 && Q4 && Q9 && Q13 exist then “NQASS” pattern

If Q1 && Q4 && Q9 && Q12 exist then “SQASS” pattern

If Q1 && Q4 && Q8 && Q13 exist then “QASS” pattern

If Q1 && Q4 && Q8 && Q12 exist then “HQASS” pattern

2) Level 2:

If “NQASS” && “SQASS” && “QASS” && “HQASS” patterns exist then sentence set belongs to “ASS” sub-class. Patterns are substituted with membership values. The sample rule base for the proposed SCM is presented below in the structured format:

SCM = (Classification, IPS, InPS, OPS, RS)

SCM.IPS = {Q1, Q2, Q3, Q4, Q5, Q6, Q7, Q8, Q9, Q10, Q11, Q12, Q13}

SCM.InPS = {NQASS, SQASS, QASS, HQASS,}

SCM.OPS = {ASS}

SCM.RS = {R1, R2, R3, R4, R5}

SCM.RS.R1 = {Rule1, P1$\wedge$ P3$\wedge$ P9$\wedge$ P13, P14, 0.25}

SCM.RS.R2 = {Rule2, P1$\wedge$ P3$\wedge$ P9$\wedge$ P12, P15, 0.50}

SCM.RS.R3 = {Rule3, P1$\wedge$ P3$\wedge$ P8$\wedge$ P13, P16, 0.75}

SCM.RS.R4 = {Rule4, P1$\wedge$ P3$\wedge$ P8$\wedge$ P12, P17, 1}

SCM.RS.R5 = {Rule5, P14$\vee$ P15$\vee$ P16$\vee$ P17, P18}

1.1.1 Verification process

For verifying the rule base, it is mapped to an FPN as shown in Fig. 11. A reachability graph is generated using the ω-Net concept algorithm presented in [26]. The verification parameters, such as incompleteness, inconsistency, circularity, and redundancy, are verified using the analysis presented in [26].

Fig. 11

Fuzzy petri net representation of the sentence classification model

Based on the obtained FPN, the reachability graph is presented in Fig. 12 and the following conclusions are drawn:

Fig. 12

Reachability graph

• 1. All the places (P) and transitions (T) exist, so there are no incompleteness errors.

• 2. P14, P15, P16, and P17 are different states of one property, so no inconsistency.

• 3. There is no loop in the reachability graph, and thus there is no circularity.

• 4. Finally, there is no redundancy; since, there are no transitions underlined.

It is observed that the fuzzy rules proposed in this paper (Table 6a, b) are verified using FPN for identifying anomalies in the rule base. The reachability graph obtained using well-known algorithm found that there are no anomalies in the rule base.