What’s the point of insurance policies if not to file a claim on the compensation when things go wrong? Claims are a stressful procedure, especially after one has heard the horror claim filing experiences from friends and families. Insurance companies have indeed stepped up in their efforts to minimize the stress and time between filing your claim and having it settled successfully. This is most notable with 48hrs being the industry standard.

According to the Insurance Regulatory Authority’s Annual Report found on their website www.ira.go.ke, Kenya has 54 registered insurance companies in August 2019 with the insurance penetration rate at 2.43%. The total gross direct premium received in the year 2018 was approximately Ksh.127.323 billion shillings which were shared across all the insurance companies. Now for some interesting statistics. The insurance Regulatory Authority report shows that the regulator registered 2,233 complaints in 2018 compared to the 2,126 registered in 2017. 74 % of these were from the General insurance business. The main reasons for the complaints were:

1. Delayed settlements being the biggest chunk of complaints of declined claims
2. Erroneous deductions
3. Unsatisfactory compensation
4. Motor Vehicle Insurance and Medical Insurance were the main segments with the largest share if incurred claims.

The main question becomes, “How do insurers minimize the rising number of complaints, without being defrauded?”. The claims department exists to protect against wrongful or fraudulent claims and as such, there may be unintended errors or slowing down of the process. The answer lies in leveraging increased computing power, specialized data management techniques and an organization’s internal controls to solve this problem.

The recent developments in AI, particularly in Natural Language Processing, can be harnessed to improve the paperwork associated with claim processing. Natural language processing (NLP) can enhance the paperless insurance workflow by automating document flow. NLP could easily replace the manual process of the claim processing for most general insurance claims in terms of the documents involved.

The benefits of Adopting Natural Language (NLP) Processing Technology.

1. Minimizes human error.
2. Improves claims processing time leading to improved customer retention and satisfaction When integrated with various machine learning algorithms the program can flag
3. Fraudulent claims.
4. Enhances data privacy and security.

How to use Adopting Natural Language (NLP) to process an insurance claim form?

Machine Learning and rule-based logic can be utilized to process data from the claim forms. Some of the documents required in claim processing are template claim form, police abstract, copy of your driving license of the driver at the time of the accident, copy of national identity card and copy of Kenya Revenue Authority certificate. The process depends on the data collected by the insurer. Internal policy and regulations also affect the order and necessity of some of these steps.

1. Document unwrapping; The claim forms scanned into a mail folder are monitored to determine the document type e.g. claim form, police abstract, invoice.
2. Image processing; Basic image enhancements are applied to the claim documents to make them text searchable and readable.
3. Classification and extraction; Assemble all pages split in the above process into single documents and Categorize them based on full content analysis. No template for every single document needs to be built.
4. Identify and extract key pieces of metadata using pre-configured rule-based logic in machine learning.

This pre-defined rule-based logic include;

• Key-value pair extraction; This feature allows a user to specify a ‘Key-Value pair’ which can be used for extracting document level index field values based on the relative location of ‘value’ against a specified key.
• Pattern Matching using regular expressions - Involves checking and locating of specific sequences of data of some pattern among raw data or a sequence of tokens.
• Table Extraction; Table extraction has three key elements. The first is to locate a table in a document. The second is to determine the structures within the table, such as headers, footnotes, etc. The third is to associate the various elements of a table with their related data cells to create cell documents.
• Validation; Verification is done based on the information retrieved by the above methods. A confidence score can be established to check the level of correctness of the information. If the score falls below a certain number, the system may require human intervention.

1. Validation- verification is done based on the information retrieved by the above methods. A confidence score can be established to check the level of correctness of the information. If the score falls below a certain number, the system may require human intervention.
2.  Export content to a record management system- The texts extracted are sent to the client’s email in the form of a report or to a repository for action.

References

1.Table extraction for answer retrieval (2006) by Xing Wei, Bruce Croft, Andrew McCallum

2.https://ephesoft.com/docs/2019-1/moduleplugin-configuration/extraction-module/key-value-extraction-4040/

Introduction

Unstructured data is data that is not stored in a fixed record length format. This includes documents, social media feeds, digital pictures, and videos, etc. 80% of most organizational data is unstructured. This article aims to explore basic steps of Natural Language Processing.

What is Natural Language Processing?

Natural Language Processing is a branch of artificial intelligence that deals with how computers communicate with human beings through a natural language. It gives the computer the ability to understand interpret and utilize a human language.

What are the applications of Natural Language Processing?

Natural language processing is the driving force in the following applications;  

• Speech Recognition
• Machine translation
• Search automation
• Survey analytics
• Messenger bots

How can Natural Language Processing improve my business?

Most businesses have online platforms or manual forms through which they receive customer feedback on their products and services. The feedback is in terms of texts and audio clips which are generally unstructured. Knowledge on what customers say about a particular brand helps a company provide great customer experience and respond to any challenges that may affect their clients. Natural Language Processing is used in applications such as chatbots that interact with clients directly. NLP is further used to analyze customer responses. Speech recognition can be used as an authorization code/password to give access to a company service since an audio print is unique to each client.

Summary of basic steps in Natural language processing for text analysis.

• Structure Extraction-identifying fields and blocks of content based on tagging.
• Identify and mark boundaries- this includes sentence, phrase and paragraph boundaries which act as breaks within which the analysis is conducted.
• Language identification- This helps to determine which linguistic algorithms and dictionaries to use.
• Sentence segmentation-breaking down the text to individual sentences. This assumes that each sentence is a separate idea.
• Tokenization- dividing up the characters of a sentence into tokens (Words, punctuation, identifiers, and numbers)
• Predicting parts of speech for each token- figuring out the role of each token in the sentence.
• Lemmatization- This involves figuring out the most basic form of each word in a sentence.
• Identifying stop words-these are words that one may consider filtering before performing any statistical analysis. They are words that appear way more frequently than other words.
• Dependency parsing- this involves looking at how the words in the sentence relate to each other.
• Finding noun phrases- Use dependencies from the above step to automatically group together words that are talking about the same thing.
• Named Entity Recognition- identifying and extracting names, places etc. in order to simplify downstream processing. The goal is to detect and label these nouns with the real world concepts they represent.
• Co-reference resolution-checking out for sentences that often refer to previous objects. To achieve the highest possible coverage, it’s important to identify these references and resolve them.

Some python libraries for Natural Language processing.

• SpaCy
• NLTK
• PyNLPI
• Stanford Core NLP python

References

How computers understand Human Language

https://medium.com/@ageitgey/natural-language-processing-is-fun-9a0bff37854e

Natural Language Processing (NLP) Techniques for Extracting Information

https://www.searchtechnologies.com/blog/natural-language-processing-techniques