Ad Entity Extraction using NLP
Advertisements are at core of a company’s function, they put the brand’s ideas and proposals out to the public and other companies to enrich the engagement of it’s products. Adverts have been the most titular form of product selling for the past 150 years, it is even considered to be an art form of its own. The importance of an advert can sometimes make or break a company, there’s a need for deeper analytical tools to understand what’s effective in an advert.
United Kingdom
Client
Event Management Industry
CLIENT
Our client is an event management firm based in the United Kingdom, they needed an advert analysis tool to properly streamline their adverts and advance them in more clear directions. This tool, developed by us, provides a more clearer understanding of print adverts by extracting the vital information related to the brand name, brand image, and also the marketing taglines. This information can be used for proper indexing and market analysis. It can also provide vital information about the brand in a single click.
CHALLENGE
There are certain challenges pertaining to various modules in this tool, they are listed below:
-
Extraction and Ad Resolution Challenges - Image and text extraction can be hindered by various design issues such as improper background colors, misaligned and incoherent typography, and etc. The ad’s scan resolution can also affect the process of Optical Character Recognition.
-
Phrasal Ambiguity - The multiple slogans and flavor text inside an ad can severely affect the phrasal extraction process regarding natural language processing.
-
Feature Extraction - The amount of image datasets, the quality of the images, and various other issues can cause deviations in the feature extraction process of the model.
SOLUTION
Solution summary:
-
The adverts are compiled and organized in an ordered manner for easy analysis.
-
An image processing ML model is created to identify the textual and non-textual components of the ads.
-
A trained NLP model is used to extract the needed information for the client.
-
Web Sources are then used to corroborate the information for concise and accurate information gathering.
With the challenges in mind, we arrived at the following steps for a solution :
-
The collected adverts are organized and an image processing machine learning model is created to extract and demarcate the text and non-text entities in the print or image advert.
-
Natural Language Processing is then used to analyse the extract the needed information pertaining to disclaimers, location, time, and etc. The rules and regulations created by the client gives us the conditions to analyze the data.
-
Feature/Detail Extraction is used to derive specific details from the advert as needed.
-
External Web sources are used to make a quick check-up regarding the veracity of the information presented in the advert.
RESULT
This model gives our client clear and concise information needed for their event management commitments, they can organize any event at ease by understanding what the organization wants by analyzing their adverts. The image and content extraction tools present in this tool increases the versatility of customer relations. The quality of data inferred from this tool is highly valuable for their strategic marketing operations as well.