Arabic Sentiment Analysis Real-World Use Cases

0

featured image

Take a look at the top 3 real-world examples of how AI-based Arabic natural language processing (NLP) empowers businesses. Machine translations can have adverse effects when used for deriving sentiments owing to the intricacies of the Arabic language such as inflectional writing, left-to-right directions, semantic ambiguity, lack of traditional vowels, and other such issues. The examples showcased in this article prove how data gathered from varied sources can be richly harnessed for consumer insights when an Arabic sentiment analysis model is powered by native Arabic NLP.

Importance of Arabic NLP in business

Arabic is a very complex language. Having a sentiment analysis solution that relies on an intelligent Arabic NLP model rather than the traditional system of using translations, is of critical importance. Unlike most languages, Arabic has three major linguistic issues – syntactic ambiguity, diacritical marks that denote vowels that are not present in online texts, and a lack of punctuation that makes long paragraphs difficult to decipher by machine learning algorithms. When an Arabic sentiment analysis model translates words to another language, say English, it loses the nuance of the text, thus leading to incorrect aggregate sentiment analysis. This can translate to businesses adopting wrong strategies based on incorrect insights, which in turn can be detrimental to a company’s return on investment and overall business planning.

Figure 1: How native Arabic NLP platforms help machine models overcome Arabic complexities to analyze sentiment accurately.

How Arabic NLP improves sentiment analysis accuracy

Use cases of Arabic Sentiment analysis

These use cases showcase how having a native Arabic NLP platform is crucial for an Arabic semantic search solution, coupled with aspect-based sentiment analysis. They also highlight how imperative it is to have a platform that can analyze video content as efficiently and effectively as text data.

1. Nahdi Medical:

Nahdi Medical is a Jeddah-based chain that manages and operates a nationwide network in Saudi Arabia. It covers 145 cities and villages, providing advanced medical care in areas such as radiology, oncology, cardiology, and pediatrics. Repustate developed an Arabic sentiment analysis solution for Nahdi that helped them extrapolate what their customers and employees were saying. The platform provided them with a speed and accuracy that other Arabic sentiment mining solutions were not able to do because of their lack of native Arabic NLP capabilities.

Repustate’s sentiment analysis solution was made dedicatedly for the Arabic-speaking populace. The platform understands major Arabic dialects such as Gulf Peninsular, Egyptian, and Levantine Arabic. It analyzes data from patient vlogs, social media vlogs, comments, reviews, and surveys through its video and audio content analysis capabilities, thus giving Nahdi a comprehensive view of the consumer’s experience.

2. Appenza Studio & the Ministry of Education, Egypt:

Appenza Studio is a specialized mobile app and web development company based in Cairo, Egypt. They are a partner to the Egyptian Ministry of Education for developing mobile learning apps for students and their parents so they can access the Ministry’s knowledge bank of academic material. While the academic content was rich, it was nearly impossible to discover since none of it was structured by any theme or logic. Further, the content was a mix of Arabic and English, which made organizing more difficult. At the same time, it was imperative that Appenza understood the feedback from students and parents to measure the performance of the app on an ongoing basis. They needed accurate and automated Arabic NLP based sentiment analysis that was especially made for Egyptian Arabic for insights from social media data.

Repustate developed a customized semantic video content search & sentiment analysis solution for Appenza and the Ministry of Education using Arabic NLP. The platform helped bridge the gap between the Ministry and the Egyptian population in the efficacy of education outreach. The content is now semantically organized by language, subject, themes and other aspects. Appenza is able to gather insights from the public regarding the efficiency of the app. Through Repustate’s Arabic sentiment analysis model, the Ministry now caters to both its Arabic and English speaking population in a more holistic way. They can also now nurture connections between academic institutions by giving them equal attention and better levels of engagement.

HealthLinks is a specialized healthcare consultancy company based in Jeddah, Kingdom of Saudi Arabia (KSA). They partner strategically with the Ministry of Health, KSA, healthcare leaders, and involved stakeholders to improve the overall quality of healthcare in the Gulf region. They locate invisible dysfunctions and gaps in care services, and map each and every phase of the patient’s journey within a hospital through data-backed knowledge and provide the information to the relevant leaders.

Repustate provided HealthLinks with a robust cloud-based Arabic sentiment analysis solution that could analyze hundreds of comments in seconds while providing sentiment scores for each of the themes specified in the Saudi Complaints Taxonomy. The solution can analyse unstructured Patient Voice data gathered by HealthLinks from more than 12 million surveys that they conduct on an annual basis. Through the native Arabic NLP model, the Ministry of Health can form well-rounded connections between various data streams connecting patient experience and operational data, and design better targeted improvement strategies.

How we do Arabic sentiment analysis at Repustate

A massive corpus of varied Arabic data is collected to train the Arabic sentiment analysis model. A part of the resulting data is tested and then compared to an existing dataset. The Arabic NLP model is trained again until it gives the highest accuracy scores. The steps are as follow:

  • Step 1: Part-of-speech tagger

Each Arabic word is classified at a grammatical level to identify conjunctions, subordinate clauses, prepositional, and noun phrases. This helps the model understand the text’s true meaning.

  • Step 2: Lemmatization

In this step, the rules of conjugating nouns and verbs based on gender, tense, etc. are applied in the model. This helps the tool determine the root of a word. For example, “reading” and “reader”, are based on the root word “read”.

  • Step 3: Prior polarity

We determine the positive and negative context of a word and calculate the intensity of the polarity. For example, excellent (+1), good (+0.5), average(0), and poor(-0.5). This is what helps the Arabic sentiment analysis solution to provide scores.

  • Step 4: Grammatical constructs

We determine nuanced grammatical constructs like negations and amplifiers so the model can understand sentiment scores.

  • Step 5: Sentiment scores

When all these steps using Arabic NLP come together, the sentiment scores are fed into machine learning models. Now the model can assign scores related to an aspect or entity when it reads a text.

Read more: How YouTubers Can Use Social Media Listening

 

 

LEAVE A REPLY

Please enter your comment!
Please enter your name here