Sentiment analysis has gone mainstream. Based on natural language processing (NLP) technology, this tool extracts sentiment by mining a large volume of publicly available data, including social media. At its most basic level, a sentiment (or opinion) analysis objectively ascribes a polarity – from positive to negative – to any text source.
The applications are many. Politicians can track their standing in the court of public opinion in real-time. When monitoring communications, government agencies can build on sentiment analysis capabilities to develop ‘anticipatory intelligence’ skills to detect anomalies.
Firms can track employee sentiment by analyzing internal or external communications (e.g., Glassdoor in ‘Open Kitchen,’ 2019.) They can also assess the evolution of the sentiment attached to their reputation or the opinion linked to specific products or product categories.
The same approach applies to corporate finance, including M&A or IPOs. Sentiment analysis can be integrated into due diligence processes.
In the capital markets, investors can assess management sentiment by reviewing earnings call transcripts and the evolution of such sentiment over time – an attempt to read between the lines about the sentiment related to issues such as China, pricing, or sustainability.
Similarly, it would be theoretically possible to run FOMC statements through powerful NLP algorithms to gauge the underlying sentiment of the Fed – dovish or hawkish – with the outcome sending immediate market orders according to pre-defined trading rules.
In all these cases, sentiment analysis helps obtain objective performance feedback or gather a unique perspective on specific trends in real-time.
The pitfalls include the difficulty of capturing irony and sarcasm, negations, and word ambiguity. Anyone learning or operating in a foreign language will appreciate these linguistic challenges. More generally, NLP tools struggle to capture the context that gives words their meaning, as established in ‘Beyond Words’ last month. Research assessing the reliability of sentiment analysis tools, including ‘The Validity of Sentiment Analysis’ (2021), suggests caution.
Limitations notwithstanding, sentiment analyses carry value when used appropriately. For now, they require human intervention as part of a hybrid machine-human discovery process. The balance will shift as sentiment analysis algorithms become increasingly sophisticated thanks to rapidly advancing machine learning capabilities (check the GTP-3, for example.)
Sentiment analysis is set to become a tool of high relevance to many corporate functions, including human resources, public relations, investor relations, and marketing. Understanding and learning about ways to use it will bring many competitive advantages.
PS - This note has received a perfectly neutral sentiment score when run here. It has achieved the required balance