Blogs

Sentiment or Sediment?

Governance of highly disposable social sentiment data can create a quality conundrum

Big Data Evangelist, IBM

You care deeply about how your customers regard your company, your products, and your quality of service. You may be listening to social media to track how your customers—collectively and individually—are voicing their feelings. But do you bother to save and scrutinize every last tweet, Facebook status update, and other social utterance from each of your customers? And if you are somehow storing and analyzing that data—which is highly unlikely—are you linking the relevant bits of stored sentiment data to each customer’s official record in your databases?

If you are, you may be the only organization on the face of the earth that makes the effort. Many organizations implement tight governance only on those official systems of record on which business operations critically depend, such as customers, finances, employees, products, and so forth. For those data domains, data management organizations that are optimally run have stewards with operational responsibility for data quality, master data management, and information lifecycle management.

However, for many big data sources that have emerged recently, such stewardship is neither standard practice nor should it be routine for many new subject-matter data domains. These new domains refer to mainly unstructured data that you may be processing in your Hadoop clusters, stream-computing environments, and other big data platforms, such as social, event, sensor, clickstream, geospatial, and so on.

The key difference from system-of-record data is that many of the new domains are disposable to varying degrees and are not regarded as a single version of the truth about some real-world entity. Instead, data scientists and machine learning algorithms typically distill the unstructured feeds for patterns and subsequently discard the acquired source data, which quickly become too voluminous to retain cost-effectively anyway. Consequently, you probably won’t need to apply much, if any, governance and security to many of the recent sources.

Blindly accepting sentiment as honesty is folly

Where social data is concerned, there are several reasons for going easy on data quality and governance. First of all, data quality requirements stem from the need for an officially sanctioned single version of the truth. But any individual social media message constituting the truth of how any specific customer or prospect feels about you is highly implausible. After all, people prevaricate, mislead, and exaggerate in every possible social context, and not surprisingly they convey the same equivocation in their tweets and other social media remarks. If you imagine that the social streams you’re filtering are rich founts of only honest sentiment, you’re unfortunately mistaken.

Second, social sentiment data rarely has the definitive, authoritative quality of an attribute—name, address, phone number—that you would include in or link to a customer record. In other words, few customers declare their feelings about brands and products in the form of tweets or Facebook updates that represent their semiofficial opinion on the topic. Even when people are bluntly voicing their opinions, the clarity of their statements is often hedged by the limitations of most natural human language. Every one of us, no matter how well educated, speaks in sentences that are full of ambiguity, vagueness, situational context, sarcasm, elliptical speech, and other linguistic complexities that may obscure the full truth of what we’re trying to say. Even highly powerful computational linguistic algorithms are challenged when wrestling these and other peculiarities down to crisp semantics.

Third, even if every tweet was the gospel truth about how a customer is feeling and all customers were amazingly articulate on all occasions, the quality of social sentiment usually emerges from the aggregate. In other words, the quality of social data lies in the usefulness of the correlations, trends, and other patterns you derive from it. Although individual data points can be of marginal value in isolation, they can be quite useful when pieced into a larger puzzle.

Consequently, there is little incremental business value from scrutinizing, retaining, and otherwise managing every single piece of social media data that you acquire. Typically, data scientists drill into it to distill key patterns, trends, and root causes, and you would probably purge most of it once it has served its core tactical purpose. This process generally takes a fair amount of mining, slicing, and dicing. Many social-listening tools, including the IBM® Cognos® Consumer Insight application, are geared to assessing and visualizing the trends, outliers, and other patterns in social sentiment. You don't need to retain every single thing that your customers put on social media to extract the core intelligence that you seek, as in the following questions:

  • Do they like us?
  • How intensely?
  • Is their positive sentiment improving over time?

In fact, doing so might be regarded as encroaching on privacy, so purging most of that data once you've gleaned the broader patterns is advised.

Fourth, even outright customer lies propagated through social media can be valuable intelligence if we vet and analyze each effectively. After all, it’s useful knowing whether people's words—"we love your product"—match their intentions—"we have absolutely no plans to ever buy your product”—as revealed through their eventual behavior—for example, buying your competitor's product instead.

If we stay hip to this quirk of human nature, we can apply the appropriate predictive weights to behavioral models that rely heavily on verbal evidence, such as tweets, logs of interactions with call-center agents, and responses to satisfaction surveys. I like to think of these weights as a truthiness metric, courtesy of Stephen Colbert.

What we can learn from social sentiment data of dubious quality is the situational contexts in which some customer segments are likely to be telling the truth about their deep intentions. We can also identify the channels in which they prefer to reveal those truths. This process helps determine which sources of customer sentiment data to prioritize and which to ignore in various application contexts.

Misinterpreting sentiment creates risks

Last but not least, apply only strong governance to data that has a material impact on how you engage with customers, remembering that social data rarely meets that criterion. Customer records contain the key that determines how you target pitches to them, how you bill them, where you ship their purchases, and so forth. For these purposes, the accuracy, currency, and completeness of customers’ names, addresses, billing information, and other profile data are far more important than what they tweeted about the salesclerk in your Poughkeepsie branch last Tuesday. If you screw up the customer records, the adverse consequences for all concerned are far worse than if you misconstrue their sentiment about your new product as slightly positive, when in fact it’s deeply negative.

However, if you greatly misinterpret an aggregated pattern of customer sentiment, the business risks can be considerable. Customers’ aggregate social data helps you compile a 720-degree portrait of the behavioral tendencies and predispositions of various population segments. This compilation is essential market research that helps gauge whether many high-stakes business initiatives are likely to succeed. For example, you don’t want to invest in an expensive promotional campaign if your target demographic isn’t likely to back up their half-hearted statement that your new product is “interesting” by whipping out their wallets at the point of sale.

The extent to which you can speak about the quality of social sentiment data all comes down to relevance. Sentiment data is good only if it is relevant to some business initiative, such as marketing campaign planning or brand monitoring. It is also useful only if it gives you an acceptable picture of how customers are feeling and how they might behave under various future scenarios. Relevance means having sufficient customer sentiment intelligence, in spite of underlying data quality issues, to support whatever business challenge confronts you.

If you have a thought or question about this topic, please share it in the comments.

[followbutton username='jameskobielus' count='false' lang='en' theme='light']
 
[followbutton username='IBMdatamag' count='false' lang='en' theme='light']