| copyright | lastupdated | subcollection | ||
|---|---|---|---|---|
|
2019-03-07 |
personality-insights |
{:shortdesc: .shortdesc} {:new_window: target="_blank"} {:tip: .tip} {:important: .important} {:note: .note} {:deprecated: .deprecated} {:pre: .pre} {:codeblock: .codeblock} {:screen: .screen} {:javascript: .ph data-hd-programlang='javascript'} {:java: .ph data-hd-programlang='java'} {:python: .ph data-hd-programlang='python'} {:swift: .ph data-hd-programlang='swift'}
{: #guidance}
Users are applying the {{site.data.keyword.personalityinsightsshort}} service to an increasing number of use cases. They apply the service to offer personalized product recommendations to customers via in-store kiosks. They explore and analyze differences in the personalities of US presidents as inferred from their State of the Union Addresses. And they integrate the service with another product in the {{site.data.keyword.IBM_notm}} {{site.data.keyword.watson}} portfolio, {{site.data.keyword.watson}} Explorer, to demonstrate how an investment advisor can offer suitable options based on investors' personality portraits. (For more information, see Use cases.) {: shortdesc}
Throughout the development of these and other prospective use cases, {{site.data.keyword.IBM_notm}} has talked with banks, healthcare providers, customer experience management companies, and federal agencies. These conversations often generate questions about applicable usage scenarios.
- What type of text is best for inferring an individual's personality? Does the text have to be reflective in nature? If so, how reflective and how does one measure reflection? See What type of text is optimal for inferring personality?
- How does one interpret text that is written by one person about another person? Can an author's personality be inferred from the author's fictional works? Can text that is written by multiple individuals be combined to obtain a personality portrait? See Interpreting results from different types of text.
- Can personality portraits be inferred from text that is generated by transcription or translation services? See Inferring personality from generated text.
- Can personality portraits be applied to applications such as matching individuals, monitoring and predicting mental health, and monitoring radical and rogue elements via social media? See Using personality portraits for specific applications.
- Does an individual's personality evolve as they age? See Does a person's personality change over time?
{: #optimal}
What type of text is best suited for inferring personality? Does the text need to be reflective, self-reflective, formal, or informal? How does one measure the words that are used in daily life? Are such words reflective in nature? The answers to these questions can help you select the most appropriate input to apply the service most effectively.
{{site.data.keyword.IBM_notm}}'s work on the {{site.data.keyword.personalityinsightsshort}} service is based on the fundamental premise that the words one uses in daily life reveal one's personality (Pennebaker and others, 2001, and Pennebaker and others, 2007). The service can analyze words that are written by individuals about themselves or about any topic. For the service to generate an accurate personality portrait, the individuals must choose and write the words.
{{site.data.keyword.IBM_notm}} believes that the text that is used to infer an individual's personality ideally needs to be reflective. Reflective writing exposes the author's personal experiences and responses. It requires that the author put a certain amount of thought into the words that are chosen. For instance, the text can include the writer's opinions, attitudes, sentiments, and observations about someone or something. It can also express the writer's likes and dislikes. But it must reflect the writer's selection of words.
{{site.data.keyword.IBM_notm}} did not find explicit references in the psycholinguistics literature about the need for reflection in the text that is used to infer personality. However, {{site.data.keyword.IBM_notm}} did observe that some studies use words from text that was collected in a lab setting in which people were asked to write short essays on specific topics. This type of writing implicitly demands a certain amount of reflection. Other studies used text from natural samples such as blogs on various topics, tweets, email, or even communication that is extracted from the game World of Warcraft. All such studies assume that the words that are used in daily life reveal personality because they tend to reflect the writer's thoughts.
Other studies show that emotional writing, control writing, blogs, and speech transcripts are well suited for inferring personality. Scientific articles, conversely, are only marginally suited for inferring personality. Additionally, the literature indicates that arriving at an accurate measure of personality is further complicated by text that
- Lacks authenticity, such as sarcasm, irony, intense editing, or multiple authors
- Includes misspellings, technical words, alternate meanings for words, negation, and phrases rather than single words
- Includes inappropriate comparison groups, such as professional versus personal and technical versus emotional writing
{: #interpreting}
The type of text to be analyzed can have a significant effect on the quality of the results of the {{site.data.keyword.personalityinsightsshort}} service. The following sections discuss how to interpret results that are obtained from different sources:
- Interpreting results from text about others
- Interpreting results from fictional works
- Interpreting results from text by multiple individuals
{: #writingaboutothers}
To reliably infer an individual's personality, does the text have to be written by the individual or can it be written about the individual by someone else? When one person writes about another person, whose personality is inferred from the text?
{{site.data.keyword.IBM_notm}}'s intuition is that writing always reflects the author's personality, regardless of the subject matter. For example, if individual A writes text about individual B, an analysis of the text infers the personality of individual A. Although individual A is writing about individual B, it is individual A who is choosing the words to express things about individual B. However, {{site.data.keyword.IBM_notm}} has not explicitly tested this scenario.
{: #fictionalworks}
{{site.data.keyword.IBM_notm}} does not recommend inferring an author's personality from their fictional work. When they write fiction, authors purposefully portray each character differently, they construct dialog to reflect their characters' personalities, and they predefine each character's personality in their minds.
Using text that is meant to reflect the personality of a personified fictional character to infer the personality of the creator of that character is questionable. While each author has a unique style, the characters are still purposefully preconfigured. However, an author's personality can be inferred from the author's non-fiction essays, interview transcripts, or other pieces of work such as introductions, prologues, acknowledgements, or dedications.
{: #multipleindividuals}
A common use case of the {{site.data.keyword.personalityinsightsshort}} service is to analyze the followers of a brand or company. Followers are defined as people who follow a company on Twitter or on a public Facebook page. In this scenario, the objective is to derive the overall personality of a company's followers. A preferred method is to collect sufficient text that is written by a group's members to calculate the personality of each individual member. The personalities are then clustered into distinct groups. These groups represent personae with distinctive personality characteristics who follow the company.
If insufficient text is available from a group's members, text that is written by many members can be combined and analyzed to produce an average or composite personality portrait for the group. This approach can yield an indication of the group's dominant qualities, but the accuracy of this method decreases with the diversity of the population. Exercise caution when you interpret portraits that are obtained from aggregating the text of multiple individuals. {{site.data.keyword.IBM_notm}}'s work in this area is still in its initial stages; {{site.data.keyword.IBM_notm}} will report its findings as it uncovers convincing results.
{: #inferring}
Analyzing text that is generated by transcription or translation services can impact the reliability of the results of the {{site.data.keyword.personalityinsightsshort}} service. The following sections discuss the effects of inferring personality from generated text:
{: #transcription}
Speech transcription engines, such as the {{site.data.keyword.IBM_notm}} {{site.data.keyword.speechtotextshort}} service, generate written text from spoken words. Different transcription engines have different accuracy ranges. Customers who transcribe speech to text for use as input to the {{site.data.keyword.personalityinsightsshort}} service need to be aware that results can vary widely depending on the performance of the engine. Specifically, {{site.data.keyword.IBM_notm}} advises clients and business partners to determine the quality of the transcription against two types of errors:
- Drop out: A spoken word is omitted from the transcription.
- Substitution: A spoken word is transcribed incorrectly.
Substitution can be a more serious issue because it can introduce words that were not spoken but that match words that are used to determine personality. Before you use transcribed text, consider manually correcting the text of a test corpus and counting the errors that you find. Then, compare the results of the automatically generated text with the manually corrected version to determine the variance in results due to transcription errors.
{: #translation}
Language translation services translate text that is written in one language to another language. As with speech transcription, the question arises whether translated text can be used as input to the {{site.data.keyword.personalityinsightsshort}} service. {{site.data.keyword.IBM_notm}} does not recommend that you use text that is obtained from translation services as input to the {{site.data.keyword.personalityinsightsshort}} service. Depending on the translation service, the results of both the translation and the personality inference can vary widely. Words, their senses, and cultural sensitivities tend to get lost in translation, yielding invalid results.
{{site.data.keyword.IBM_notm}} recommends that you use as input only text that is written in languages in which the {{site.data.keyword.personalityinsightsshort}} service is enabled. Language-enabled versions of the service
- Parse the input text in its native language.
- Use native language dictionaries to identify personality characteristics.
- Use models that are calibrated for the native language to produce statistical results.
If you must analyze translated text, {{site.data.keyword.IBM_notm}} recommends that you first manually translate some sample text by using the services of a human language expert. You can then compare the results that the {{site.data.keyword.personalityinsightsshort}} service obtains from both manually and automatically translated text to understand the variance in results.
{{site.data.keyword.IBM_notm}} continues to add more languages as business demand increases. {{site.data.keyword.IBM_notm}} understands that the {{site.data.keyword.personalityinsightsshort}} service might not support your native language fast enough for your purposes. {{site.data.keyword.IBM_notm}} is conducting tests to compare results from native language enablement with results from language translation services and will report its findings as they become available.
{: #applying}
The {{site.data.keyword.personalityinsightsshort}} service can be applied to innumerable use cases. The following sections describe the use of personality portraits for a few specific purposes:
- Matching individuals
- Monitoring and predicting mental health
- Monitoring radical and rogue elements via social media
{: #matching}
Making a good match between people can improve interaction and outcomes in relationships. This notion also applies to team building within a company and to interaction with customers across a wide range of industries. In a study of doctor-patient matching, researchers found that patients prefer doctors who are similar to themselves. Effectively matching doctors and patients builds confidence and encourages communication, which can improve compliance and result in more successful treatment (Godager, 2012).
Personality also influences interaction preferences between professionals and customers. For example, patients with a high degree of conscientiousness and openness prefer being actively involved in deciding their course of treatment. Patients with high levels of agreeableness or neuroticism prefer that doctors take the lead in making important health decisions (Flynn and Smith, 2007).
{: #mentalhealth}
{{site.data.keyword.IBM_notm}} believes that disease diagnosis is best performed by a trained medical professional. It might be possible to detect some signs of individuals' mental status from their word usage. But {{site.data.keyword.IBM_notm}} has not conducted ground truth studies or explored the possibility of establishing a scientific basis for such work.
Clients and business partners who are interested in using the {{site.data.keyword.personalityinsightsshort}} service for mental health diagnosis are welcome to design and conduct ground truth experiments for such use cases. Active research in this area includes work that relates personality to health outcomes (Israel and others, 2014). It also includes work that aims to predict postpartum and other forms of depression from social media (De Choudhury and others, 2013a, and De Choudhury and others, 2013b).
{: #monitoring}
Government agencies across the world are constantly searching for ways to detect early signals of the radicalization of individuals or groups of individuals via social media channels. Inferring personality from individuals' writing is a traditional application of the {{site.data.keyword.personalityinsightsshort}} service. So it comes as no surprise that using the service to infer radical and rogue elements via social media is a viable use case. Reliable inferences require not only personality characteristics but a host of other attributes such as gender, age, and geography. {{site.data.keyword.IBM_notm}} has no studies to validate or invalidate this use case. Clients and business partners are welcome to conduct studies to explore this use case based on their specific goals and requirements.
{: #evolve}
Does a person's personality evolve over time? If so, how often should the {{site.data.keyword.personalityinsightsshort}} service be used to infer a person's personality? Different studies report evidence for and against the theory that a person's personality stabilizes at adulthood. In his seminal 1890 work about measuring the stability of personality, The Principles of Psychology, Harvard psychologist William James noted, "In most of us, by the age of thirty, the character has set like plaster, and will never soften again." But more recent studies report that personality does evolve over time:
- Helson and others (2002) report that "the idea that personality change is most pronounced before age 30 and then reaches a plateau received no support." The authors conducted a longitudinal study on two cohorts over a period of 40 years. They note that period of life and social climate are significant factors in personality change over the adult years. They comment that "Scores on Dominance and Independence peaked in the middle age of both cohorts, and scores on Responsibility were lowest during peak years of the culture of individualism."
- Scollon and Diener (2006) set out to examine the individual differences in change in extraversion, neuroticism, and work and relationship satisfaction over time. The authors note that increased work and relationship satisfaction was associated with decreases in neuroticism and increases in extraversion over time.
- Roberts and Mroczek (2008) comment that evidence exists for "mean-level change in personality traits, as well as for individual differences in change across the life span." The authors note that people show increased self-confidence, warmth, self-control, and emotional stability with age. These changes predominate in young adulthood (ages 20 to 40). Moreover, mean-level change in personality characteristics occurs in middle and old age, showing that personality characteristics can change at any age.
Based on these studies, {{site.data.keyword.IBM_notm}} recommends that users of the {{site.data.keyword.personalityinsightsshort}} service always work with the latest data and with as much available data as possible. {{site.data.keyword.IBM_notm}} further recommends that users refresh personality portraits at regular intervals to capture individuals' evolving personalities. While {{site.data.keyword.IBM_notm}} tends to believe that personality evolves within certain bounds, it has conducted no studies to examine the upper and lower borders of this evolution.
One might ask how regularly to refresh the personality portraits of individuals. {{site.data.keyword.IBM_notm}}'s guidance is to look for the availability of new data and text from an individual. If an individual authored a substantial amount of new text since their personality portrait was obtained, it might be worth refreshing the portrait. This approach captures the evolving nature of personality, if one chooses to accept that personality evolves. It also gives the {{site.data.keyword.personalityinsightsshort}} service the benefit of working with more words, which in turn can increase the precision of the service's results.