What`s new

Blog

Breakthrough in the Automatic Diagnosis of Motives

Apr 17, 2020 10:09:46 AM / by Simon Tschuertz

Automatic speech analysis is often criticized. It is not valid and so far nobody has proven that it actually works. Already in September 2019, we have reported in an article of Algorithmwatch (https://algorithmwatch.org/story/sprachanalyse-hr/) about working on a large validation, which marks the scientific breakthrough. With pride, we can now present the results.

This article shows the approach and our unique results.

Why are particularly motives important to understand?

Motives are important internal action instructors. They cause a person to think in a certain direction and to behave in a certain way. They express an aspiration: "I would like to be...." or "I would like to...". Generally, three basic motives are distinguished. They are the motives for power, status, and leadership, for performance and development and harmonious relationships with other people.

Unlike other personality models such as the Big 5, motives have a very high predictive power of human behaviour. Motives can be used, for example, to increase sales opportunities or predict professional success. A person with high-performance motives needs solution-oriented arguments for the purchase decision. A person with a high power motive on the other hand needs prestige-oriented arguments. Professional success can also be predicted. Dörr was able to show in 2016, for example, that managers have more satisfied employees under a certain motive. These motives were formulated by the social psychologist David McClelland and researched in psychology. Unfortunately, they have not yet been widely transferred into practical applications.

Why motives are not yet applied in full?

Up to now, motives have been captured with a procedure that can only be carried out and evaluated with a lot of effort. Texts that are written by people and provide information about the implicit motive structure of a person have to be examined in painstaking detail by two experienced psychologists. It goes without saying that this method is not scalable and cannot be used in communication processes.

Therefore, we have dealt as 100W to develop solutions that bring the power of motives in various practical applications to bear. Be it in the understanding of employees, customers, and interested parties. But also to get in better contact with these groups. This ensures successful cooperation and communication.

Validation studies are important to successfully predict motives. For us, these are an essential part of our corporate philosophy. To measure personality traits in high quality is something everyone can claim.

Implementation and results of the latest validation

For the prediction of motives we used the core of our software. The so-called Psychological AI, which we have been researching and developing since 2014. For this task we had the more than 66,000 records of Schönbrodt et al (2020) at our disposal. Each sentence in this dataset was accompanied by information that provides information about the motive characteristics. This label was awarded by psychologists trained for this task. As usual, we divided the data set into a training and a test data set. We repeated this several times by randomly selecting different training and test data sets. And each time we had our Psychological AI predict which motive would be expressed in a given sentence. To decide how well our Psychological AI correctly classified the motives, we collected the so-called Accuracy. This gives the percentage of all correct decisions to all decisions. We achieve an accuracy of 85%. This means that 85% of our decisions are consistent with those of people. The F-score is even better, 86%. This also takes into account the prediction. The prediction is the percentage of all wrong decisions to all decisions. But what does this mean exactly?

Importance of this scientific breakthrough

In order to better understand the meaning, we compare the achieved values with those of the two psychologists. On average, these reach a value between 75% and 85%. With 86 % of our Psychological AI we are thus above the values of the two most similar psychologists. If we now also take into account objectivity, i.e. the influence of measurement fluctuations, and cost-effectiveness, our approach is clearly superior to the standard procedure. Thus, for the first time, motifs can now be used in a broad application in two ways. On the one hand, to build understanding among customers, employees, and interested parties and, on the other hand, to improve the relationship based on this information. This creates completely new possibilities for positively shaping relationships between people.

Our recommendations

- How well do 100W predict our motives? - Our motive study

- The 100W of text analysis to measure implicit motives

 

Tags: Psychological AI, implizite Motive, Textanalyse, implicit motive, 100W, artificial intelligence, motive analysis, Speech interpretation

Written By Simon Tschuertz

Lists by Topic

see all