Whether You Have a Mental Illness or Not, You Can Confirm It by Talking to This Model

There are some subtle relationships between mental illness and language, but even experienced doctors are not able to fully grasp this connection. However, from a data perspective, machine learning algorithms are expected to be used in language analysis to find individuals who may develop mental illness from abnormal language features, thereby helping to build and prevent mental health.

Mental health is gradually threatening the lives of more and more people. Some professionals even assert that mental illness is the biggest threat in the 21st century.

Mental illness, also known as mental disorder or psychological illness, mainly refers to the disorder of brain function, which leads to problems in cognition, emotion, will and behavior. Common mental illnesses include schizophrenia, depression, autism, dementia, personality disorder, etc.

Mental illness seriously affects personal life and also becomes a destabilizing factor for society.

Traditional diagnosis of mental illness: relying on experts

The diagnosis of mental illness is also very important. If it can be diagnosed in time, it can help patients cope with and receive treatment better. For those who want to use "mental illness" as an excuse, evidence can make them accept justice.

However, conventional diagnostic methods are rather complicated.A thorough psychiatric examination, physical and neurologic examination, brain imaging, and neuropsychological evaluationAfterwards, the doctor gave a preliminary diagnosis.Combined with complete medical history, especially personal life history, medical history, and related social and psychological factors, are analyzed and summarized to give the final judgment result.

The traditional approach also has its drawbacks, such as the lack of directly observable biological objective indicators and reliance on clinical observation of symptoms and the personal experience of professional doctors.

Some studies have shown that subtle features of language can be used to predict whether a person is at risk of mental illness. However, these features are usually not observable by humans and require some technical means, and machine learning has become the best choice.

"Even for experienced doctors, trying to hear these nuances in a conversation is very difficult, just like trying to see tiny bacteria with your eyes, which is almost impossible."

Neguine Rezai, a researcher at the Department of Neurology at Harvard Medical School, said,But computational methods such as machine learning are feasible to reveal subtleties hidden in language."These methods are like adding a microscope to accurately detect those signs," she added.

Nuture, a sub-magazine that specializes in reporting on mental illness NPJ Schizophrenia Magazinesuperior,Emory University and Harvard UniversityResearchers published a paper titledA machine learning approach to predicting psychosis using semantic density and latent content analysis》They used machine learning technology to uncover the hidden connection between language patterns and mental illness, and were able to accurately predict the early symptoms of mental illness.

Article link address:

https://www.nature.com/articles/s41537-019-0077-9

New diagnostic method: Finding the secrets of language from text analysis

To extract the basis for judgments from language, they used two linguistic variables:Semantic DensityandUse of sound-related vocabulary.

Semantic density is a measure of "lack of content" or ambiguity. We use the mathematical method of vector unpacking to obtain a linguistic marker of semantic density: breaking down the meaning of a sentence into its core ideas.

In order for the model to establish a judgment benchmark, Reddit On the website 30,000posts, capture the conversation content, use Word2Vec The program analyzes the words in the conversation. It processes and analyzes the words so that those with similar meanings are close together in the "semantic space", while those with very different meanings are far apart.

Next, speech samples from 40 patients in the North American Prodrome Longitudinal Study (NAPLS) at Emory University were input and followed up for two years. To train the model, information was collected from an additional 30 participants in the second phase of the study.

Finally, the team compared the conversational information from NAPLS with baseline data using machine learning analysis, and then compared it with follow-up data, including those who eventually developed mental illness, to find the link between language and the risk of illness.

The results showed that among the analyzed population, those who eventually developed mental illness had some common characteristics. In their conversations,The use of sound-related words (such as modal particles) is higher than the normal standard, and words with similar meanings are used frequently..

In their approach, the content of language was used to predict mental illness, andThe accuracy rate reached 93% .

Professor Elaine Walker, one of the researchers, said that if at-risk individuals could be identified earlier and preventive interventions could be taken, significant improvements could be made in the treatment of mental illness.

More than just mental illness, the mysteries of the brain lie ahead

Although the experiment achieved high accuracy, due to the limited number of samples, the experiment is still a success at the research level. The researchers also said that they are moving towards the goal of perfection and productization, and plan to use more data to test and improve this technology in the future.

In any case, the application of new technologies has once again confirmed that we can hear the "hidden meaning" of natural language from data.

In a recent Nuture magazine cover story, an amazing achievement was featured.Columbia University researchers spent eight years mapping all the neural networks and connections of the nematode Caenorhabditis elegans..

One-third of the cells in this nematode are brain cells, so mapping the neural connections between nematode brain cells is the first time that humans have deciphered the detailed workings of the brain.

Research on mental illness can also help explore the brain. Although the development of AI that imitates neural connections has now declined, exploring the mysteries of the brain has been a dream that has existed since the birth of artificial intelligence.

These explorations into mental illness are just small discoveries, butIn addition to revealing information about mental illness, it also helps us understand how the brain works, such as deducing how the brain puts various thoughts together.It is undeniable that these discoveries, piled together, will eventually bring about a new chapter.

The use of more powerful algorithms and technologies has brought new breakthroughs in diagnosis and treatment, but technology has not yet become so magical that it can take over everything. We just hope that the intervention of more methods will enable more and more people to have a healthier life.