Ocular Sentiment Analysis Library (OSAL)
Before the internet, there were two primary mediums that informed Americans. These two mediums were radio stations and newspapers. Newspapers were usually received in the morning, while radio shows were usually played in the evening. The primary difference between the two is that newspapers allow writers to include a lot more information. I cannot imagine how simple it would if there were only two news sources in 2021. Nowadays, there are now hundreds, if not thousands of places that we gather information every day. Even though the simplicity is enticing, there are also potential downsides to having limited information sources.
The Los Angeles Times - May 5, 1941.
I think the primary drawback is that it becomes easy for an overpowering institution to take over and manipulate the mass media. An “overpowering institution” could mean a lot of things. It could be a corporate conglomerate, a wealthy individual, or even a government. The fewer information streams, the easier it becomes for one group to take control of how the masses perceive the world.
Let’s go all the way back to September of 1939- when under the leadership of Hitler, Nazi Germany invaded Poland. Using Blitzkrieg tactics, the country was quickly overtaken. Britain and France declared war on Germany shortly after but took very little action. Over the next two years, Germany took over Denmark, Norway, Belgium, The Netherlands, and France. This brings us to December 1941, when the Japanese executed a surprise attack on Pearl Harbor.
The Los Angeles Times - December 16, 1941. A few weeks after the Japanese attack on Pearl Harbor.
It was this tragic event that pulled the US into the war. The attack on Pearl Harbor took the United States by surprise. Roosevelt was forced to respond quickly. In the paper the next day were phrases like: “Japanese Alien Roundup Starts”, “FBI Seizes Aliens”, “Los Angeles is Stunned”, “They started it, we’ll finish it”, and even calls to action. From the very start, there was the creation of the “us and them” mentality. The idea that the enemy was inhuman, and we needed to get rid of them.
Six months after the attack on Pearl Harbor, the United States formed the Office of Wartime Information (OWI). The OWI was a government agency that was formed to control both domestic and foreign information. The OWI produced films, created radio shows, and wrote newspapers that were distributed all around the world with a common goal to spread US propaganda.
We know that the OWI was in charge of inflows and outflows of information in the US, but we don’t really know the extent to which that happened. It is very possible that we have not yet discovered the full power of the OWI’s voice. The agency operated for a total of only 4 years.
The Office of Wartime Information in 1942.
There are a few ways that one could test the extent of the OWI’s power:
Analyze newspapers and see if there are any irregularities from the start to the end of the OWI’s existence.
Same thing as number 1, but with NLP and radio shows.
You could ask every leader in the OWI what their goals were and try and connect the dots.
Number 3 is pretty much impossible at this point, which leaves numbers 1 and 2. I could *probably *do number 2, but it would take a lot longer, and I’m not convinced that the data would be very useful. I assume most of the radio shows that the OWI created stopped after the agency was disbanded, which means it would be hard to collect data before and after the war to scan for irregularities.
Hence… newspapers! If we perform sentiment analysis of newspapers from 1939 – 1945, we may be able to notice some differences/irregularities in the general sentiment on the front page of the papers. The reason I chose to only scan the front page, is because that is where the general sentiment for the entire paper is stored. That’s also where the most important information is.
To accomplish this, I created the “Ocular Sentiment Analysis Library” (OSAL), an open-source library that combines different tools offered by Google and Microsoft. The library allows the user to batch-input any image, convert it to text, and perform sentiment analysis to get a result from 0 to 1.
OSAL can be used for more than just newspapers, it can actually determine the sentiment of any input image (assuming the image has text). For example, if you wanted to determine the sentiment of a page in a book (or an entire book), you could run every page through the program and collect your data.
As of the time of this writing (March 3, 2021), I have analyzed 140 newspapers. Keep in mind that this project would be MUCH more complete if I had more data. For this project to be at its full potential, I would have to analyze every single newspaper from every publication in the United States. The primary limitation of this project is that I am writing about the sentiment analysis of 140 newspapers from the Los Angeles Times.
Sentiment Graph with labeled events
Usually, what sticks out the most when collecting data is outliers. This is true for this project as well. It’s fascinating to look at both the negative and positive outliers. For example, the most positive newspaper front page from all the data I collected was on May 10th, 1947.
Most Positive Front Page
The fascinating thing about this front page is that nothing really sticks out as *good news. However, *I noticed that if you compare it to some of the other papers, this ended up being a rather good paper. Most of the papers had wartime information that included topics such as bombings, killing, and other generally violent language. I did a bit of further research into this particular week in history and found that it was also the week that many German officers that ran nazi camps were indicted for war crimes. This makes me wonder about the relationship between the newspaper sentiment and the general sentiment for that week. This is yet another limitation of only having 140 datapoints. It would be rather fascinating if you plotted each datapoint as the average sentiment over a 7 day period. To do that for the LA Times for 10 years, you would need to compile, scan, and record the sentiment data from 3,650 newspapers. An implementation such as this would likely decrease the number of outliers in the data and increase the accuracy of the project overall.
Additionally, here is the most negative day in the LA Times. The difference between this and the aforementioned front page is that there is a lot of violent language in this paper. (“Raid”, “Death Battle Rages”, “Starts Blaze”, “US Army Bomber Crashes”, “…Hotel Bombed”, and much more in the small text.)
Most Negative Front Page
If we take a look at the sentiment chart, we can see that average sentiment is decreasing up until around mid-1942. It is incredibly important to point out that the OWI was created in June of 1942. Seemingly immediately after their creation, there is a general increase in sentiment. Now, it’s actually difficult to analyze this because this increase in sentiment could be a result of the OWI’s US-based wartime propaganda, or it could be a result of the US beginning to win the war. If we wanted to conclusively say that the OWI had significant power in manipulating domestic mass media, we would have to do the same exact sentiment analysis on another war and see if the trends are similar.
Sentiment Chart From LA Times Front Page
My goal with this research is to pioneer a new type of quantitative historical inquiry. As researchers, we may be able to generate answers to questions we didn’t even know existed. Many of us believe in the notion that it is important to look back at the lessons of history in order to better understand the world today. With this research, I hope to demonstrate that we can also apply modern technologies to better understand the past.