Can we use tweet sentiment analysis to predict the NFL draft?

Since the NFL draft is tomorrow, I thought it would be a little fun to see if Twitter sentiment analysis could be used to predict where the top prospects will be selected. Twitter sentiment analysis is when computer programs look at each tweet mentioning certain keywords (for example, “browns Allen”) and provides each one with a sentiment score (low sentiment scores being negative tweets and high sentiment scores being positive tweets). The higher the sentiment score, the more positive a tweet is. A tweet with a low sentiment score would be something like this: “Swear if the Browns take Sam darnold tomorrow I’ll be so shocked. Would be worst mistake ever”. On the other hand, a tweet with a high sentiment score would be similar to this: “I really like Darnold — level headed -smart — very good player. Would be an excellent top pick.”

Note that I will only be analyzing the top fifteen picks of the NFL draft. Around this point, the NFL draft has less predictability. I won’t be predicting any trades in this mock draft.

In order to determine which player will be selected at each spot, I am using some of the prospects that have been predicted to be chosen by various teams in the past rather than analyzing every single prospect for every single team. The prospect with the highest average sentiment score for a team will be the player selected. A prospect’s average sentiment score will change as we evaluate him with the team selecting. For example, Saquon Barkley’s sentiment score is higher with the Giants than it is with the Cleveland Browns since tweets mentioning the Giants and Barkley are more likely to be positive than tweets mentioning the Browns and Barkley.

I will evaluate my mock draft after Thursday based on the following point system. If a player is chosen by the team I predicted that would choose him as well as at the spot where I predicted he would be picked, two points get awarded. For example, if the Browns select Josh Allen with their first pick, then that pick will receive two points. If a player is chosen by the team that I predicted would choose him but not at the pick where I had him being picked, then one point is awarded. In addition, if a player is chosen at the predicted pick but not by the correct team, then one point is awarded. Therefore, if the Browns select Josh Allen at 4 or the Giants trade up to draft Josh Allen with the first pick in the draft, then that pick will receive one point. If a team drafts the predicted position but not the correct player with their pick, or if a predicted position is selected at a pick, then half a point is awarded. For example, if the Browns select Sam Darnold with their first pick or if the New York Giants trade up to select Baker Mayfield with the first pick, half a point will be awarded. If none of these conditions are fulfilled, then no points will be awarded.

Mock Draft Generated from Tweet Sentiment Analysis

1. Cleveland Browns: Josh Allen, QB (Wyoming)

I’m not exactly surprised by this pick. It’s no surprise that the Browns have a huge crush on Josh Allen. Despite being a raw prospect, Allen has a very high ceiling. The average sentiment for a tweet mentioning both the Browns and Josh Allen was a 0.2318, shattering that of the closest prospect (Baker Mayfield’s 0.1032).

Prospects looked at (in order of likelihood to be drafted using my model): Josh Allen (QB, Wyoming, Mean Sentiment Score = 0.2318), Baker Mayfield (QB, Oklahoma, Mean Sentiment Score = 0.1032), Josh Rosen (QB, UCLA, Mean Sentiment Score = 0.1018), Saquon Barkley (RB, Penn State, Mean Sentiment Score = 0.05667), Sam Darnold (QB, USC, Mean Sentiment Score = 0.04582)

2. New York Giants: Saquon Barkley, HB (Penn State)

Excellent pick for the Giants here. I don’t think they are ready to replace Eli Manning quite yet, and with a new head coach and one of the best running backs in the league being drafted to them, I think the New York Giants are a dark horse candidate to get a playoff spot in the NFC. After all, this team was 11–5 just two years ago.

Prospects looked at (in order of likelihood to be drafted using my model): Saquon Barkley (HB, Penn State, Mean Sentiment Score = 0.185), Josh Rosen (QB, UCLA, Mean Sentiment Score = 0.148), Quenton Nelson (G, Notre Dame, Mean Sentiment Score = 0.1007), Bradley Chubb (DE, NC State, Mean Sentiment Score = 0.08148), Baker Mayfield (QB, Oklahoma, Mean Sentiment Score = 0.01974), Sam Darnold (QB, USC, Mean Sentiment Score = 0.016)

3. New York Jets: Josh Rosen, QB (UCLA)

This is the first selection that actually shocked me. It has not been a big secret that the Jets have been interested in Mayfield, and many sources have been pointing to them selecting Mayfield at 3. Josh Rosen, however, is a better prospect than Mayfield, and I think Rosen will be able to lead the Jets to the Super Bowl once Brady retires. The average sentiment scores for Rosen and Mayfield are extremely close. The Jets could go either way here.

Prospects looked at: Josh Rosen (QB, UCLA, Mean Sentiment Score = 0.1158), Baker Mayfield (QB, Oklahoma, Mean Sentiment Score = 0.1103), Sam Darnold (QB, USC, Mean Sentiment Score = 0.002784)

4. Cleveland Browns: Bradley Chubb, DE (NC State)

Bradley Chubb is a sure-fire prospect who will terrorize offensive lines alongside Myles Garrett for years to come. The Browns’ defense will be very scary in 2018.

Prospects looked at: Bradley Chubb (DE, NC State, Sentiment Score = 0.1381), Quenton Nelson (G, Notre Dame, Mean Sentiment Score = 0.005634), Mike McGlinchey (T, Notre Dame, Mean Sentiment Score = -0.125)

5. Denver Broncos: Baker Mayfield, QB (Oklahoma)

The Broncos get their quarterback of the future in Baker Mayfield. Mayfield will be able to sit under Case Keenum for a couple of years and fully develop before taking the field. Mayfield is a top prospect who already will be better than Paxton Lynch and Trevor Siemian.

Prospects looked at: Baker Mayfield (QB, Oklahoma, Mean Sentiment Score = 0.2038), Quenton Nelson (G, Notre Dame, Mean Sentiment Score = 0.094), Josh Rosen (QB, UCLA, Mean Sentiment Score = 0.014), Sam Darnold (QB, USC, Mean Sentiment Score = -0.07566)

6. Indianapolis Colts: Roquan Smith, LB (Georgia)

The Colts obtained a massive haul by trading back with the Jets. They could continue to rake in by selecting Roquan Smith here at six, who will surely overhaul their defense.

Prospects looked at: Roquan Smith (LB, Georgia, Mean Sentiment Score = 0.1898), Tremaine Edmunds (LB, Virginia Tech, Mean Sentiment Score = 0.06), Minkah Fitzpatrick (S, Alabama, Mean Sentiment Score = -0.08182), Denzel Ward (CB, Ohio State, Mean Sentiment Score = -0.1648)

7. Tampa Bay Buccaneers: Derwin James, S (Florida State)

It’s not exactly a secret the Buccaneers are in love with Derwin James. The Florida State product is the best safety in this draft, and he will be a welcome addition to the Bucs defense.

Prospects looked at: Derwin James (S, Florida State, Mean Sentiment Score = 0.0484), Mike McGlinchey (OT, Notre Dame, Mean Sentiment Score = 0.03704), Minkah Fitzpatrick (S, Alabama, Mean Sentiment Score = -0.065), Quenton Nelson (G, Notre Dame, Mean Sentiment Score = -0.06955), Denzel Ward (CB, Ohio State, Mean Sentiment Score = -0.2343)

8. Chicago Bears: Quenton Nelson, G (Notre Dame)

The second best prospect in this draft falls into the Bears’ laps at eight. Nelson is a sure-fire Hall of Famer and he will transform the Bears’ offensive line and help progress Trubisky into a Pro Bowl QB.

Prospects looked at: Quenton Nelson (G, Notre Dame, Mean Sentiment Score = 0.2395), Minkah Fitzpatrick (S, Alabama, Mean Sentiment Score = 0.1835), Denzel Ward (CB, Ohio State, Mean Sentiment Score = 0.114), Josh Jackson (CB, Iowa, Mean Sentiment Score = 0.1048), Tremaine Edmunds (LB, Virginia Tech, Mean Sentiment Score = 0.0075)

9. San Francisco 49ers: Calvin Ridley, WR (Alabama)

The 49ers need a shiny new wide receiver for Jimmy Garoppolo to throw to. Calvin Ridley is the best wide receiver in this class and will be terrorizing NFC West defenses for years to come.

Prospects looked at: Calvin Ridley (WR, Alabama, Mean Sentiment Score = 0.2222), Josh Jackson (CB, Iowa, Mean Sentiment Score = 0.2062), Courtland Sutton (WR, SMU, Mean Sentiment Score = 0.1698), Tremaine Edmunds (LB, Virginia Tech, Mean Sentiment Score = 0.1431), Denzel Ward (CB, Ohio State, Mean Sentiment Score = 0.1076), Minkah Fitzpatrick (S, Alabama, Mean Sentiment Score = 0.04308),

10. Oakland Raiders: Rashaan Evans, LB (Alabama)

Rashaan Evans had a very impressive sentiment score. It was higher than any prospect seen up until this point. In fact, both Mike McGlinchey and Josh Jackson also had extremely high sentiment scores as well. Either one of these three prospects could be selected here at the 10th pick.

Prospects looked at: Rashaan Evans (LB, Alabama, Mean Sentiment Score = 0.3396), Mike McGlinchey (OT, Notre Dame, Mean Sentiment Score = 0.3271), Josh Jackson (CB, Iowa, Mean Sentiment Score = 0.2927), Denzel Ward (CB, Ohio State, Mean Sentiment Score = 0.06967), Minkah Fitzpatrick (S, Alabama, Mean Sentiment Score = 0.05063), Tremaine Edmunds (LB, Virginia Tech, Mean Sentiment Score = 0.008264), Vita Vea (NT, Washington, Mean Sentiment Score = -0.05518)

11. Miami Dolphins: Da’Ron Payne (DT, Alabama)

Interesting pick here. With Ward, Fitzpatrick, and Vea still on the board, the Dolphins select Da’Ron Payne. Although he will be able to fill the hole left by Suh, I don’t think he’s the best prospect to select here.

Prospects looked at: Da’Ron Payne (DT, Alabama, Mean Sentiment Score = 0.4124), Denzel Ward (CB, Ohio State, Mean Sentiment Score = 0.3741), Minkah Fitzpatrick (S, Alabama, Mean Sentiment Score = 0.2727), Tremaine Edmunds (LB, Virginia Tech, Mean Sentiment Score = 0.1852), Vita Vea (NT, Washington, Mean Sentiment Score = 0.1287), Josh Jackson (CB, Iowa, Mean Sentiment Score = 0.08955)

12. Buffalo Bills: Sam Darnold, QB (USC)

I think at this point it is well known that the Bills are going to get a quarterback in the draft, whether is it at pick 12 or an earlier pick due to a trade up. The only quarterback left in this no-trade mock draft is Sam Darnold, which makes me believe that the Bills will most likely trade up for him. At this point, I won’t even do a sentiment analysis on the quarterbacks.

13. Washington Redskins: Denzel Ward, CB (Ohio State)

With the Eagles’ receiving core being a constant threat and Odell Beckham returning from injury, Washington will need an elite corner to cover them. Denzel Ward fits the bill.

Prospects looked at: Denzel Ward (CB, Ohio State, Mean Sentiment Score = 0.2416), Tremaine Edmunds (LB, Virginia Tech, Mean Sentiment Score = 0.2258), Minkah Fitzpatrick (S, Alabama, Mean Sentiment Score = 0.2038), Josh Jackson (CB, Iowa, Mean Sentiment Score = 0.1395), Vita Vea (NT, Washington, Mean Sentiment Score = -0.1962)

14. Green Bay Packers: Minkah Fitzpatrick, S (Alabama)

Selecting Fitzpatrick at the 14th pick would be a steal for the Packers. His presence on the defense next to Haha Clinton-Dix will help defense the Vikings’ stellar core of wide receivers as well as the Bears’ newest wide receiver addition, Allen Robinson.

Prospects looked at: Minkah Fitzpatrick (S, Alabama, Mean Sentiment Score = 0.3518), Vita Vea (DT, Washington, Mean Sentiment Score = 0.2041), Tremaine Edmunds (LB, Virginia Tech, Mean Sentiment Score = 0.1465), Josh Jackson (CB, Iowa, Mean Sentiment Score = 0.1208), Marcus Davenport (DE, Texas-San Antonio, Mean Sentiment Score = 0.09552)

15. Arizona Cardinals: Courtland Sutton, WR (SMU)

This pick surprised me. With all of the hype that the Cardinals are selecting a quarterback as well as Bruce Arians’s pushing for Lamar Jackson to be selected at 15, I was under the assumption the Cardinals would go for Jackson here. However, they go for Larry Fitzgerald’s successor instead. Depending on Sam Bradford’s health, this could actually be the right move. That being said, I think this is too high to draft a wide receiver like Courtland Sutton.

Prospects looked at: Courtland Sutton (WR, SMU, Mean Sentiment Score = 0.05556), Vita Vea (DT, Washington, Mean Sentiment Score = -0.07143), Lamar Jackson (QB, Louisville, Mean Sentiment Score = -0.08916), Tremaine Edmunds (LB, Virginia Tech, Mean Sentiment Score = -0.1045), Mason Rudolph (QB, Oklahoma State, Mean Sentiment Score = -0.1579), Marcus Davenport (DE, Texas-San Antonio, Mean Sentiment Score = -0.1892)

Conclusion

I was pretty pleased with how the draft was generated. A lot of the picks actually made sense. That being said, a few things surprised me. I was concerned to see Vita Vea not picked after fifteen picks in the draft. I doubt the Dolphins would select Da’Ron Payne when Vita Vea is still on the board. In addition, I was surprised that Courtland Sutton was selected by Arizona at the 15th pick over a quarterback like Lamar Jackson.

I believe that my model can be pretty accurate in predicting who will be selected for the top five picks. A vast majority of tweets are about players who will be selected with the top few picks in the draft. After the top five, it becomes harder to predict the rest of the draft because we have a lot less data to analyze. However, this could be a good model for determining where the top prospects will be drafted.