Wang Fan closed the webpage and felt that his mood seemed to be a little better. This online store is still more attentive, there is really a customer service chatting, and it is very interesting! Back to work, Wang Fan felt that he suddenly came to inspiration, and then put himself into a state of selfless work.

The recommendation algorithm of the e-commerce industry is to speculate on the products that users may like through some program algorithms. Mainly based on the user's browsing, purchasing, collection, sharing and other data, to profile the user, and use the consumption needs of similar users to predict the user's interest.

There are many kinds of recommendation algorithms, but the mainstream and practical recommendation algorithms are mainly combination recommendation algorithms, which are the synthesis of various recommendation algorithms and the purpose is achieved through various combination ideas such as weighting, transformation, mixing, feature combination, and cascading.

After years of development, the data dimensions that can be analyzed by the recommendation algorithm have been basically fixed, and there are roughly more than 100 kinds. Although Wang Fan's inspiration came, there was no subversive creativity, but he had a clearer understanding of the weight of each dimension.

In nearly thirty minutes, he finished the code adjustments. After the recommendation algorithm is completed, an offline evaluation will be conducted first. There is not much to expect, after all, there has been no major progress in a long time. But as soon as the offline assessment results came out, he was taken aback. 10%。 Is there a mistake? It's not like this before! So he updated the offline data packet and did another offline evaluation. 9.9%。 At present, the average success rate of the recommendation algorithm of the entire industry is about 6%, and the success rate of large companies such as Ali is about 8%. Wang Fan's company is limited by the level of personnel and the small amount of data, and the success rate is about 5%.

Next, he changed the packet five or six times, with an average success rate of more than 10%.

He was completely confused!

Really?

Accidentally created the history of the company's recommendation algorithm?

He suppressed his excitement and kept comforting himself: calm down, calm down, maybe the online real-person assessment effect is not ideal. Push the code to the team leader and let him decide.

Despite this, he had a strong hunch: maybe he could really make history this time!

The team leader is drinking coffee when the code is pushed to the team leader through the version management tool. He was a little surprised to see Wang Fan submit the code so early, after all, he likes to submit before work every time. This submission is too early. Click on the code description, and it is noted that the offline evaluation is 10%. The team leader immediately squirted out the coffee with a smile.

"Wang Fan, are you engaged in wool, 10% of the offline evaluation?" said the team leader with a smile, "Now is not the period of the Great Leap Forward, you don't need to put satellites!"

Wang Fan said unconfidently, "I also suspect that there is a problem with my offline data package, so I quickly submit it to the team leader and let the team leader check it!"

"I don't care so much! If it doesn't reach 7%, you can invite all the team members to eat Haidilao!" said the team leader with a grin. He is usually a joker, and he has a very good relationship with everyone.

"Team leader, you are too ruthless! Who doesn't know that I am the poorest, I just bought a house, and I have to raise a girlfriend, how can I compare with you big guys? Can you spare me?" Wang Fan said bitterly.

The team leader cut off, "No!" Download the code submitted by Wang Fan and start the offline evaluation. A few minutes passed, and the results came out, 11%.

"Holy", the team leader scolded like a ghost. Then I unbelievably updated the offline assessment package and re-evaluated, 11.3%.

The team leader stood up abruptly and said loudly, "Everyone stop what you are doing and listen to me." Just now Wang Fan submitted a code that has been evaluated more than 10% offline! I have just done an offline evaluation, and the result is more than 11%. ”

After the team leader finished speaking, the other team members immediately exploded.

"It's not April Fool's Day!"

"It's not like that in kidding!"

"Team leader, didn't you wake up? Why don't you go home and sleep again?"

"Wang Fan just let himself go, how can the team leader let go of himself?"

"Someone has floated recently, if it weren't for the two balls dragged by below, they would have to fly into the sky! Who? Of course it's Wang Fan!"

The team leader waved his hand and said solemnly, "I repeat, no one is joking! Now let's get to work!Who is, deduct the bonus!" After that, he sat back in his seat and shared the code with the group. After a while, the group members shared the results in the group, with the highest 12%, the lowest 9.5%, and the average 10.1%.

Everyone is starting to fry again.

"It's like a dream!"

"I just twisted my thigh and it really hurts. It doesn't look like it's a dream!"

"Accidentally became the number one in the industry?"

"Wang Fan is burning incense at home today, right? It must be possessed by a god, or how about a battle to become a god!"

Wang Fan said weakly, "Team leader, can you put it on the online evaluation! What if the online effect is not good?"

In general, the results of online and offline assessments are consistent, but they are also uncertain. Sometimes, the results can vary widely. Wang Fan very much hopes that today's code will make history, but the more this is the case, the more he is afraid that the final result will not be good. As the saying goes, the greater the hope, the greater the disappointment. Therefore, he deliberately lowered his expectations, for fear that the final result would disappoint him too much. After all, today's situation is hell-like hell!

The team leader thought for a while and solemnly deployed the code to run online. The online assessment took a long time, but all the team members were not in the mood to work, and they all gathered around the group leader and watched the data fluctuate.

One hour, the average referral success rate is 10.2%.

Two hours, with an average success rate of 9.8%.

Three hours, with an average success rate of 9.7%.

Four hours, with an average success rate of 10.3%.

...

Until the end of work at 6 p.m., nearly 8 hours of online assessment, with an average success rate of 10.1%. Maybe for others, it is impossible to understand the meaning. But for all the team members, they understood that they knew that they were witnessing a miracle! They didn't eat alone at noon, they didn't even order takeout, just to keep an eye on the data.

The team leader announced before leaving work that due to Wang Fan's excellent performance, he would be rewarded with 100,000 yuan on the spot. If the average success rate can be stable at more than 9% within a month, 900,000 yuan will be rewarded again. It is not uncommon to give such a reward! The high success rate of referral directly brings about an increase in the company's profits, and it is a substantial increase. It is conservatively estimated that it can bring hundreds of millions of profits to the company every year. Wang Fan is very happy, after all, 100,000 yuan is not a small amount, and it is very likely that he will get a huge sum of 900,000 yuan in the future, how can he not be happy?

When I got home, I saw that my girlfriend was already home. The meal was ready and waiting for him. He hugged her three times and kissed her on the cheek countless times, until she wondered if he was crazy, and then excitedly recounted the day's events.

"Why is my husband suddenly so powerful today?"

"I suspect it," Wang Fan said hesitantly, "I suspect that the IQ recharge card is working!"