How would you make Google search better?

Below is my understanding of google search at high level from layman's vantage point.
- Google search is web based query for any topic, it returns you with information regarding the "Search Topic" with a list of findings in the order of relevance. The main classifications are ALL, Books, Video, Images, Maps and few more.

It helps the general public to find the topic(s) of interest on the internet without having to keep track of things themselves. It also updates itself with recent updates

Before we tackle the improvements, I like to get clarification on "Make it better"
Following are a few among numerous things to improve, let me know which one or a few you like me to focus. Given the time constrain, we have to scope it down.
(1) Search time : Improve the speed of search.
(2) Help : Help the user with auto completion of the search topic so as the right key words are used.
(3) Increase the accuracy & relevancy of the search results.
(4) Expand the languages to reach to increase the audience
(5) Weed-out/Flag false/incorrect/questionable content.
(6) Google's efficiency of managing the search in terms of compute, storage or new crawling algorithms
(7) Let me also know if there is geographical or topic(science, politics) in focus
(8) Analytics for business users

Its assume the interviewer gave you options, the following are the areas of improvement I will be focusing.
Let me pick the most popular criteria of common user.
(1) Search time (3) Accuracy & relevancy of the search results

The goal is to improve the search time and Accuracy, relevancy of a search result.
As I described earlier, the majority of the Google Search user care for time it takes to search and how accurate are we the results. The major success of Google came about is from this two criteria. Then competitors failed to match the speed and accuracy of google and rest was history.

For the goal we have chosen for search time and accuracy, the following are the possible user searches. I am not going by the user groups as there is overlap of people between the different levels of searches. The focus is also going to remain on the human interface not machine interface.

1) Common search : Users looking for day to day life events i.e current news, shopping, maps
2) Complex topic search : Users looking for more complex work i.e researchers, higher grade (college or grade school) students looking uncommon things like papers published etc.

There are other ways of slicing this duck i.e Urban, suburban and rural communities. For now I will be sticking to the Group(Search) I mentioned above.

Common search :
1) Like to see the results the quickest. (Current times of sub secs)
2) Like to see the most relevant results to be within the first 3-4 items.
3) Need to match the current selection trend at the time for the location.
3) Be able to distinguish between fake and real news/results.

Complex topic search:
1) Speed may not be the topic requirement.
2) Most relevancy for the topic of search. I.e Student may be looking for the answer which fetches them the best score on their exams or homework
3) Research topics, need the variety in answers so they can do the appropriately evaluate the paper.

1) Identify the type of search by some of the following means
a. Use the Topic of search to map the category of common or complex.
b. Provide check boxes on the simple search (not advanced) for common-Default or complex.
c. Let users create profiles and you can derive the most likely search type from the profile
d. Use location as also one of the parameter to Id the search type. University, Shopping district, residential
2) Build profiles from the search history of the computer. So the appropriate algorithm/critiera can be set for the searches
3) Use the ML to learn from the selection of the results by the users and time they stay on it as input to improve the search. Also co-relate them with profiles

Lets use the RICE model and score the various option. There will be three phases of rollout suggestion for the above provide.
RICE model is defined as REACH, IMPACT, CONFIDENCE and EFFORT.

IMPACT is defined as follows:
* H is >= 90
* M is 70 - 89
* L is 1 - 69

Effort in person month (we are talking of scale implementation not a prototype)

(R, I, C, E)
1a) 70, H, 80%, 10
1b) 90, H, 100% , 1
so on and so forth.

Based on these model and results.
The first MVP would be combination of 1b, 3
The second MVP 1a, c, 2
Final product - improvement of the first and second MVP and learnings from them.

The metrics for success of improvements are
1) Number of choices made by the user of the first result and top 4 results. (Success criteria)
2) Number of re-searches with altered topic but still the core search remains the same in next 2 minutes (Failures)