User:Onse/Search Engine Ecosystem part 2

 List of abbreviations 
 * CTR: Click-through rate
 * PPC: Pay per click
 * GSP: Generalised second-price auction
 * CPC: Cost per click
 * IP: Internet Protocol
 * 3G: Third generation telecommunications system standard
 * HTTP: Hypertext Transfer Protocol

Introduction
After a general overview of the history, the technical architecture and operating principles in the previous chapter, this chapter presents an economic view on search engines as well as a critical view on search result personalisation. The first section 2 presents who may influence or is affected by the operation of search engines. Further, section 3 focuses on the economical background of search engines. The goal of this very section is to understand the general relationship between search engine owners and advertisers. This also includes how advertising on search engines is done, how advertisers can sell advertisements and how their charges are calculated. The final section 4 deals with the personalisation of search results. Personalisation is an upcoming trend in information technology that allows customising the information shown to the user by her interests. For search engines, it has the advantage that search results can be of higher interest to the user. This means higher user satisfaction as well as the opportunity to advertise more successfully by targeting specific user groups. The goal here is to understand some basic technical possibilities for personalisation, its theoretical background and also to reflect critically on personalisation.

Multi stakeholder system
In this section, a definition for stakeholders will be given and subsequently the most important stakeholders of search engines will be identified and described. There have been several attempts to give a full definition of stakeholders, which were of varying success. A widely used definition is given by Freeman : "A stakeholder in an organization is (by definition) any group or individual who can affect or is affected by the achievement of the organization's objectives." The following five paragraphs describe one stakeholder each. These are the users, the site owners, the advertisers, governments and the search engine owner. The stakeholders and their interactions are also given in Figure 1. Both the text and the figure may not be complete due to the broad scope of the given definition of stakeholders. At the end of this section, the reader should have a general understanding of the multi stakeholder system of a search engine. She should know the most important participants as well as their interactions with the search engine and each other. Furthermore, the reader should get a general understanding of the difficulties of satisfying all stakeholders.

Search engines are information retrieval programs built to ease the discovery of content on the Web. Therefore, the first stakeholders are the users, who are interested in receiving good results. A good result can be defined as a result, which matches the user's search input in a way that she is satisfied with it. Satisfaction is coupled with how relevant the search results seem before clicking any links. This relevance derives from evaluating the information presented on the search result page. Satisfaction is further coupled with the actual relevance of the results to the user. The user can evaluate the actual relevance only after having clicked on results. Hence in this definition, good results need to both seem and be relevant, with the first result being the most relevant one. Considering advertisements, Pickhardt argues that users are also interested in unbiased results. In this context, he describes that users want their results purely ranked by relevance and not influenced by advertisers paying for a better ranking. The notion of good results can cover this case in a less specific way.

Besides the actively searching users, people who create and own websites can also be determined as stakeholders. These can be called site owners. For being a stakeholder, it is of no relevance whether site owners want their sites to be discovered or not. The reason for this is that a site owner must declare if webpages should not be indexed e.g. by specifying it in a file named. Hence, site owners are affected by search engines in either way and are therefore by definition stakeholders. Site owners are also of high importance for the search engine as they provide the content. Without (enough) content, search engines would not be of interest for users.

Revenue is being made by selling advertisements to appear together with the search results. This makes advertisers stakeholders too. They are interested in a low price for placing advertisements. At the same time, advertisers are interested in a high conversion rate or at least a low bounce rate. The conversion rate describes how many visitors of a webpage actually perform a desired action, e.g. bought something. Whereas the bounce rate describes the amount of people who accessed a webpage and left it immediately. In the later section 3.1 it becomes clear that a high bounce rate is very expensive for an advertiser and should be minimised.

As the view of the available content on the Web is heavily shaped by search engines, it may be of interest for governments to prevent biased results or even to influence results themselves. A reason for trying to keep the results unbiased is to guarantee citizens the unfiltered access to information like it is stated e.g. in German law (Art. 5 Abs. 1 | GG (Germany)). But on the other hand, many governments are already censoring the Internet for various reasons like the fight against child pornography or copyright infringement. This makes governments another main stakeholder of search engines. Most of the time, search engine owners accept the local interference by governments for not losing any markets. The only greater exception to this was Google's decision in 2010 to redirect Chinese Google Web Search users to Google Hongkong, which was out of control of the Chinese government. But even in this case, Google Inc. eventually quit the efforts to prevent interaction by the Chinese government.

Search engines have turned their owners like Google Inc. and Yahoo! Inc. into multi-million dollar corporations. Hence, the companies operating search engines are also stakeholders. As profit-oriented companies, their interest is to maximise profits. Yet, due to being in full control of the search engine, the search engine owner can be seen as the head of the multi stakeholder system. As a result, it is her task to try satisfying all other stakeholders' interests at the same time. This can be complex, because as described, these interests are not only diverse but are sometimes even in conflict with each other. An additional difficulty with satisfying all stakeholders is that the roles of stakeholders can overlap. For example a site owner might also be an advertiser.

Economics of a search engine
This section presents the general concepts of monetisation of search engines. As mentioned before, the selling of advertisements is a very profitable business for big search engines and one of their key concepts. Therefore, it is a mandatory background for understanding the interests of most search engine owners. Furthermore, the following subsections provide insight into the auctioning system which allows advertisers to buy placements of their advertisements. After the first subsection, the reader should know the general concept of keyword-based advertising. She should understand the advantages and disadvantages of the discussed payment model for the advertisers and search engine owners. The reader should also extend her understanding of the complexity of the multi stakeholder system. After the second subsection, the reader should be familiar with the basic concept of auctioning of keywords. This includes understanding why a lower bid can result in a higher revenue for the advertiser in the later introduced generalised second-price auction.

Basics of keyword-based advertising
Advertisers are given the chance to buy keywords that will trigger their advertisements to be displayed alongside the search results. This results in advertisements, which are related to the search query. One advantage is that many search engine users do not find these advertisements annoying but at times even helpful. Another advantage is that users seeing an advertisement are likely to be interested in the advertised product and are thus more likely to click it. Search engines like Google Web Search or Yahoo! Search display several advertisements at once for a given search term. Each advertisement is placed in a so called slot, where the top and therefore first slot is has the highest click-through rate (short: CTR). Click-through rate describes the amount of clicks on a link divided by the times it was displayed. The advertisers pay per click (short: PPC) of an advertisement and can sometimes also define a daily maximum budget cf. Google Inc (2014). Pay per click servicehas an advantage for the search engine owner. Higher valued slots on the one hand lead advertisers to pay more in order to have their advertisement displayed in a profitable position. On the other hand, these slots can be resold to a new advertiser faster as the links get clicked more often.

For the advertiser, pay per click has the advantage that she does not have to pay every time her advertisement is being displayed but only, when it is being clicked. Similar to the given description of good results, an advertisement must seem relevant for the user to click on it and be pleased with the search results. However, a high click-through rate is no guarantee for a low bounce-rate or even a high conversion-rate, because the user might find the linked webpage to be irrelevant. This has a negative effect on the user's perceived quality of the results. Additionally, the advertiser will have to pay for people clicking on her advertisements and directly closing the webpage afterwards. So, while the advertiser might not make make any profit from a click, the search engine owner will. This illustrates how the search engine owner as the head of the multi stakeholder system will have to try to satisfy the advertisers and the users. Consequently, the search engine owner will try to maximise the pre-click relevance a user thinks an advertisement has as well as the actual post-click relevance cf. Hillard et al. (2010).

Auctioning of keywords
This section covers an overview of the buying process of keywords for keyword-based advertisement. There, advertisers choose a set of keywords to be associated with their advertisements, which are matched against the users' search terms. Some search engines also offer personalised advertisements to target specific user groups cf. Google Inc. . Examples for such personalisation are limiting the display of advertisements to users of a single gender, from a particular geographic location or having a specific interest. The buying process of personalised advertisement slots is not topic of this section as it is to concrete for a general overview. Still, section 4 deals with personalisation of search results and hence gives an insight about profiling users which can then also be used for selling advertisements.

Advertisers cannot directly buy advertisement slots. Instead, a later explained auction is done for every keyword. The best slot is not guaranteed to be given to highest-bidding advertiser. Instead, the search engine owner chooses the advertisements to be in the best slot, which are predicted to return the highest revenue. This is done by multiplying the estimated advertisement CTR with the money an advertiser pays per click, called cost per click (short: CPC). It is important to note that the CTR of an advertisement must be estimated individually for every advertisement. It cannot be directly derived from the general CTR of a slot. Therefore, Ragno defines the estimated revenue for an advertisement in general as $$E_{ad}[revenue]=P_{ad}(click)\cdot CPC_{ad}$$. Due to its complexity, the concrete calculation of $$P_{ad}(click)$$ is not being discussed. Thus, the following explanation of the keyword auctioning will be simplified so that advertisers bet for the best slot directly instead of just a high probability to gain the best slot.

Search engine advertisement auctions are mostly in the form of a generalised second-price auction (short: GSP). There, the advertisers $$a_1,\dots,a_n$$ are each bidding one price $$b_i$$ willing to pay per click with $$ i \in [1,n]$$. At the same time, each advertiser has one value $$v_i$$ which describes the personal worth of her advertisement being clicked once. There exist the slots $$s_1,\dots,s_m$$ with $$s_1$$ being the most valuable to advertisers and $$s_m$$ the least. The first slot will be given to the highest bidder. In GSP, the advertiser's CPC $$p_i$$ is the price of the bid of the next lower bidder. For example, for three advertisers $$a_1,a_2,a_3$$, and two slots $$s_1,s_2$$ where $$b_2 > b_1 > b_3$$, the highest bidder will have to pay $$p_2 = b_1$$, the second $$p_1 = b_3$$ and the third $$p_3 = 0$$. Here, we assume that there are less slots than advertisers for a given keyword, so $$m < n$$ cf. Varian (2007). As $$a_3$$ does not get a slot, he does not have to pay. If $$m \geq n$$, the search engine owner can choose to set a fixed price for every bidder outside the $$m-1$$ highest bidders so that all cases are defined. For every slot $$s_j \mid j \in [1,m]$$, there is a general CTR $$c_j$$ times the formerly introduced advertisement-specific CTR $$P_{ad}(click)$$. Again for simplicity reasons, any advertisement-specific effects on the CTR will be ignored, so that here only $$c_j$$ as probability for a slot to be clicked will be respected. The expected revenue for the advertiser $$r_i$$ on the advertisement can then be calculated of the value $$v_i$$, price $$p_i$$ the advertiser has to pay for it and the slot CTR $$c_j$$: $$r_i = v_i\cdot c_j - p_i\cdot c_j$$. This formula ignores daily budgets, which an advertiser might have set.

If all advertisers choose to bid their value $$b_i=v_i$$ this is called truth-telling. In GSP, truth-telling is not a Nash equilibrium. Nash equilibria come from game theory and describe a state in a strategic game, where every player is satisfied with her decision in a way that she would not want to change it after inspecting the other players' choices. For GSP this means that there are cases where if every advertiser bid truthfully, some advertisers could have generated a bigger revenue if they chose a different, not-truthful bid. The following example will illustrate such a case. Table 1 shows the possible outcome of a GSP-auction where there is no Nash equilibrium. As there are as many slots as there are advertisers, the lowest-bidding advertiser $$a_3$$ pays a predefined price of $$1$$. In this example, all advertisers bid truthfully. If instead, $$a_1$$ bid $$5$$, she would lose the first slot $$s_1$$ and would be assigned the second $$s_2$$. In this case, the price would shrink to $$p_1 = 2$$, resulting in a revenue of $$r_1 = 2.5$$. As $$a_1$$ could more than double her revenue, truth-telling is not a Nash equilibrium.

Search engine advertising and especially the best strategies for search engine owners as well as advertisers are a complex topic. This section for example ignored daily budgets set by advertisers, advertisement-specific effects on CTR and personalised advertising. The next section deals with personalised search results, which can also be of interest for personalised advertising.

Personalisation of search results
Search engine users are interested in good results. They want to reach webpages whose contents match their search input as quickly and easily as possible. This is reflected in user search behaviour with keyword-based search. On average, users only look at around eight results per search query, before they either have found a webpage they are interested in or rephrase their search query. One factor to achieve this is to learn what the user likes, what she dislikes, what she is interested in, what she is not interested in and then sort or filter the search results accordingly. This is called search personalisation and is the topic of this section. Search personalisation is an example for the efforts made to satisfy the search engine users. The users are essential to the multi stakeholder system due to their relation to site owners, advertisers and search engine owner. As users mostly discover new websites through search engines, it is of interest for many site owners to be indexed by search engines. Advertisers can acquire new customers from users discovering their product through advertisements. While doing so, users generate revenue for the search engine owner.

There are two main approaches for personalisation. One can be called location-based personalisation and the other interest-based personalisation. The first one influences the search results order via the physical location a search query is sent from. However, interest-based personalisation influences the search results order by evaluating the user's interests. Naturally, the two forms of personalisation can be used in combination. They will be explained in greater detail in the following subsections. A reader should be able to distinguish these two forms and understand the relevance of personalisation for future search engine development. Furthermore, she should understand the basic technical concept behind interest-based personalisation. The reader should know how interests can be represented through an interest graph, how collaborative filtering works and what downsides can come with personalisation.

Location-based personalisation
Location-based personalisation implements the idea that there can be relationships between a user's geographical location and her interests. For example, if a user searches for "weather" in a keyword-based search engine, she will likely be interested in the weather situation in his area. For this, location-based personalisation tries to determine the geographic location that the query is sent from. For this, the IP-address of the query sender is often being used. In 2011, Pickhardt demonstrated that Google Web Search personalises searches e.g. for weather via the user's IP-address. By using a proxy server, he unintentionally also showed that locating a user by her IP-address may not always return the real location. When using a proxy server, the search engine will not obtain the IP-address from the user herself because for the search engine the query was sent by the proxy server. Another difficulty with an approach which simply maps IP-addresses to locations, is determining the location of users connected to 3G mobile networks. Balakrishnan et al. showed that cell phones can have the same IP-address-space although they might be separated by hundreds of kilometres. The authors were also able to show that such devices can be localised again by analysing their latencies. Summarised it can be said that locating a user for personalisation can be of interest. It is possible via her IP-address as long as she does not use a proxy-server to send the queries.

Interest-based personalisation
Unlike location-based personalisation, interest-based personalisation cannot be done without any previous knowledge about the user. The reason is that interest-based personalisation sorts search results according to user interests. User interests can be gathered for example by monitoring the user's behaviour on the Web including her prior interactions with the search engine itself. This information must be stored by the search engine and mapped to a user. For a mapping to users, a search engine must be able to identify them. There are several solutions to it. The easiest is, when a user has a user account for the search system and sends a query while she is logged in. Whenever a user is logged out or has no account, she can still be identified. Compared to location-based personalisation, trying to identify users by their IP-addresses is not a good practice. That is because end-users normally only get dynamic IP-addresses assigned by their internet service providers. This means that a normal user's IP-address will change (ir-)regularly e.g. once per day. Instead, cookies or more complex and advanced techniques like browser fingerprinting are being used. Subsequently, the widely used technique of identifying users through HTTP-cookies is being explained.

HTTP-cookies
HTTP-cookies are a realisation for storing information locally on a user's machine and sending this information via the Hypertext Transfer Protocol (short: HTTP) to servers to identify users. This allows storing information associated with a user beyond an HTTP-session. When a user requests a document from a Web server via HTTP, the server may ask the user's browser to create a cookie by using a -field in the HTTP-header. If the user's browser allows it, it will set a cookie with the information given by the server. It will send the cookie the next time requesting the resource declared as value in the -attribute if it is still valid. The validity is specified as - or  -attribute. As mentioned beforehand, cookies allow storing information locally on the user side. A search engine might set a cookie with a unique identifier for each new user. As long as the user does not delete the cookie, the search engine will be able to recognise her. Amongst other things, HTTP-cookies are therefore used for identifying logged in users as well as users who have not authenticated themselves. This makes HTTP-cookies the technical basis for personalisation.

Interest graphs
Interests of a user can be represented through the use of an interest graph like it is depicted in Figure 2. Such a graph has users and interests as nodes and relationships as edges. It goes beyond the social graph, which only models explicit connections to other people. Interests can be various things like places, celebrities, consumer products, art and so on. Such a graph must be directed, because the average Jane can be interested in a celebrity but the celebrity will most likely not be interested in her. When it comes to inanimate things, the interests logically cannot even be mutual. The graph's edges can be weighted to indicate how highly developed an interest is. Interest graphs are built around one or multiple users and can be very valuable for online marketing. The reason is that a large interest graph is of great value for user personalisation e.g. of ads or search engine results. An example would be a search query for the ambiguous term "noise". If an interest graph shows that a user is highly interested in experimental music, she would be shown webpages about the noise music genre ranked higher. Whereas if a user's interest graph shows a high interest in acoustical engineering, webpages about sound like noise control could have been ranked higher. These interests can be extracted from social networks with directed links or from usage activity tracking like analysing search queries. Even more information can be extracted from interest graphs, when combined. This is explained further in the next subsection.

Collaborative filtering
Collaborative filtering describes a technique to predict interests for individual users by analysing the interests of a group of users. "Collaborative" describes the fact that decisions are based on information of multiple users. The term "filtering" describes that this technique will pick specific information for individual users from a vast amount of information. When it comes to search engine result personalisation, it can be argued whether personalised sorting can be called filtering. When defining collaborative filtering as a technique for predicting and recommending instead of explicit filtering of information, it can instead better be called a recommender system. Recommender systems work best for any arbitrary user, when there are many users with similar interests. User interests can be extracted from a (weighted) information graph. Similarity means not only a great overlap in interests but also - if available - in their weights. Comparing the weights is especially important because a very low - or negative - weight might even express a dislike and is therefore contrary to an interest. Hence, having the weights $$w_1, w_2$$ their difference $$w_1-w_2$$ should be close to $$0$$. A similar user can then be recommended any interest she does not share with others of similar interests. Therefore, if a user searches for something that does not match her interests perfectly, results can still be personalised in a way that she is likely to find them relevant and interesting. This way, search recommendations can also be personalised. Table 2 shows an example, where user 3 should be recommended a new artist. The ranking in table 2 ranges from strong dislike (1) over indifference (3) to strong like (5) and no information about interest (-). Similar to the previous example, user 3 is interested in experimental music, which is expressed by her liking Sonic Youth, Merzbow and especially Karlheinz Stockhausen. A matching interest graph is depicted in Figure 2. The colours in Table 2 signify the similarity of other users' interests to user 3, with green being a high and red a low similarity. A search engine could also use this knowledge for auto-completions while typing a search query. If user 3 starts typing "Kylie Mino", the system may then recommend a search for the noise artist "Kylie Minoise" instead of the more popular singer "Kylie Minogue". Not, because the user knew of Kylie Minoise when doing the query but because the users 2 and 3 who have similar interests did. The problematic side to this is being discussed in more detail in the subsection about filter bubbles.

Filter bubbles
Personalisation can help to return good results for the user. To realise this, webpages matching the interests of the user or similar users are preferred in the sorting of the results. As a search query can easily return more results than a user will look at cf. Lorigo et al. (2008), personalisation can be perceived as a form of filtering. Webpages that are not related to the user's interests therefore can get buried after hundreds of other results. This distorted perception of results is called a filter bubble, which could influence the development of a confirmation bias. Confirmation bias describes how people can develop a bias by remembering or searching mainly for information that matches their ideas and beliefs. Pariser (2011) described an example for the filter bubble where two people searched for "BP" shortly after the Deepwater Horizon oil spill on Google Web Search. One of the two got information about the oil spill while the other only saw general and promotional links to BP. He also describes that one person got less results indicating that personalisation can also mean active filtering of information opposed to only perceived filtering by resorting. Another problem with personalisation and filter bubbles is that it can be hard to tell whether search results have been personalised. In his book, Pariser (2011) claims that filter bubbles are inescapable. Albeit, a study by Hannak et al. (2013) showed that remarkable personalisation for Google Web Search only happened by geographical location and for logged in users. For now, this means that at least for Google Web Search, users can still bypass massive personalisation which would result in filter bubbles. Still, many people trust Google blindly to rank the most relevant and important results the highest instead of results that a user is interested in. Thus, many people might not even be aware of filter bubbles. Alternative search engines like DuckDuckGo are trying to win over customers by raising awareness of filter bubbles. In sum, filter bubbles are a growing problem, which can endanger the access to unbiased information. At the same time, personalisation is not necessarily negative. Personalisation can be helpful but the user must be informed about this and she should be able to deactivate it.

Conclusion
This chapter has shown several things. First and most important, search engines affect and are affected by many different groups and individuals. The interests of those stakeholders are often contrary to each other. Thus, satisfying most of them adequately is a complex task for the owner of the search engine. The reward for this work is the possibility to make great profits from advertising.

Second, slots for advertising are indirectly bought through generalised second-price auctions. This chapter simplified these auctions by leaving out advertisement-specific effects, daily budgets and personalised advertising. Still, it should have become clear that maximising the profit for the advertisers and the search engine owner is not trivial.

Third, search engines may personalise their search results to better satisfy users and to allow advertisers targeting specific groups of people. The forms of personalisation can mainly be separated into location- and interest-based. There are various types of ways to recognise users with using HTTP-cookies being the most widely used. This allows the tracking of user behaviour from which interest graphs can be built. These graphs are the basis for personalised search results but can also be used for recommender systems. The advantage is that search results then can not only be personalised based on what the search engine knows a user is interested in. Instead, the results can also be personalised by using information about what she could be interested in. This lead to the problems of potential filter bubbles. As it was discussed in this chapter, personalisation can hamper the access to unbiased information and the formation of unbiased opinions in general.

All in all, search engines are powerful programs that significantly shaped the way we look at the Web today. Due to their intermediate position between users and the content created by site owners, they play an important role to the perception of content on the Web and the possibility to discover information. This special position may bring great profits to the search engine owner but also great responsibilities. The future will show how the multi stakeholder system will adapt to changes like the further development of search engines or interest changes of stakeholders.

Quiz questions
{What rate is of interest for the advertisers?} - A high bounce rate. + A low bounce rate. + A high conversion rate. - A low conversion rate + A high click-through rate. - A low click-through rate.
 * A high bounce rates indicates that users who reach a page will leave it afterwards immediately. This results in no profit for the advertiser. If the user reached the webpage through a banner, the advertiser might even make a loss because she has to pay for the user's click.
 * The conversion rate describes how many users reach the website and can be converted to paying customers. Therefore, it is the most valuable score.
 * Although a click-through rate alone gives no explicit information about profits, it is still important for the advertiser as a subset of it describes paying customers.

{How can a user prevent being tracked?} - She can make her Web browser set the Do Not Track field in the HTTP-header  to. - She can disallow HTTP-cookies. + As of now, she cannot fully prevent being tracked.
 * Do not track is only a proposed header field. Respecting it, is only a gentleman's agreement. Ignoring this field has no legal consequences.
 * Disabling cookies is a way to make tracking more difficult. There are still many ways to track a user e.g. through pixel-graphics, flash-cookies or installed fonts.

{In theory, who would profit from the following business models? +- Pay per click (the advertiser pays whenever a user clicks an advertisement) +- Pay per lead (the advertiser pays when a user clicks an advertisement and gets in contact with the advertiser) +- Pay per sale (the advertiser pays when a user clicks an advertisement and the user buys something) -+ Pay per view (the advertiser pays whenever an advertisement is displayed)
 * type="[]"}
 * Advertiser | Search engine owner
 * The advantage for the advertiser is that she will only have to pay for people actually seeing her side through clicking the advertisement. For search engine owners this needs additional efforts. They have to ensure that unsuccessful advertisements will not be placed in slots with a generally high CTR. Else, expensive slots cannot be resold as fast and profits shrink. This is the reason why the most valuable slot is not given to the highest bidder but to the advertiser whose advertisement is likely to generate the highest revenue for the owner.
 * This model is in the interest of the advertiser because she will only have to pay for users that are actively interested in buying.
 * This is the cheapest model for the advertiser, because she will only have to pay if she sells something. At the same time, the search engine owner will not earn much.
 * This allows the search engine owner to bill the advertiser for every display of an advertisement, no matter if any user even clicks on it.

{Regarding table 2, which artist is or which artists are likely to be recommended to the sixth user?} - Sonic Youth - Stockhausen - Merzbow - Kylie Minoise - Madonna + Kylie Minogue - Britney Spears - No recommendations can be made
 * User 4 and 5 are quite similar to user 6. They both have a strong interest in Kylie Minogue, which user 6 does not know yet. Therefore, this would be a likely recommendation.

{What can be said about search results personalisation?} - Personalisation needs information about the user's interest. + Information gathered from search results personalisation can be used for targeting advertisement to specific groups of people. + Personalisation may involve the collection of sensitive information. + Personalisation may help discovering e.g. webpages that are of a user's interest. - Personalisation always helps the user find the information she is interested in faster. + Personalisation may induce a biased perception towards certain topics.
 * No, personalisation can also happen in a more generic way by analysing the user's physical location.
 * Yes, data from personalisation is very interesting for advertisers and therefore a valuable good for the search engine owner.
 * Yes, information gathered for personalisation purposes may contain information about sexual orientation, religious or political beliefs, health, finances and more.
 * Yes, this is due to the power of recommender systems.
 * No, although this is one of its core intentions, it does not always help the user finding interesting information faster. If a user decides to search on topics she formerly disliked or ignored, previous personalisation information might make the relevant webpages harder to find.
 * Yes, this may be true for people who mainly or exclusively obtain their information for certain topics from the Web and through search engines. Filter bubbles are one of the downsides to personalisation as they may impede the forming of an unbiased opinion and the equal access to information.