Web search behavior: user goals, query strategies and search success
Good search is not only a ranking problem. Users differ in knowledge, vocabulary, goals and strategy, so search interfaces must support reformulation, exploration and quick evaluation.
Experts
Use more precise vocabulary, filters and iterative reformulation.
Beginners
Need clearer feedback, suggestions and visible paths to refine the query.
Strategies
Top-down, bottom-up and mixed strategies depend on the task and prior knowledge.
Success
Result design, snippets and interaction cost shape whether users find what they need.
Searching the Web
1.Web search behavior of Internet experts and beginners
The accelerated growth of the World Wide Web has turned the Internet into an immense space of information without any organization. This fact has caused users when they perform their searches to They are overwhelmed with information or simply do not know how to do them correctly. To resolve this need, it has been analyzed What type of knowledge is most relevant to Web search and Which structures and strategies produce better results.Following this objective, one of the most prestigious research (link) I carried out an experiment using groups of people with different knowledge computer scientists, where each group was made up of:
Computer experts: These are people related to the world of computing or the Internet. In this group We will find people who have worked in the last 3 years in information search programs or They are simply Web page designers or WebMasters.
Diversification of people: The second experiment is used with people who meet the following characteristics:
Knowledge of computer science and economics.
Knowledge of economics but not computer science
Knowledge of computers but not economics.
They do not have knowledge of computers or economics
1.1 Experiment with experts
The subjects selected for the tests are asked about their behavior when they have to carry out a search on the Internet. The interviewer guides them in questions to obtain the most accurate information possible. Once finished, they are asked to perform a small diagram where they outline their behavior in a search graphically through a diagram of states.Once all the information is collected, the researchers analyze the common concepts, heuristics and common strategies in the subjects analyzed to establish guidelines and models.
1.2 Experiment using searches on the internet
Users have been provided with a device with Internet access and an audio recorder. The purpose of the process is for each subject to carry out a series of real searches, such as searching for a file audio or certain information on a given topic with the device and an observer write down all the processes of search you perform. In addition, the user records their procedures with the recorder to prevent them from being lose any steps. Of course, in addition to the steps, the queries made in each query are stored. The The number of searches carried out has been 1956 queries.Once the results are obtained, we divide them into two groups, the steps to perform a search and the statement of queries used where we obtain the following results:
Steps to search for information
First of all, of the 1956 consultations carried out, two thirds were carried out successfully. The User responses were processed and common unit tasks were obtained. The results can be analyzed in the figure [1] where we also see interesting data such as:
67 percent of subjects use a search engine to help them obtain the desired information
41 percent of users who make a query in the search engine end up repeating the process
80 percent of users when they access a website, it is the one that contains the information they are looking for
1 Graph that shows us the states obtained from the experiment and their percentages As we have mentioned, a large number of users use search engines to make queries (67%). The following figure is shown from the result of this interaction [2] : 2 Graph that shows us the interaction of users with the search engines search.
Preparation of queries
Regarding the queries made, it is worth highlighting an average of 3.64 words and the use of Boolean elements that allow the search engine to perform more precise searches and obtain results more valid.
1.3 Experiment with user diversification
As we have mentioned previously, in In this experiment, the selected users have very diverse knowledge about the subject to be searched and also about computing. In this case now the search data will be stored on a server to be later analyzed and users will have a time restriction of 10 minutes to carry out their searches. Of course, in the first phase of the experiment, the subjects to be analyzed have been classified into 4 groups. different that are:
Group 1: Knowledge of computer science and economics.
Group 2:Knowledge of economics but not computer science
Group 3:Knowledge of computers but not economics.
Group 4:They do not have knowledge of computers or economics
Results
Regarding the results obtained in the experiment, they have been very revealing. The The results obtained have been the following:
Group 1: They have been the fastest in obtaining the solution to the search.
Group 2: These users did not have computer knowledge but they made the queries rather than users who did have knowledge and therefore easily obtained a response validates your problem.
Group 3: Although they knew which search engines they should use or some search methods and elaboration of advanced queries, ignorance about the subject made them reformulate their queries until find the desired information.
Group 4: This group is the one that had the most problems.
2. Research on web search behavior
Another article that analyzes the behavior of users in Internet search is the article Research on Web search behavior prepared between 1995 and 2000. This text studies the patterns, needs and methods used by two very different social groups. to perform internet searches. The first group is made up of subjects between 8 and 18 years old and the second is made up of adults between 18 and 65 years old. The purpose of these studies is to analyze what type of information they search, in what environment and also how they analyze.
2.1 Introduction
Is it really important to study the user search patterns? The answer is resounding Yes. Of all the users who access Internet, 57 percent access daily to search for information through search engines and, therefore, we can affirm that search engines have come to be considered the greatest source of information for many millions of people. users.
2.2 Information Search
A recovery task is composed of three aspects:
The content: The information that is recovered is the interface with which the user obtains this one.
Its capacity: At this point the different algorithms that we can use, the criteria that we can apply.
Users: It is very important to consider the level of knowledge that the user has on the subject to prepare the queries or even the computer knowledge you have to use the search engine.
Web Search Behaviors
According to Spink, Bateman and Jansen (authors of the article discussed above) demonstrated through logs from a Telnet server that 77 percent of users are successful when performing their search. To increase this percentage, the article analyzes different methodologies and approaches that They allow increasing the efficiency of the process or reducing errors, such as:
The queries: Of those 23 percent who do not succeed, 80 percent try again by reformulating their query. It has been demonstrated through various studies that professionals in the sector use elements such as Booleans or queries with essential words without connectors that increase the chances of success. Normally, The vast majority of failed queries have been created using requests with multiple words in natural language. From another point of view, those users who make queries about topics they are unaware of also act differently. While more experienced users usually perform searches with few words and analyze the documents provided to learn about the topic and carry out new, more specific searches, the More novice users do not act this way. What they do is formulate very long requests, with a lot of data. that in many cases provide them with documents that are too general.
Users: Many articles on this topic analyze the type of Internet users and how we can subdivide them into groups to analyze their different interactions, such as:
Children: Despite the terminology in which we refer to this group, we must include all children. users with very poor navigation and many difficulties finding the desired information. Really This type of user does not intend to search for information but rather to navigate certain pages and analyze the information that appears there.
Adults with experience: Which in turn can be subdivided into:
search professionals
web-related workers
students.
It is interesting to study their behavior of this group of adults. According to the server logs, these Users access the website through a very interesting backtraking process. Find a website that interests you, They browse through it and if they don't like it, they go back and look for a different option. In addition to that, you can also divide the time they remain on a site, that is, we could have:
Users who access for a short period of time (as an intermediate step to access another website or by mistake)
Those who access for a long period of time
Those who access to make a query.
The search engine interface: Group the results into types, use a friendly and easy to understand design and Furthermore, differentiating the results with different styles helps guide the user and improve the percentage of successes in the searches.
Search strategies: According to the article Shneiderman (1997) they found three strategies used for the searches that are:
Top-down: The search engine starts with the most general area and descends the search to a more specific point.
Bottom-up: The search engine starts from a specific word and scans the results.
Mixed: Use both strategies.
By defining these models, researchers have managed to generate more systematic search engines in their tasks and in the search execution. These advances have been mainly applied in the interface to help users understand the results obtained.
The tools used. The most expert users always use the same search engines, the same techniques regardless of the interface and even perform very similar queries. Normally they always access to several documents in their searches and also consult their history and visit pages they had already visited previously. While new users are the opposite: they access diverse search engines, use many visual elements for searches and use very diverse queries.
2.3 Web study methodology
In this section we will analyze the problems we encountered when carrying out studies on user behavior with the Web. The commented article already expresses its concern about this issue by defining the web as a testing ground to analyze behaviors and apply scientific methods. social. It is quite easy to assimilate that any valid analysis will have to have interviews, reports, surveys, audio and video recordings and server logs to collect all possible information. Another very important aspect is what type of sample we take. It is very important in statistical analysis take very varied groups of people to analyze the largest group of diverse people.
3 Understand user goals in web search
Daniel E.rose and Danny Levinson (link) They focus on explaining how and what users who interact with the Internet are looking for. To do this, both scientists analyze like other colleagues have carried out their studies, analyzing the queries shared in other projects with similar characteristics. Understanding the "why" of user behavior is essential to satisfy the information they need. Users do not think of searches as an end, but as a process to obtain a objective and, therefore, the objective of the researcher will be to analyze their behavior according to a purpose. Following this philosophy, when we implement a search engine and its methods, an important part of the development will be to improve and adapt the software for the purposes of the users who use it. A very typical case in searches is those very similar or even identical queries, but with different objectives depending on the user who invokes them. By For example, a user searching for clothing stores can search for online locations or physical stores closest to their location. neighborhood everything depends on the user who makes the query,Using this new methodology, the following model has been generated for Web search machines in three tasks that are:
Create a tool for user goals
Create a process to associate search engines with user goals and queries
Modify the machines to use the search objectives.
To finish this introduction we must highlight a very interesting fact. Of all the works published up to 2002 only Brode's article ( Taxonomy of web search (2002)) is the only one that deals with this topic with this approach. The rest of the analyzed works simply analyze the behavior, without studying the purpose.
3.1 Tools to search for objectives
To create these tools, Levinson and Rose have decided to use a space of current objectives, that is, they have searched a set of queries from the AltaVista search engine and have mixed in set of possible objectives based on your own experience. The result is an extensive collection of objectives. Using this list we have classified 200 queries manually. The next step was to review the tool to accommodate the results to that classification made. We have also classified the objectives into different categories. An example is, for example, "resource" which are queries like "song supersubmarina" who do not want information about the music group Supersubmarina but rather to listen to their songs. Other classifications of objectives that may be very interesting are:
Navigational: For example, go to the main brand page and then access your device.
Information: Obtain information about a topic
Location: Locate a place in space
More information: Searches to perform other searches.
