Headline invasion, Baidu struggle, Ali sneaked in. Under the giant “search” battle, “HOW searched well” thought: “The next generation of conversational question and answer search engine” is near, and the future has come.
With the “Today’s headline” announcing the entry search as a sign, the “Internet traffic dispute” of the Chinese Internet was finally placed in the public eye. Internet competition entered the second half, The disappearance of the mobile Internet dividend made the short-term search in the search field come earlier and more cruel – “Search engine” determines the living space of enterprises under the mobile Internet: Baidu proposes a knowledge-enhanced semantic representation model ERNIE, AI launches dialogue-based human-computer interaction DuerOS, attempts to use “All in AI” to turn the tide; Ali quietly Using Quark to develop a full-network search, its exploration and precipitation based on intelligent dialogue has gradually surfaced and successfully applied to the “Tmall Elf”; the byte is dedicated to open up the internal content production chain, and watch the multi-modal search and clench Information portal in the era of mobile Internet…
No matter whether it is success or failure, progress and retreat, the development direction of the next generation of search engines is the same: “Dialogue question and answer” will undoubtedly become the future of search engines.
In addition to the giants, there are also startups in the industry track that are actively exploring the next generation of search engine models and expanding the boundaries of China’s Internet imagination.
Contacted “HOW Good Search” is to seek based on “conversational question and answer”, establish a “new generation search engine” to meet the user’s long tail search needs in multidimensional demand scenarios, and realize the combination of human and computer information. Retrieve the output. The recent “HOW Good Search” announced that it has completed the Pre-A round multi-million dollar financing. This round of financing was led by Blue Lake and GGV Capital, and Palm Capital was the exclusive financial advisor; It has completed the multi-million dollar angel round financing led by Jingwei.
Iteration from the Q&A community to the search engine
“HOW Search” is a two-way pain point for information input and output in search scenarios:
-
Input: In the multi-dimensional life scenario, the search requirements of the user’s long tail cannot be satisfied: the traditional search engine has inaccurate positioning of the long tail search keywords in the life scene, and the information feedback is poor. Problem;
-
Output:The content production threshold is too high, and the value of long tail UGC content related to daily life has not been exploited.
In the multi-dimensional life scenario of users, “HOW Good Search” is dedicated to satisfying users’ long-tailed question-and-answer requirements and establishing a human-machine combined information retrieval and output mechanism:
-
For the search pain points of users cannot filter, extract, and correct search keywords, “HOW Good Search” developed a long tail search sentence keyword recognition matching algorithm: the system induces the user to perform the pole The description of the fine grain size problem, the algorithm identifies the scene keyword with the highest probability, and assists the search, so that the user can conduct natural language questioning on the platform;
-
Based on user response, “HOW Good Search” matches the question to the closest existing response after keyword extraction, and distributes it to the matching user for answering, achieving high matching Content distribution and information output.
Combined with human-machine, “conversational question-and-answer search” based on UGC content not only develops and utilizes the content value of UGC, but also improves the high relevance and diversity of the answers in the search scene, thus assisting “intelligent identification, questioning The function of the “Required Answer” is implemented to meet the “More Thousand Faces of Life Scene Requirements” in the new generation of search engines.
“HOW Good Search” screenshot inside the app
Starting from the establishment of the Q&A community, accumulating valid questions and answers for fragmentation and logical hierarchy processing and using Q&A interaction information for model training, “HOW Good Search” improves the long tail sentence keyword extraction, auto-filling and The content matching and distribution functional framework is used to meet the information retrieval and acquisition requirements under the user’s life needs scenario. In order to build a “new generation of conversational question and answer search engine” architecture, “HOW Good Search” is further developing speech recognition, input and dialogue functions, and improving system stability.
According to the “HOW Good Search” data, the current cumulative number of problems on the platform exceeds 13 million, replying to 18 million. The user asks for a frequency of 1.5 times/day, 3-4 times, and the response rate is 90%, and the average response time is 15-30 minutes.
Two-way empowerment, reverse search
After using the model and algorithm to solve the problem of information input, “HOW Good Search” has developed the function of Share Content Reverse Search. Through the semantic understanding algorithm, the user’s life sharing content published in “HOW Friendship” in “HOW Good Search” can be matched to the appropriate question and adapted to the question. Under this mechanism, the user-generated multi-scene fragmentation information, real life experience, academic knowledge will be developed and utilized. Connect to the real needs question and make “Users share content valuable“.
According to the “HOW Good Search” data, the current cross-probability of the responder and the respondent in the App is 70%, that is, 70% of the users ask questions on the platform and answer questions on the platform. In the decentralized settings, the user is the questioner and the answerer. Through brightening issues, replying to encouragement, “HOW search” encourages users to meet their needs Question response and content output:On the one hand, the user threshold is minimized, the user’s participation is increased, and the user’s stickiness is increased; on the one hand, “useful” is used instead. “Fun”, improve the utilization value of fragmented information, and develop the UGC sharing mechanism in depth.
“HOW Circle” Share Content Reverse Search
for furtherImprove user’s sense of participation and connection, “HOW Good Search” also developed Reply to share, polish issues and other functions, and refine “HOW discovery”, “HOW hot question and answer” and other functions to assist users Operation.
“HOW is good” demo interface
Dialog Q&A: Decentralization
“HOW Good Search” believes that the core of the conversational Q&A is “Decentralization”. Under the decentralized distribution mechanism, a large percentage of users participate in Q&A, and “Q&A” users maintain a high crossover rate can drive the Q&A community Gradually evolving to search, seeks to transform from “from question and answer community to search engine”. According to “HOW Good Search”, the current 45% users in the app are Q&A providers, not viewers. The 30-day retention rate reaches 25 %. The user’s secondary question rate is 62%, and the secondary response rate is 64%.
“Q&A is not a supplement to search engines, question and answer should be the next form of search engine”, “HOW search well” to introduce.
Committed to “Let search as easy as speaking“, “HOW search well” believes that the “next generation conversational Q&A search engine” is nearing the goal and the future is here. From the multi-dimensional life scene, for the long tail search dialogue, with the “user participation and connection sense” “question-answer” closed-loop construction, it is “HOW good search” to create a “next generation search engine” innovation attempt.
The founder of “HOW Good Search” is graduated from Princeton University and is mainly engaged in research on optimization and forecasting. He has worked for a startup in the Smart Grid Forecasting and Optimization Distribution area in Silicon Valley.